[
https://issues.apache.org/jira/browse/CASSANDRA-16337?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17253060#comment-17253060
]
Yifan Cai commented on CASSANDRA-16337:
---------------------------------------
I have run the test that compares the system performance with and w/o running
{{garbagecollect}} during the steady state load.
h3. Test Plan
# Run data prepopulation
# Run steady state load for X seconds as Phase 1.
# Make one test case dependent change, e.g. triggering garbagecollect, altering
table, etc., and continue the steady state load for Y seconds as Phase 2.
# Compare the performance metrics beween Phase 1 and Phase 2.
Result report link (saved as a pdf):
https://github.com/yifan-c/CASSANDRA-15581-COMPACTION-TEST/blob/main/CASSANDRA-16337/7019-Test:%20Perf%20Comparison%20%5BLCS%20-%20garbagecollect%5D.pdf
We can observe that both read & write latencies (avg., p95, p99) increased
significantly when {{garbagecollect}} is running. Read avg. was doubled, p99
was 3x. For the write latency, its avg. was ~6x and p99 was over 100x.
We can also observe several write timeouts during {{garbagecollect}}.
The number of L0 sstables increased very fast after triggering
{{garbagecollect}}. Meanwhile, the number was kept at a low level during just
the steady state load.
The conclusion is that for LCS, {{garbagecollect}} has a significant impact on
the system performance. We could observe latency increment and throughput
decrement when running the command in the production.
> LCS steady state load vs. steady state load with garbagecollect running
> performance test
> ----------------------------------------------------------------------------------------
>
> Key: CASSANDRA-16337
> URL: https://issues.apache.org/jira/browse/CASSANDRA-16337
> Project: Cassandra
> Issue Type: Sub-task
> Components: Test/benchmark
> Reporter: Yifan Cai
> Assignee: Yifan Cai
> Priority: Normal
>
> The testing cluster should be pre-populated with ~200GB data in each node.
> Run the steady state workload to compare the read, write and compaction
> performance between before and during garbagecollect is running.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
