[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16563885#comment-16563885 ] Eshcar Hillel commented on HBASE-20188: --- Hi, [~stack] I noticed you posted some additional profiling results from July in the above link. I was wondering, is there a timeline toward the 2.2.0 release? What are the trajectories? Also, are you planing to run YCSB performance benchmarks again? Would it be load A, run A run C? Any other benchmarks you would consider? We hope you will be able to benchmark IMC memstore and to consider making it the default. Last results we presented following HBASE-20542 show that IMC read performance is comparable to no-compaction, and is much better in the write-only workloads x. Please let us know if there is anything else we can do to help drive this effort. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Priority: Blocker > Fix For: 3.0.0, 2.2.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, HBase 2.0 performance evaluation - throughput > SSD_HDD.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16561446#comment-16561446 ] stack commented on HBASE-20188: --- Profiling log link: https://docs.google.com/document/d/1vZ_k6_pNR1eQxID5u1xFihuPC7FkPaJQW8c4M5eA2AQ/edit > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Priority: Blocker > Fix For: 3.0.0, 2.2.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, HBase 2.0 performance evaluation - throughput > SSD_HDD.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16520473#comment-16520473 ] stack commented on HBASE-20188: --- bq. Please note new patch and benchmarks results are available in HBASE-20542 Thanks. Will take a look. bq. So what should I do here before releasing 2.1.0? Compare the performance between 1.4.x and 2.1.0 and give a report? 2.1.0 is same as 2.0.x I'd say. Not sure there is much for you to do [~Apache9]. I'm still working on it... > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Priority: Blocker > Fix For: 3.0.0, 2.1.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, HBase 2.0 performance evaluation - throughput > SSD_HDD.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16520197#comment-16520197 ] Duo Zhang commented on HBASE-20188: --- So what should I do here before releasing 2.1.0? Compare the performance between 1.4.x and 2.1.0 and give a report? > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Priority: Blocker > Fix For: 3.0.0, 2.1.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, HBase 2.0 performance evaluation - throughput > SSD_HDD.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16516256#comment-16516256 ] Eshcar Hillel commented on HBASE-20188: --- Please note new patch and benchmarks results are available in HBASE-20542 > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Priority: Blocker > Fix For: 3.0.0, 2.1.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, HBase 2.0 performance evaluation - throughput > SSD_HDD.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16507045#comment-16507045 ] stack commented on HBASE-20188: --- Update. Have been working on perf in background focused on writes. Writes were bottlenecking on flush; our flush in hbase2 was 2x slower than hbase1s. It was also erratic in that it was flushing sometimes at the limit, other times at well in excess of the limits. With input from the likes of [~ram_krish] and [~anoop.hbase], flushes are 'regular' now with same profile as hbase1. See HBASE-20483 for detail. Our writes are still slower. The bottleneck now seems to be our WAL writing. While the dfsclient is a ball of synchronization knots, it is able to take in bigger blobs than our async WAL writer and so in a simple benchmark where region count is small, the old FSHLog does better (hbase2 is up to 30% slower than hbase1 in certain setups). But if you up the contention and up the region count so it resembles a real deploy, the async WAL starts to shine. At hundreds of regions, it can write faster, and almost as importantly, requires way less resources (To learn more see messy experiments here abouts: https://docs.google.com/document/d/1vZ_k6_pNR1eQxID5u1xFihuPC7FkPaJQW8c4M5eA2AQ/edit#heading=h.niiqwjd247t4). A few of us are working on it. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Priority: Blocker > Fix For: 3.0.0, 2.1.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, HBase 2.0 performance evaluation - throughput > SSD_HDD.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16507016#comment-16507016 ] huaxiang sun commented on HBASE-20188: -- A minor issue found by cpu profiling during PE randomWrite testing. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Priority: Blocker > Fix For: 3.0.0, 2.1.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, HBase 2.0 performance evaluation - throughput > SSD_HDD.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16467480#comment-16467480 ] stack commented on HBASE-20188: --- Good one [~eshcar] Makes sense. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Priority: Blocker > Fix For: 3.0.0, 2.1.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, HBase 2.0 performance evaluation - throughput > SSD_HDD.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16467272#comment-16467272 ] Eshcar Hillel commented on HBASE-20188: --- Thanks Anoop (: Meanwhile I have opened HBASE-20542 to elaborate on the new solution. We can continue the discussion there. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 3.0.0, 2.1.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, HBase 2.0 performance evaluation - throughput > SSD_HDD.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16467239#comment-16467239 ] Anoop Sam John commented on HBASE-20188: [~eshcar] That is excellent observation. Ya we can tune many areas and make ideal perf for IMC feature. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 3.0.0, 2.1.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, HBase 2.0 performance evaluation - throughput > SSD_HDD.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16467229#comment-16467229 ] Eshcar Hillel commented on HBASE-20188: --- Hi, wanted to share some interesting insights and benchmark results. We tried to understand why the benefit of in-memory compaction decreases when MSLABs are used. We finally realized this is due to internal fragmentation that causes under utilization of the memory. For example, setting the active segment threshold to A=0.02 means it stores 0.02*128MB=2.56MB. Each such 2.5MB segment utilizes 2 chunks (spanning *4MB*) which are carried in the compaction pipeline until the data is flushed to disk. Each 2.5MB data taking 4MB space means IMC heap utilization is roughly at 65%. Not ideal. We therefore experimented with A=0.014, namely active segment of size 1.8MB, which fits into a single chunk (leaving some space for overflow etc). Running workloadx+workloada+workloadc show improvement in performance in all these workloads wrt the default parameters of IMC (results are attached [^HBase 2.0 performance evaluation - throughput SSD_HDD.pdf] ). While the new default improves performance we believe there are still cases where an overflow may cause using 2 chunks instead of one. We have and idea how to circumvent this problem. Suggesting to move in two phases: (1) in Jira HBASE-20390 set IMC default parameters to best utilize memory also when using MSLABs. (2) in a new Jira present and implement a solution that avoids the chunk overflow problem. In addition, we are also considering more optimization in HBASE-20480 that potentially reduces overhead of temporary cell objects in while searching in a CCM segment. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 3.0.0, 2.1.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, HBase 2.0 performance evaluation - throughput > SSD_HDD.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16448883#comment-16448883 ] Sean Busbey commented on HBASE-20188: - there's a ton of great discussion included in this jira. Could I ask y'all to distill some of the consensus around how folks should look at simulating application performance in the ref guide? In particular [we have this anemic appendix on YCSB|http://hbase.apache.org/book.html#ycsb] where we could benefit from even an explanation of the tradeoff for {{clientbuffering=true}}. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16446069#comment-16446069 ] stack commented on HBASE-20188: --- Patch on HBASE-20411 removes it. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16446063#comment-16446063 ] Anoop Sam John commented on HBASE-20188: See this one Stack {code} public void incMemStoreSize(MemStoreSize memStoreSize) { if (this.rsAccounting != null) { rsAccounting.incGlobalMemStoreSize(memStoreSize); } long dataSize; synchronized (this.memStoreSize) { this.memStoreSize.incMemStoreSize(memStoreSize); dataSize = this.memStoreSize.getDataSize(); } checkNegativeMemStoreDataSize(dataSize, memStoreSize.getDataSize()); } {code} > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16446041#comment-16446041 ] stack commented on HBASE-20188: --- [~anoop.hbase] There is HBASE-20411 for the Segment lock. Is that what you are seeing? Which HRegion synchronize sir? > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16446035#comment-16446035 ] Anoop Sam John commented on HBASE-20188: We can have sub tasks for write perf issue and read perf issue. Just adding some more note on the write only case perf issue. 1. It is very clear that (in my test setup) the flush op taking much more time now and that in turn affects the write perf also. 2. When we avoid the flush part (temp I changed the flush to just take one row and write that only), the 2.0 perf is just half of that of the 1.4. The main issue is the synchronized blocks in HRegion and Segment write path for updating the memstore size. I did an immediate patch to avoid this synchronized and that makes the perf almost similar to that of 1.4. We should open a sub task for that alone. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16444317#comment-16444317 ] stack commented on HBASE-20188: --- Make a subissue [~ram_krish] and [~Apache9]? (Thanks for the great digging as per [~davelatham]) > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16444172#comment-16444172 ] Dave Latham commented on HBASE-20188: - Bravo, [~ram_krish] and [~Apache9]! Fun to follow along with the sleuthing and see the answer. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16443861#comment-16443861 ] Duo Zhang commented on HBASE-20188: --- OK, checked the code again, the condition we use is the size before filtering, which is good. So I think the problem is, we can only switch from pread to stream when we finish an rpc call. We need to make the rpc returns earlier when we find that we want to switch from pread to stream. Let me open an issue. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16443856#comment-16443856 ] ramkrishna.s.vasudevan commented on HBASE-20188: Exactly. this could happen in other cases also where i want a fraction of data to be returned and not the whole data (like some filters). We track the bytesRead but we take action based on shipped() and not on actual bytesRead. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16443854#comment-16443854 ] Duo Zhang commented on HBASE-20188: --- {quote} I did get some more info on the tests. Since my full table scan was for loading the cache i was filtering all the Cells that were coming out of the scan. {quote} So this is the problem. Actually we want to switch from pread to stream when we find out that we are going to scan large amount of data, but the condition we use is the data we returned, not scaned. In your case, we return nothing and for a whole region, and you can finish in one rpc call, so we have no chance to switch to stream... > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16443760#comment-16443760 ] ramkrishna.s.vasudevan commented on HBASE-20188: [~Apache9] I did get some more info on the tests. Since my full table scan was for loading the cache i was filtering all the Cells that were coming out of the scan. So this means we will never try to ship anything so there was no need for the switch from pread to STREAM. Since by default any scan will use HDFS pread() now (even if the scan type is DEFAULT) we are seeing that when we do scan region by region evey time there is tcp connection setup and then the read happens in preads where as in STREAM read we don have this connection setup. The blockreader already has the connection in place. That is what the flamegraphs show. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, hits_with_fp_scheduler.png, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16443082#comment-16443082 ] stack commented on HBASE-20188: --- For [~ram_krish], on fastpath scheduler vs no fastpath scheduler. I did a run last night and confirmed that stuff runs faster for me when fastpath scheduler is enabled. I see about ~4% more throughput. 23081 reads/s vs 22252. See attached hits_with_fp_scheduler.png for image showing two workloadc runs, first w/ fp on, and then with it off. Also shown are cache hits and misses which seem about same adjusted for difference in throughtput. Let me try with all from cache as you had. Will report back here. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442452#comment-16442452 ] Duo Zhang commented on HBASE-20188: --- {quote} True. As I said give me some time will be back here. {quote} Good. Wait for your good news. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442450#comment-16442450 ] ramkrishna.s.vasudevan commented on HBASE-20188: bq.So this is a bug we need to fix, not something we should add to the ref guide. True. As I said give me some time will be back here. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442451#comment-16442451 ] Duo Zhang commented on HBASE-20188: --- {quote} For 2.0 comparison between STREAM and PREAD it is 30 secs vs 97 secs for a simple scan of 10G. {quote} Wait a minute, as you say PREAD, you did not specify ReadType.PREAD in your test right? It should be ReadType.DEFAULT? > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442447#comment-16442447 ] Duo Zhang commented on HBASE-20188: --- And what I want to say is, by design the switch will not impact performance too much, at least for your full table scan case. So this is a bug we need to fix, not something we should add to the ref guide. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442446#comment-16442446 ] Duo Zhang commented on HBASE-20188: --- {quote} For 2.0 comparison between STREAM and PREAD it is 30 secs vs 27 secs for a simple scan of 10G. 97 vs 27 is for 2.0 vs 1.2. {quote} I'm a little confused, so for 2.0, STREAM is 30s and pread is 27s? Then what is the 97s? The feature is written by me and I've tested it long ago and the result is really different from yours. I've posted a possible problem and since you said that switch to use STREAM directly could help so I think it is the root cause, but then you said that set to a smaller result size does not help... Thanks. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442439#comment-16442439 ] ramkrishna.s.vasudevan commented on HBASE-20188: I am not before system now. For 2.0 comparison between STREAM and PREAD it is 30 secs vs 27 secs for a simple scan of 10G. 97 vs 27 is for 2.0 vs 1.2. As I said I will dig in deeper as to what exactly is happening in 1.2 and 2.0 and report back here. Just give a day or two. HDFS is 2.7.3 and I have some HDD and not SSDs in my small cluster. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442420#comment-16442420 ] Duo Zhang commented on HBASE-20188: --- And in HBASE-17917, the result shows that the performance is the same... {noformat} ./bin/hbase pe --rows=1000 --cacheBlocks=false --caching=30 --scanReadType=pread/stream --nomapred scan 1 ./bin/hbase pe --rows=100 --cacheBlocks=false --caching=30 --scanReadType=pread/stream --nomapred scan 10 {noformat} And the result {noformat} One Thread Test PREAD : ~220s STREAM : ~187s DEFAULT: ~187s Ten Theads Test PREAD : ~32s STREAM : ~27s DEFAULT: ~27s {noformat} And the strange thing is, for me, pread is slower if you scan all the data, but the performance is 220s vs 187s, but in your test, the result is 97s vs 27s, could you please share more things about your test? Hardware? HBase config? HDFS setup? etc. Thanks. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442407#comment-16442407 ] Duo Zhang commented on HBASE-20188: --- {quote} But one thing is sure that this pread / stream change makes scans slower (the long running scans) and the same has been discussed on that original JIRA. {quote} The point here is that, you have 10G data, and only 256KB is pread, then I do not think the switch will impact the performance too much. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442258#comment-16442258 ] ramkrishna.s.vasudevan commented on HBASE-20188: I did that setMaxResultSize(2MB). It did not help much. It was still around 97 secs. How ever I have not dived in to the logs as to exactly what is happening and what is the exact cause of the slowness. But one thing is sure that this pread / stream change makes scans slower (the long running scans) and the same has been discussed on that original JIRA. I will be back with more data here. Thanks [~Apache9]. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442226#comment-16442226 ] Duo Zhang commented on HBASE-20188: --- {quote} In this case there is no offheap cache. It is just simple LRU based cache. {quote} But we do not use this information... See HBASE-18055. {quote} I changed from DEFAULT to STREAM and the time taken to do a full table scan came down from > 95secs to 30 secs. {quote} OK, seems it is the problem I described above. But still a bit slower than 1.2? Oh shit, I checked the code, {code} public static final long DEFAULT_HBASE_SERVER_SCANNER_MAX_RESULT_SIZE = 100 * 1024 * 1024; {code} The default buffer size is 100MB... Could you please use scan.setMaxResultSize to set the receive buffer to 2MB to see if it can lead to a better result? I think maybe we can quit earlier in StoreScanner if we reach the maxPreadSize size so that we can switch to streaming immediately... > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442213#comment-16442213 ] ramkrishna.s.vasudevan commented on HBASE-20188: In this case there is no offheap cache. It is just simple LRU based cache. I changed from DEFAULT to STREAM and the time taken to do a full table scan came down from > 95secs to 30 secs. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442206#comment-16442206 ] Duo Zhang commented on HBASE-20188: --- One possible reason for the slow down I can imagine is that, since we can only switch from pread to streaming in the shopped method of StoreScanner as the read path offheap approach hold the references to some buffers in StoreFileScanner, if the receive buffer is 2MB, then we can only use pread to read the 2MB data, although we want to change to stream after 4 * blockSize(usually 256KB). And things could be worse if you set a even larger receive buffer. But first, I think we need to confirm that this is the problem. So I need to know what is the result if just use ReadType.STREAM instead of DEFAULT. [~ram_krish] Thanks. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442189#comment-16442189 ] Duo Zhang commented on HBASE-20188: --- {quote} A simple full table scan with 10G data takes 97 to 100secs with 2.0 and the same 10G full table scan on a 1.2 takes 27 secs. We need to document on the pread to stream switch over for scans that happens in this case. Just noting down here so that we are able to track it down as part of the release. {quote} What is the result if you just set the ReadType to STREAM? > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442184#comment-16442184 ] ramkrishna.s.vasudevan commented on HBASE-20188: Just noting down based on offline discussion with [~stack]. A simple full table scan with 10G data takes 97 to 100secs with 2.0 and the same 10G full table scan on a 1.2 takes 27 secs. We need to document on the pread to stream switch over for scans that happens in this case. Just noting down here so that we are able to track it down as part of the release. The pure read workload and when all data in cache I was seeing around 2k to 3k performance lesser between 1.2 and 2.0. After some profiling, changed the FastPathExecutor and just used the BalancedQueueRpcExecutor and that 2k to 3k difference disappered. Now both 2.0 and 1.2 are able to do around 32500 ops/sec. Previously it was aroudn 29k to 30 k ops on an average. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442031#comment-16442031 ] stack commented on HBASE-20188: --- I added a few graphs of flush time by size up on https://docs.google.com/spreadsheets/d/1sihTxb4aCplR3Rr_GGXkPlwhMIm-CbB9j_5339AS0Zc/edit#gid=1016758826 Flushed take longer in 2.0.0, about 2x longer. In 1.2.7, flush sizes are clustered around 128M. In 2.0.0 the size varies over the full range up to blocking. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16441995#comment-16441995 ] stack commented on HBASE-20188: --- [~anoop.hbase] Then I should be able to see the 'slower flushing' by graphing the flush amount against the fourth column, the time taken to flush? > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16441981#comment-16441981 ] Anoop Sam John commented on HBASE-20188: This flush change is visible in 2.0.0 not only CompactingMemstore. My tests are on DefaultMemstore only. The initial flushes (in both versions) happen with memstore size ~128MB only. But what I can see is the 2.0.0 flush op takes >2x time for the flush to finish. So the new flush requests get assigned to a flusher thread much later and by that time the memstore size grows much larger than 128MB. Which again takes even more longer for the flush. This is the issue with write perf for sure. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16441945#comment-16441945 ] stack commented on HBASE-20188: --- Report comparing flush histories of 1.2.7 vs 2.0.0 with client batching enabled (2MB) > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16441683#comment-16441683 ] stack commented on HBASE-20188: --- I tried enabling YCSB clientSideBuffering, see https://github.com/brianfrankcooper/YCSB/issues/1136, but it made the regionserver go into GC hell. Default buffer size is 12MB. Setting it to 2MB made it so the test would pass. I see that 1.2.7 is better at writing and its a wash when mixed or read-only. I do see an instance of memstore bloat as @anoop talks of above.Will be back w/ detail. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16441396#comment-16441396 ] stack commented on HBASE-20188: --- I added another sheet to the report (Truncate 04/17). It provides detail on summary given above, that hbase2 -- when the table is truncated when we go from 1.2.7 to 2.0.0 runs -- is a good bit faster reading but still slower when pure reads. In this default YCSB hbase2 is good enough for now. There is a bunch we can do to optimize but no easy win after study; will require effort. As per an off-list comment by [~ram_krish], indeed, the YCSB default is no client-side batching -- see below where payload is 1.7k though we are set to use bufferedmutator. Let me try PE, the tool [~anoop.hbase] has been using. It does batching. {code} 501411 2018-04-17 11:07:34,037 TRACE [RpcServer.default.FPBQ.Fifo.handler=46,queue=1,port=16020] ipc.RpcServer: callId: 2477247 service: ClientService methodName: Multi size: 1.7 K connection: 10.17.240.20:58304 deadline: 9223372036854775807 param: region= ycsb,user5139312943861817391, 1523988283780.8a9442655986021253d3acc69dde60df., for 1 actions and 1st row key=user6571173983002699380 connection: 10.17.240.20:58304, response regionActionResult { resultOrException { index: 0 result { } } } regionStatistics { } queueTime: 0 processingTime: 6 totalTime: 6 {code} > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16440606#comment-16440606 ] Anoop Sam John commented on HBASE-20188: PE command hbase org.apache.hadoop.hbase.PerformanceEvaluation --nomapred --presplit=40 --size=30 --columns=10 --valueSize=100 --writeToWAL=false --inmemoryCompaction=NONE randomWrite 100 The exact numbers I need to calc. But the net PE run time is 3x for 2.0 compared to 1.4.2 > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16440582#comment-16440582 ] Eshcar Hillel commented on HBASE-20188: --- [~anoop.hbase] can you share the exact command line you are running. Can you report the throughput numbers for 1.4.2 and for 2.0.0. Thanks. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16440561#comment-16440561 ] Anoop Sam John commented on HBASE-20188: With PE (batched writes) , 2.0.0 write only case is way slower. As said before the issue is the slow flush op. For same load and total data size, I can see that 1.4.2 doing >2x flushes# compared to 2.0.0. 128 MB is the flush size and 512 MB is the blocked memstore size. We never cross 500 MB mark in 1.4 but in 2.0, many a times we cross 512 MB and so the batched mutate op fails and there will be many retry. This tries are also having a sleep before the trial. This makes the net run time for PE to be very slow. So there is throughput dips in btw and net avg throughput is way low. Examined the GC also once, there is no GC issue in 2.0 compared to 1.4 > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16440550#comment-16440550 ] stack commented on HBASE-20188: --- Chatting with [~ram_krish] and [~anoop.hbase] they remind me that I need to test w/ PE; its batched writing vs the write-per-row that YCSB is doing (a row per rpc rather than batches of rows as PE does). I'll let them comment in here. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16440510#comment-16440510 ] stack commented on HBASE-20188: --- Been distracted (HBASE-20411 and it seems like we do more allocation in pb3.5.1 than we did in pb2.5... though we have the nice new UnsafeByteOperations facility <= Need to dig more). Main finding from today is that I have been missing a line from my deploy script -- the truncate ycsb table after the 1.2.7 run and before 2.0.0 starts running. This pulled down read throughput (we were reading twice the data). I ran into this because HBASE-20411 'improved' reads -- which made no sense... it was just that I'd truncated the table before I made HBASE-20411 runs. I'm running more tests but it looks like, while hbase2 is currently about 10-15% slower on writes (and more erratic), it is much better on mixed read/write, and on pure read. Will report better tomorrow. This is all defaults (IHC is on set to its current defaults) and a 16G heap. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, hits.png, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, perregion.png, run_ycsb.sh, > total.png, tree.txt, workloadx, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16438530#comment-16438530 ] stack commented on HBASE-20188: --- In my YCSB runs, comparing 1.2.7 to 2.0.0 loads, I see that we carry more in hbase memstores but when I look by region, we seem to carry less per region and spike sizes more often. See attached graphs (the hits is so you can see the overall profile of load, mixed, pure-read for 1.2.7 and then for 2.0.0) > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt, workloadx, > workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16437794#comment-16437794 ] Eshcar Hillel commented on HBASE-20188: --- [~anoop.hbase] can you post the command line or script + settings you are using so we can re-produce these results? > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt, workloadx, > workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16435998#comment-16435998 ] Anoop Sam John commented on HBASE-20188: I tried to test the write perf using PE RandomWrite tests. (Pure writes only). This is with not writing to WAL. (Mutations mark writeToWal = false). I see 2.0.0 perf much lower compared to 1.4.2. (It is DefaultMemstore only). On investigation further I can see that the flush operation takes too long time now. The initial flushes kicked off at ~128 MB mark only. But this takes much longer time than the 1.4.2 case. There are 5 flusher threads in my test. So later on the flush entries for regions are added at 128 MB mark only but when really the flush happens it will grow to 2x and 3x sizes.. Those flushes takes even more. This makes the region size to grow >4x mark and so writes failed and client to retry it. This makes the throughput at these time very low. For flush we make a StoreScanner instance on top of ImmutableSegment. Fetching cells from this Scanners seems to take time! (Added more logs to each op to each which takes time). Not sure what all changes happened in this area. SQM any significant changes? > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt, workloadx, > workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16435236#comment-16435236 ] Eshcar Hillel commented on HBASE-20188: --- We also ran the same experiments in HDD machines and indeed we see that in HDD the lift of IMC over None is smaller. We now repeat all experiments again to eliminate noisy results. We will post a summary of both SSD and HDD results by EoD tomorrow. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt, workloadx, > workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16434384#comment-16434384 ] Eshcar Hillel commented on HBASE-20188: --- vq. I should add workloadx to the suite of tests I'm trying? Yes please do. In my script [^HBASE-20188-xac.sh] I run it after (pre)splitting into 10 regions. Opening a sub-jira to update defaults. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt, workloadx, > workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16434227#comment-16434227 ] stack commented on HBASE-20188: --- [~eshcar] I should add workloadx to the suite of tests I'm trying? If there are better defaults, lets open an issue to get them in. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt, workloadx, > workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16433710#comment-16433710 ] Eshcar Hillel commented on HBASE-20188: --- And another thing, might be just a distraction but we think it is worth mentioning – Backward scans should run faster with IMC since the flat segments are simply arrays and it is very simple to go to prev in O(1), while in skip list prev is implemented by seeking the prev key which is done in O(log n). We don't have an easy way to verify this since YCSB does not support testing this kind of operation, but we are looking into it and will report if something interesting comes up. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt, workloadx, > workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16433704#comment-16433704 ] Eshcar Hillel commented on HBASE-20188: --- re-attaching [^workloadx] from the most recent experiment > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt, workloadx, > workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16433700#comment-16433700 ] Eshcar Hillel commented on HBASE-20188: --- Just to summarize the results again -- We see that in write workloads IMC improves performance; it delays flush to disk and hence reduces number of disk compaction. When values are small IMC reduces memory occupancy by reducing metadata size (regardless of workload distribution), when the distribution is skewed IMC reduces memory occupancy by eliminating data duplication (regardless of value size), when the values are big and workload is uniform IMC doesn't help. For reads IMC is either comparable or slightly worse than None (no in-memory compaction). In addition measures we did in past experiments show that IMC reduces write amplification, again due to reducing number of disk compaction. I am opening a new Jira to change the default to the parameters that showed best performance in the recent benchmarks. Namely, IMC policy = ADAPTIVE, active segment porition = 0.02, limit on number of segments in pipeline = 2. We are continuing with our experiments to see if any additional changes can help improve the performance. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt, workloadx, > workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16433458#comment-16433458 ] ramkrishna.s.vasudevan commented on HBASE-20188: bq.If it is a long scan then we will switch to stream later. Yes. that seems to take some time. bq.I see this transition happening in tests, from pread to streaming. In randomreads you should not see this happening. In the report that you have attached it is randomReads only na even for R + W case? > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16433452#comment-16433452 ] stack commented on HBASE-20188: --- I see this transition happening in tests, from pread to streaming. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16433379#comment-16433379 ] Duo Zhang commented on HBASE-20188: --- {quote} Scans in 2.0 are slower because scans are also like preads now. {quote} If it is a long scan then we will switch to stream later. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16432783#comment-16432783 ] Eshcar Hillel commented on HBASE-20188: --- Attaching additional benchmark results where we update and read single column rows [^HBase 2.0 performance evaluation - 8GB(1).pdf] In workloadx (write-only) with a single (wider) column -- Adaptive outperforms None by 15%. In workloads a and c with a single wide column – Adaptive and None are comparable. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB(1).pdf, HBase 2.0 > performance evaluation - 8GB.pdf, HBase 2.0 performance evaluation - Basic vs > None_ system settings.pdf, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16431148#comment-16431148 ] Eshcar Hillel commented on HBASE-20188: --- I attached the results of some additional experiments we ran with 8GB heap. We created a new workload [^workloadx]. It is a write-only skewed distribution, with client side batching. Results show that IMC with 2% active and 2 pipeline segments has an advantage over none in write-only workloads. Script to run the experiments is here [^HBASE-20188-xac.sh], and relevant hbase-site configuration here [^hbase-site.xml] we would still like to investigate the read latency with IMC we have some direction we plan to explore; will come back with results. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188-xac.sh, > HBASE-20188.sh, HBase 2.0 performance evaluation - 8GB.pdf, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > hbase-site.xml, lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt, workloadx > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16427119#comment-16427119 ] stack commented on HBASE-20188: --- bq. The latest results with 8G cache are also with short circuit reads ON? Is there any variation in the stack trace? [~ram_krish] Its on, yes. I'm sure there is variation in stack traces but I've not done the analysis yet (and the two have strayed significantly so hard to make correlations but I'll get to it). bq. Scans in 2.0 are slower because scans are also like preads now. I'd doubt this makes a difference when 48 threads contending on the one server but I can check. bq. Could you please post the complete results for 8G so that we could see the difference between the reads and the writes? Will do [~ebortnik] > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16426793#comment-16426793 ] Anastasia Braginsky commented on HBASE-20188: - [~stack], what tool do you use in order to produce the graphs that you have a the bottom of your sheets with results? > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16426569#comment-16426569 ] Edward Bortnikov commented on HBASE-20188: -- Michael, thanks for all the diligence, apparently you are a step ahead of us with 8G. Could you please post the complete results for 8G so that we could see the difference between the reads and the writes? The workloada result is weird - the writes are skewed, and IMC should really shine. Apparently, the read and the write paths have very different (and independent) issues. With workloadc, there is no reason IMC would work faster (multiple segments to look up), but let's understand workloada first. Thanks again. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16426488#comment-16426488 ] ramkrishna.s.vasudevan commented on HBASE-20188: The latest results with 8G cache are also with short circuit reads ON? Is there any variation in the stack trace? Scans in 2.0 are slower because scans are also like preads now. bq. 'dfs.client.read.shortcircuit.streams.cache.size' and 'dfs.client.socketcache.capacity' values? These values were increased because the default size was causing some issue with ShortcircuitCache. {code:java} 2017-07-18 22:52:28,969 ERROR [ShortCircuitCache_SlotReleaser] shortcircuit.ShortCircuitCache: ShortCircuitCache(0x122da202): failed to release short-circuit shared memory slot Slot(slotIdx=26, shm=DfsClientShm(f0cce51b1df7a0c887c2b708b1bf702d)) by sending ReleaseShortCircuitAccessRequestProto to /var/lib/hadoop-hdfs/dn_socket. Closing shared memory segment. java.net.SocketException: read(2) error: Connection reset by peer {code} WE have not written any detail doc but just collected the observations that we got. As I said when you have enough RAM and all data is in page cache and you have lot of threads reading from HDFS then Short circuit cache was really needed because TCP connection was a problem. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16426469#comment-16426469 ] stack commented on HBASE-20188: --- Added a 5th sheet to our doc. I ran compare of 1.2.7 to 2.0.0 and 2.0.0 w/o in-memory compaction w/ 8G of heap; i.e. lots of cache misses. 1.2.7 is > 2x the throughput for read-only loads and 10-20% better on writes. On mixed-load, 2.0.0 with in-memory-compaction OFF is better than 1.2.7 but with it on, its much worse. Something is wrong here when lots of cache misses (With previous runs with heap of 31G, most reads were from cache). > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16426262#comment-16426262 ] stack commented on HBASE-20188: --- [~eshcar] I tried it here and seems to run. I don't get your complaint. I'm at commit d655a1ca3e208a829641837d027ced59ead243fc Author: Sean Busbey Date: Thu Mar 22 11:55:56 2018 -0500 Add HBase 2.0 binding. Are you running a released version? Could you try adding guava to your CLASSPATH to satisfy the java.lang.NoClassDefFoundError: com/google/common/base/Preconditions I made a summary fourth sheet in the document of where we are at the moment. We are 20% slower writing, 11% slower reading and 15% faster doing 50/50. If we disable in-memory compaction we are 9%/7%/17%. Am waiting now on a good story to tell about in-memory compaction. I'll try with smaller heaps in meantime to see if it helps but currently its detrimental in all of these basic YCSB runs. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16426168#comment-16426168 ] Eshcar Hillel commented on HBASE-20188: --- I am running hbase12 client This is the code of the client {code} package com.yahoo.ycsb.db.hbase12; /** * HBase 1.2 client for YCSB framework. * * A modified version of HBaseClient (which targets HBase v1.2) utilizing the * shaded client. * * It should run equivalent to following the hbase098 binding README. * */ public class HBaseClient12 extends com.yahoo.ycsb.db.HBaseClient10 { } {code} The difference from hbase10 is just in the pom.xml file I believe, which includes shaded-client instead of hbase-client - could this be to blame? {code} com.yahoo.ycsb hbase10-binding ${project.version} org.apache.hbase hbase-client ... org.apache.hbase hbase-shaded-client ${hbase12.version} {code} > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16426154#comment-16426154 ] stack commented on HBASE-20188: --- Try hbase12 client instead of hbase10 [~eshcar]? > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16426134#comment-16426134 ] Eshcar Hillel commented on HBASE-20188: --- This is the code that initiates {{bufferedMutator}} in ycsb: {code:java} final TableName tName = TableName.valueOf(table); this.currentTable = connection.getTable(tName); if (clientSideBuffering) { final BufferedMutatorParams p = new BufferedMutatorParams(tName); p.writeBufferSize(writeBufferSize); this.bufferedMutator = connection.getBufferedMutator(p); } {code} so need to understand why {{connection.getBufferedMutator(p)}} returns null > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16426130#comment-16426130 ] Eshcar Hillel commented on HBASE-20188: --- I am trying to write a new workload that applies client side buffering by using {{clientbuffering=true}} However when running the workload I get the following exception in line {{Preconditions.checkNotNull(bufferedMutator);}} {code} java.lang.NoClassDefFoundError: com/google/common/base/Preconditions at com.yahoo.ycsb.db.HBaseClient10.update(HBaseClient10.java:441) at com.yahoo.ycsb.DBWrapper.update(DBWrapper.java:198) at com.yahoo.ycsb.workloads.CoreWorkload.doTransactionUpdate(CoreWorkload.java:775) at com.yahoo.ycsb.workloads.CoreWorkload.doTransaction(CoreWorkload.java:608) at com.yahoo.ycsb.ClientThread.run(Client.java:454) at java.lang.Thread.run(Thread.java:745) {code} I was able to use this ycsb property in the past. Anyone aware of changes to client implementation that result in a null {{bufferedMutator}}? > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425880#comment-16425880 ] stack commented on HBASE-20188: --- [~ebortnik] Thanks for the question. I fixed the description. It uses the default RPC scheduler which includes fastpath. bq. One other thing that puzzles me is the discrepancy between your and Eshcar Hillel's results for workloadA - her results show +27% upside for IMC, curious what's going on here? Dunno. Maybe its the amount cached. I could try downing my heap size from 31G to 8G and see if it helps. bq. Last question - do you intend to start looking at off-heap configurations? We are working on them now, too. Not as part of this effort. Will start up a new effort to try offheaping but I see that of lower priority compared to ensuring default hbase2 deploy does not regress on hbase1 perf. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425868#comment-16425868 ] stack commented on HBASE-20188: --- [~ram_krish] dfs.client.read.shortcircuit.skip.checksum helps. Let me try 1.2.7 with this flag. I'l be back. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425861#comment-16425861 ] Edward Bortnikov commented on HBASE-20188: -- [~stack] just making sure we're on the same page .. the "2 all defaults" column (col I) does not include FastPath (included in Col F), is this intentional? One other thing that puzzles me is the discrepancy between your and [~eshcar]'s results for workloadA - her results show +27% upside for IMC, curious what's going on here? Last question - do you intend to start looking at off-heap configurations? We are working on them now, too. Thanks > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425818#comment-16425818 ] stack commented on HBASE-20188: --- [~anastas] bq. If you are running again can you please also add to this run the " plus the default 2.0 RPC scheduler, FastPath"? Did that. The latter runs are using the default's which includes the fastpath RPC scheduler. It improves throughput though it looks bad when you look at locking dumps. [~eshcar] bq. Are you still using 31GB heap in your runs? 31GB heap for 25GB of data is too much. With 8GB I think the gc affect is more pronounced I can try that. I've been using 31G so most data is out of cache. I was trying to eliminate i/o. I can roll it back in now I've figured a problem w/ i/o reads. For short-circuit reads, see HBASE-20337. Shout if not clear. Thanks > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425806#comment-16425806 ] stack commented on HBASE-20188: --- I attached my hbase-site.xml, the script I'm using to drive YCSB (See below the commit hash I'm using), and hbase-env.sh. These are the configs I've been running of late. I'm at the tip of branch-2.0 80724e9ba8a295f6d1ee2161fb5ed5a920749da5 commit d655a1ca3e208a829641837d027ced59ead243fc Author: Sean Busbey Date: Thu Mar 22 11:55:56 2018 -0500 Add HBase 2.0 binding. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-env.sh, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425804#comment-16425804 ] Edward Bortnikov commented on HBASE-20188: -- [~eshcar] could you please post your YCSB 100%W benchmark code? Thanks > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, hbase-site.xml, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, run_ycsb.sh, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425792#comment-16425792 ] stack commented on HBASE-20188: --- [~anastas] and [~eshcar] I tried 0.02. I put the numbers up on the third sheet at extreme right (just the throughputs). 0.02 is worse than 0.1 in this run. At the moment, I have in-memory compaction being a drag on our throughput up to 10% or so in all cases. I need a story that shows in-memory compaction shining. Without this, I want to turn it off as default. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425146#comment-16425146 ] Eshcar Hillel commented on HBASE-20188: --- bq. can you please share how can we switch on short-circuit reads in our experiments? OK I see how to do it in HBASE-20377. Do I need to set all 5 parameters that are mentioned there or just the first? > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425142#comment-16425142 ] Eshcar Hillel commented on HBASE-20188: --- {quote}Did you use hbase defaults or did you change segment count or flush size from default? {quote} No I did not change any default except for the system settings (cms and mslab) in the second and third experiment. 2 major differences with respect to your setting is (1) I run on SSD *and* (2) I use only 8GB heap. Are you still using 31GB heap in your runs? 31GB heap for 25GB of data is too much. With 8GB I think the gc affect is more pronounced. You can run an experiment with 0.02 (this was shown to be optimal once) but I wouldn't haste in changing *any* default before we run full experiments. There are several parameters that affect each other, as I mentioned above (pipeline length, active portion CAM/CCM, etc.), and I would like to check all of them more deeply, both in the current workloada/workloadc and in an additional workload. But before we run any further experiments, can you please share how can we switch on short-circuit reads in our experiments? Thanks. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425138#comment-16425138 ] Anastasia Braginsky commented on HBASE-20188: - Thanks [~stack]! If you are running again can you please also add to this run the " plus the default 2.0 RPC scheduler, FastPath"? Looks like orthogonal to in-memory-compaction, but if this improves performance, let us see the full picture... > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425130#comment-16425130 ] stack commented on HBASE-20188: --- Let me try a run with 0.02 then [~anastas]. If good, will open issue to make it default. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425123#comment-16425123 ] Anastasia Braginsky commented on HBASE-20188: - [~stack], Great to hear that you found the reason for read performance big gap! Regarding Eshcar's result, as I understand it was just intended to show the effect of not using MSLAB and using another GC (G1GC). All other parameter were default, no change in the segment count or in-memory-flush size threshold. However, I think we should stop using and comparing in-memory-compaction with 10% threshold and change it to 2%. Eshcar's result show that in 99/95th percentile (under given workloads) in-memory-compaction doesn't show performance degradation. Also for mixed workload BASIC performs better then NONE. We still have to bring you a workload that makes the in-memory-compaction to perform in its full power. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425075#comment-16425075 ] stack commented on HBASE-20188: --- [~eshcar] Did you use hbase defaults or did you change segment count or flush size from default? I see NONE does better than BASIC always. Thanks. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425066#comment-16425066 ] stack commented on HBASE-20188: --- I added a third sheet name "Short Circuit Reads 25M Run" at [1] with timings with short-circuit read in place for hbase1 and hbase2. Here are findings: {quote} Findings: hbase1.x performs better than 2.x in pure read and pure write modes (but we are now within 10%). Mixed load (workloada 50/50), hbase2 is better no matter what combination. FastPath RPC Scheduler, the default for hbase2 is better than the hbase1 RCP scheduler though it looks ugly in the thread dumps with all threads seemingly backed up on its Semaphore coordinator. hbase2 uses more CPU but seems to have a flatter GC profile. in-memory compaction costs. For load, no-in-memory compaction is 4% slower than hbase1, but with in-memory compaction, it is 9% slower. For workloada, no-in-memory-compaction is 25% faster than hbase1 and with in-memory compaction, 17% faster. For workloadc, with no-in-memory compaction, we are 2% slower. With it, we are 5% slower. {quote} 1. https://docs.google.com/spreadsheets/d/1w2NBqAPFthG8Ib4C0pHpLARYpWoIF2Vck2vHZW77zE4/edit#gid=1651250875 > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16425019#comment-16425019 ] stack commented on HBASE-20188: --- [~ram_krish] Setting this on the client-side dfs.domain.socket.path /home/stack/sockets/stack_dn_socket This configuration parameter turns on short-circuit local reads. dfs.client.read.shortcircuit.skip.checksum makes sense. Let me try it here and see if it helps. Let me add it to the doc over on HBASE-20337. Do you recall what prompted your upping of the 'dfs.client.read.shortcircuit.streams.cache.size' and 'dfs.client.socketcache.capacity' values? Lets get that into HBASE-20337 too. You have "... We have done some detalied study on the effect of short circuit reads and have our analysis on it." Is it available anywhere boss? > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16424993#comment-16424993 ] ramkrishna.s.vasudevan commented on HBASE-20188: What was the config that was missing for short circuit reads in hbase-2? In our recent R + W experiments we had to enable short circuit reads for our tests because without that we had Port issues because the underlying drives were faster and so every local read trying to establiish the TCP was failing and retrying. (thus adding to latency). When the number of threads were lesser we did not get this issue. With short circuit reads this problem was not there but we had to explicitly enable it. We have done some detalied study on the effect of short circuit reads and have our analysis on it. How is hbase1 running with short circuit by itself? Another important thing Anoop found while doing those test was this config {code:java} dfs.client.read.shortcircuit.skip.checksum true {code} We had to enable it because in hbase we does its own checksums. Also we increased the '>dfs.client.read.shortcircuit.streams.cache.size' and 'dfs.client.socketcache.capacity' > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, > lock.127.workloadc.20180402T200918Z.svg, > lock.2.memsize2.c.20180403T160257Z.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16424894#comment-16424894 ] stack commented on HBASE-20188: --- Fixing short-circuit reads config made a big difference to hbase2 read throughput putting it close to hbase-1.2.7. Let me update the report. hbase1 seemed fine with having shortcircuit reads = true but hbase2 was complaining falling back on remote reads. The giveaway was the differing lock profiles. See attached locking profile flamegraph lock.127.workloadc.20180402T200918Z.svg for 1.2.7 workloadc. Notice how we are blocking on the ShortCircuitCache cache inside in *local* BlockReader. A run against hbase2 with same configurations had the locking profile lock.2.memsize2.c.20180403T160257Z.svg. There are a few things going on but we are sticking on PeerCache from *remote* BlockReader. Looking in hbase2 regionserver logs, it seems like we ran fine for a while and then the shortcircuit cache would throw exceptions and hold up the handler a while. Our doc on short-circuit setup is stale. Updated it here HBASE-20337 > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > cpu.127.workloadc.20180402T200918Z.svg, > cpu.127.workloadc.20180402T200918Z.svg, flamegraph-1072.1.svg, > flamegraph-1072.2.svg, lock.2.memsize2.c.20180403T160257Z.svg, > misses.127.workloadc.20180402T200918Z.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16424884#comment-16424884 ] stack commented on HBASE-20188: --- Fixing short-circuit reads config made a big difference to hbase2 read throughput putting it close to hbase-1.2.7. Let me update the report. hbase1 seemed fine with having shortcircuit reads = true but hbase2 was complaining falling back on remote reads. The giveaway was the differing lock profiles. Here is hbase1's locking profile for workloadc looked like: [^misses.127.workloadc.20180402T200918Z.svg] Notice how we are blocking on the ShortCircuitCache cache inside in *local* BlockReader. A run against hbase2 with same configurations had this locking profile: [^cpu.2.memsize2.c.20180403T160257Z.svg] There are a few things going on but we are sticking on PeerCache from *remote* BlockReader. Looking in hbase2 regionserver logs, it seems like we ran fine for a while and then the shortcircuit cache would throw exceptions and hold up the handler a while. Our doc on short-circuit setup is stale. Updated it here HBASE-20337 > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > cpu.2.memsize2.c.20180403T160257Z.svg, flamegraph-1072.1.svg, > flamegraph-1072.2.svg, lock.2.memsize2.c.20180403T160257Z.svg, > misses.127.workloadc.20180402T200918Z.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16424632#comment-16424632 ] stack commented on HBASE-20188: --- Thanks [~eshcar]. Short-circuit reads were not enabled properly for the hbase2 runs (hbase1 was able to do short-circuit reads w/o full configuration in place). This seems to put us close to hbase1 read perf. Let me add a sheet to the above report with latest numbers and then dump a summary here. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16424604#comment-16424604 ] Eshcar Hillel commented on HBASE-20188: --- Attached are the results of evaluations over *SSD* machines [^HBase 2.0 performance evaluation - Basic vs None_ system settings.pdf] , and the script to run them [^HBASE-20188.sh] (which is based on the script by Stack). The setting is also similar: 1 master, 1RS with 8GB heap, 1 ycsb client, underlying HDFS set to 3-way replication. Comparing Basic with default configuration vs None under different system settings: cms/mslab vs cms/no-mslabs vs g1gc/no-maslab Summary of results: 1) None outperforms Basic in a uniform distribution of insert-only operations that includes multiple split events 2) Basic outperforms None in a mixed workload with zipfian distribution 3) None is slightly better than Basic in read-only zipfian workload 4) not using mslab improves performance in zipfian distribution workloads and has a negative effect with insert-only uniform workload 5) g1gc performs worse in all cases; this could be due to lack of tuning It is important to note that each configuration was tested once, each of these runs can be an outlier - a good or bad outlier Next we will come up with a workload which demonstrates the advantage of in-memory compaction as well as continue with benchmarks to determine optimal default values for in-memory compaction, namely portion of active segment, length of pipeline, etc. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, HBASE-20188.sh, HBase 2.0 > performance evaluation - Basic vs None_ system settings.pdf, > ITBLL2.5B_1.2.7vs2.0.0_cpu.png, ITBLL2.5B_1.2.7vs2.0.0_gctime.png, > ITBLL2.5B_1.2.7vs2.0.0_iops.png, ITBLL2.5B_1.2.7vs2.0.0_load.png, > ITBLL2.5B_1.2.7vs2.0.0_memheap.png, ITBLL2.5B_1.2.7vs2.0.0_memstore.png, > ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16424243#comment-16424243 ] stack commented on HBASE-20188: --- Thanks [~ram_krish] It helps to have independent observation. I'm doing default configs. and basic YCSB loadings. When all is cached, the two come closer but still a significant difference; not as near as you see (15%). What happens if reads are not all from cache in your setup? Currently looking at shortcircuit reading. We don't seem to be doing it my hbase2 runs compared to hbase1 runs. Will be back. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16423788#comment-16423788 ] ramkrishna.s.vasudevan commented on HBASE-20188: The previous results were not apple to apple because offheap bucket cache does an onheap copy for hbase-1.2. So tried out the same experiment for a longer duration with L1 block cache and the avg was hbase -1.2 vs hbase -2 was 32393 ops/sec vs 28898 ops/sec. So we have a 4k difference still. In Stack's case it is much higher. Will try to figure out if there are any obvious bottlenecks that can be resolved to make up for this difference. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16423585#comment-16423585 ] ramkrishna.s.vasudevan commented on HBASE-20188: I just reran the experiment with loading around 15G of data and loading everything to cache and doing pure read workloads. I used the hbase-1.2 client only with 100 threads (all single node) hbase-1.2.6 vs hbase-2 (31099 ops/sec vs 30838 ops/sec). (Almost same). I have 15 regions in my table and there are no splits that happen. All 15G of data is served from bucket cache (offheap mode). The results are not so different even if serve from L1 block cache also. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16423581#comment-16423581 ] ramkrishna.s.vasudevan commented on HBASE-20188: bq.None really. Tried with data cached thinking it i/o that was responsible for the difference, but while it brings us closer, we are still down from hbase1.2.7 (52221.77 ops/second vs 33839.57). This is really surprising. Even after data cache you see hbase-2 to be so much slower? All are defaults in both cases? What about bloom filters? What is the cache size here? And you sure all data is served only from cache? > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16423307#comment-16423307 ] stack commented on HBASE-20188: --- Thanks for taking a look [~huaxiang] Be aware that the 9.1 is the proportion of 100% of the *current* thread, not of total CPU. If you have a patch, that'd be great but don't sweat it. Its not 11% of CPU but of whatever this one thread was consuming. See the companion flamegraphs for overview. I don't even see getMin as registering. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16423279#comment-16423279 ] huaxiang sun commented on HBASE-20188: -- Hi [~stack], I was looking at tree.txt. The following concerned me, it seems that getMin() was taking lots of cpu time. 9.1/81.8 = 11 percent of cpu. It makes me wondering the fix I did for HBASE-12148 is causing trouble for write performance. Let me prepare a patch for undoing the change in HBASE-12148 and we can rerun to see if this improves, thanks. {code:java} (t 81.8,s 0.0) org.apache.hadoop.hbase.regionserver.RSRpcServices::doNonAtomicRegionMutation(org.apache.hadoop.hbase.regionserver.HRegion,org.apache.hadoop.hbase.quotas.OperationQuota,org.apache.hadoop.hbase.shaded.protobuf.generated.ClientProtos$RegionAction,org.apache.hadoop.hbase.CellScanner,org.apache.hadoop.hbase.shaded.protobuf.generated.ClientProtos$RegionActionResult$Builder,java.util.List,long,org.apache.hadoop.hbase.regionserver.RSRpcServices$RegionScannersCloseCallBack,org.apache.hadoop.hbase.ipc.RpcCallContext,org.apache.hadoop.hbase.quotas.ActivePolicyEnforcement) (t 81.8,s 0.0) org.apache.hadoop.hbase.regionserver.RSRpcServices::doNonAtomicBatchOp(org.apache.hadoop.hbase.shaded.protobuf.generated.ClientProtos$RegionActionResult$Builder,org.apache.hadoop.hbase.regionserver.HRegion,org.apache.hadoop.hbase.quotas.OperationQuota,java.util.List,org.apache.hadoop.hbase.CellScanner,org.apache.hadoop.hbase.quotas.ActivePolicyEnforcement) (t 81.8,s 0.0) org.apache.hadoop.hbase.regionserver.RSRpcServices::doBatchOp(org.apache.hadoop.hbase.shaded.protobuf.generated.ClientProtos$RegionActionResult$Builder,org.apache.hadoop.hbase.regionserver.HRegion,org.apache.hadoop.hbase.quotas.OperationQuota,java.util.List,org.apache.hadoop.hbase.CellScanner,org.apache.hadoop.hbase.quotas.ActivePolicyEnforcement,boolean) (t 81.8,s 0.0) org.apache.hadoop.hbase.regionserver.HRegion::batchMutate(org.apache.hadoop.hbase.client.Mutation[],boolean,long,long) (t 81.8,s 0.0) org.apache.hadoop.hbase.regionserver.HRegion::batchMutate(org.apache.hadoop.hbase.regionserver.HRegion$BatchOperation) (t 81.8,s 0.0) org.apache.hadoop.hbase.regionserver.HRegion::doMiniBatchMutate(org.apache.hadoop.hbase.regionserver.HRegion$BatchOperation) (t 81.8,s 0.0) org.apache.hadoop.hbase.regionserver.HRegion$MutationBatchOperation::writeMiniBatchOperationsToMemStore(org.apache.hadoop.hbase.regionserver.MiniBatchOperationInProgress,org.apache.hadoop.hbase.regionserver.MultiVersionConcurrencyControl$WriteEntry) (t 81.8,s 0.0) org.apache.hadoop.hbase.regionserver.HRegion$BatchOperation::writeMiniBatchOperationsToMemStore(org.apache.hadoop.hbase.regionserver.MiniBatchOperationInProgress,long) (t 81.8,s 0.0) org.apache.hadoop.hbase.regionserver.HRegion$BatchOperation::visitBatchOperations(boolean,int,org.apache.hadoop.hbase.regionserver.HRegion$BatchOperation$Visitor) (t 81.8,s 0.0) org.apache.hadoop.hbase.regionserver.HRegion$BatchOperation$$Lambda$247.826973093::visit(int) (t 81.8,s 0.0) org.apache.hadoop.hbase.regionserver.HRegion$BatchOperation::lambda$writeMiniBatchOperationsToMemStore$0(long,org.apache.hadoop.hbase.regionserver.MemStoreSizing,int) (t 81.8,s 9.1) org.apache.hadoop.hbase.regionserver.HRegion$BatchOperation::applyFamilyMapToMemStore(java.util.Map,org.apache.hadoop.hbase.regionserver.MemStoreSizing) (t 72.7,s 0.0) org.apache.hadoop.hbase.regionserver.HRegion::access$600(org.apache.hadoop.hbase.regionserver.HRegion,org.apache.hadoop.hbase.regionserver.HStore,java.util.List,boolean,org.apache.hadoop.hbase.regionserver.MemStoreSizing) (t 72.7,s 0.0) org.apache.hadoop.hbase.regionserver.HRegion::applyToMemStore(org.apache.hadoop.hbase.regionserver.HStore,java.util.List,boolean,org.apache.hadoop.hbase.regionserver.MemStoreSizing) (t 72.7,s 0.0) org.apache.hadoop.hbase.regionserver.HStore::add(java.lang.Iterable,org.apache.hadoop.hbase.regionserver.MemStoreSizing) (t 72.7,s 0.0) org.apache.hadoop.hbase.regionserver.AbstractMemStore::add(java.lang.Iterable,org.apache.hadoop.hbase.regionserver.MemStoreSizing) (t 72.7,s 0.0) org.apache.hadoop.hbase.regionserver.AbstractMemStore::add(org.apache.hadoop.hbase.Cell,org.apache.hadoop.hbase.regionserver.MemStoreSizing) (t 72.7,s 0.0) org.apache.hadoop.hbase.regionserver.AbstractMemStore::internalAdd(org.apache.hadoop.hbase.Cell,boolean,org.apache.hadoop.hbase.regionserver.MemStoreSizing) (t 72.7,s 0.0) org.apache.hadoop.hbase.regionserver.MutableSegment::add(org.apache.hadoop.hbase.Cell,boolean,org.apache.hadoop.hbase.regionser
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16423224#comment-16423224 ] stack commented on HBASE-20188: --- I tried enabling FSHLog, the old WAL writer, in place of AsyncFSWAL. The write performance was slightly less in the load phase and a good bit less when doing the 50/50. It does not seem to be cause of lower write throughput in hbase2. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (HBASE-20188) [TESTING] Performance
[ https://issues.apache.org/jira/browse/HBASE-20188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16422735#comment-16422735 ] stack commented on HBASE-20188: --- bq. Any news with the directions you suggested Stack? None really. Tried with data cached thinking it i/o that was responsible for the difference, but while it brings us closer, we are still down from hbase1.2.7 (52221.77 ops/second vs 33839.57). [~ram_krish] observes that we go faster when we use the hbase20 client for some reason (45049.10) but not fast enough and I'm thinking that whatever this client-side difference is, it'd probably make 1.2.7 go faster too (can't run hbase2 client against hbase1). Ram is looking into what difference is on client. Trying to compare perf locking, cpu, and allocation traces, their profiles differ too much to be able to finger a 'culprit'. The Semaphore in the RpcScheduler gets 'blamed' in cpu and locking profiles but with it in place, our throughput goes up and thinking on it, it probably makes sense that threads coordinate around this point (if you or anyone have a better idea on how to do the handoff, I'm all ears). There are a few dumb things that I can fix but they won't gain us much. There is some macro change that I'm unable to discern at the mo. bq ASYNC_WAL does not work. SYNC_WAL is default. HBASE-16689 bq. What would this (flip to g1gc) entail? Needs an owner. Said person would do some long runs where they'd figure some conservative defaults that would likely work in most cases and then they'd evangelize our move to G1GC (Message to list with "... CMS is deprecated...G1GC is the future..."). Then we'd flip. Would be cool if said person did stuff like run the long-range tests to see if MSLAB is still needed when running G1GC. bq. It would be unfortunate to get such a big release of HBase without adjusting to the progress in jvm management. Agree. 2.0.0 would be (have been) the right place to do it. Thanks. > [TESTING] Performance > - > > Key: HBASE-20188 > URL: https://issues.apache.org/jira/browse/HBASE-20188 > Project: HBase > Issue Type: Umbrella > Components: Performance >Reporter: stack >Assignee: stack >Priority: Blocker > Fix For: 2.0.0 > > Attachments: CAM-CONFIG-V01.patch, ITBLL2.5B_1.2.7vs2.0.0_cpu.png, > ITBLL2.5B_1.2.7vs2.0.0_gctime.png, ITBLL2.5B_1.2.7vs2.0.0_iops.png, > ITBLL2.5B_1.2.7vs2.0.0_load.png, ITBLL2.5B_1.2.7vs2.0.0_memheap.png, > ITBLL2.5B_1.2.7vs2.0.0_memstore.png, ITBLL2.5B_1.2.7vs2.0.0_ops.png, > ITBLL2.5B_1.2.7vs2.0.0_ops_NOT_summing_regions.png, YCSB_CPU.png, > YCSB_GC_TIME.png, YCSB_IN_MEMORY_COMPACTION=NONE.ops.png, YCSB_MEMSTORE.png, > YCSB_OPs.png, YCSB_in-memory-compaction=NONE.ops.png, YCSB_load.png, > flamegraph-1072.1.svg, flamegraph-1072.2.svg, tree.txt > > > How does 2.0.0 compare to old versions? Is it faster, slower? There is rumor > that it is much slower, that the problem is the asyncwal writing. Does > in-memory compaction slow us down or speed us up? What happens when you > enable offheaping? > Keep notes here in this umbrella issue. Need to be able to say something > about perf when 2.0.0 ships. -- This message was sent by Atlassian JIRA (v7.6.3#76005)