[ https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15597123#comment-15597123 ]
stack commented on HBASE-16890: ------------------------------- I reran WALPerformanceEvaluation and here is what I got: Current {code} 11697968 11697969 4204706.715579 task-clock (msec) # 7.284 CPUs utilized 11697970 51,803,614 context-switches # 0.012 M/sec 11697971 6,685,043 cpu-migrations # 0.002 M/sec 11697972 18,132,994 page-faults # 0.004 M/sec 11697973 7,791,964,493,250 cycles # 1.853 GHz 11697974 <not supported> stalled-cycles-frontend 11697975 <not supported> stalled-cycles-backend 11697976 1,927,527,070,105 instructions # 0.25 insns per cycle 11697977 335,996,633,114 branches # 79.910 M/sec 11697978 8,301,975,795 branch-misses # 2.47% of all branches 11697979 11697980 577.245436180 seconds time elapsed {code} With async enabled {code} 1596 1597 2820443.128629 task-clock (msec) # 9.413 CPUs utilized 1598 27,279,634 context-switches # 0.010 M/sec 1599 4,548,444 cpu-migrations # 0.002 M/sec 1600 9,174,991 page-faults # 0.003 M/sec 1601 6,222,195,076,963 cycles # 2.206 GHz 1602 <not supported> stalled-cycles-frontend 1603 <not supported> stalled-cycles-backend 1604 1,448,534,066,827 instructions # 0.23 insns per cycle 1605 244,568,362,833 branches # 86.713 M/sec 1606 4,213,196,541 branch-misses # 1.72% of all branches 1607 1608 299.621230438 seconds time elapsed {code} With 100 threads, async finishes in half the time. Let me try with the server up. Here is what I ran: {code} for i in 100; do for j in 1; do export HBASE_CLASSPATH_PREFIX=`pwd`/hbase/lib/hbase-server-1.4.0-SNAPSHOT-tests.jar ; ./hbase/bin/hbase --config conf_hbase classpath; perf stat ./hbase/bin/hbase --config /home/stack/conf_hbase org.apache.hadoop.hbase.wal.WALPerformanceEvaluation -threads $i -iterations 250000 -keySize 50 -valueSize 100 &> "/tmp/wal.5.async.${i}.${j}.txt"; done; done {code} > Analyze the performance of AsyncWAL and fix the same > ---------------------------------------------------- > > Key: HBASE-16890 > URL: https://issues.apache.org/jira/browse/HBASE-16890 > Project: HBase > Issue Type: Sub-task > Components: wal > Affects Versions: 2.0.0 > Reporter: ramkrishna.s.vasudevan > Assignee: ramkrishna.s.vasudevan > Fix For: 2.0.0 > > Attachments: contention.png, contention_defaultWAL.png > > > Tests reveal that AsyncWAL under load in single node cluster performs slower > than the Default WAL. This task is to analyze and see if we could fix it. > See some discussions in the tail of JIRA HBASE-15536. -- This message was sent by Atlassian JIRA (v6.3.4#6332)