[
https://issues.apache.org/jira/browse/IGNITE-23616?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17896736#comment-17896736
]
Vladislav Pyatkov commented on IGNITE-23616:
--------------------------------------------
JMH benchmark result:
{noformat}
Benchmark (batchSize) (clusterSize) (fsync)
(partitionCount) (replicaCount) (switchThreadFast) Mode Cnt Score
Error Units
InsertBenchmark.kvThinBatchGet 10 1 false
16 1 true thrpt 20 2075.141 ± 149.683
ops/s
InsertBenchmark.kvThinBatchGet 10 1 false
16 1 false thrpt 20 2109.672 ± 163.907
ops/s
InsertBenchmark.kvThinBatchGet 100 1 false
16 1 true thrpt 20 993.601 ± 109.928
ops/s
InsertBenchmark.kvThinBatchGet 100 1 false
16 1 false thrpt 20 854.891 ± 99.234
ops/s
InsertBenchmark.kvThinBatchGet 1000 1 false
16 1 true thrpt 20 194.868 ± 35.556
ops/s
InsertBenchmark.kvThinBatchGet 1000 1 false
16 1 false thrpt 20 243.551 ± 37.252
ops/s
InsertBenchmark.kvThinBatchInsert 10 1 false
16 1 true thrpt 20 1413.116 ± 150.674
ops/s
InsertBenchmark.kvThinBatchInsert 10 1 false
16 1 false thrpt 20 1429.464 ± 129.690
ops/s
InsertBenchmark.kvThinBatchInsert 100 1 false
16 1 true thrpt 20 377.131 ± 97.991
ops/s
InsertBenchmark.kvThinBatchInsert 100 1 false
16 1 false thrpt 20 365.495 ± 72.314
ops/s
InsertBenchmark.kvThinBatchInsert 1000 1 false
16 1 true thrpt 20 52.264 ± 18.242
ops/s
InsertBenchmark.kvThinBatchInsert 1000 1 false
16 1 false thrpt 20 49.996 ± 17.003
ops/s
{noformat}
> Found a performance drop in Ignite benchmarks
> ---------------------------------------------
>
> Key: IGNITE-23616
> URL: https://issues.apache.org/jira/browse/IGNITE-23616
> Project: Ignite
> Issue Type: Improvement
> Reporter: Vladislav Pyatkov
> Assignee: Vladislav Pyatkov
> Priority: Major
> Labels: ignite-3
> Time Spent: 10m
> Remaining Estimate: 0h
>
> h3. Motivation
> The performance issue has appeared without expected prerequisites. We found
> out it happened in IGNITE-23257.
> We could seed the degradation only in batch benchmarks:
> {noformat}
> putAll getAll
> before 26803 60658
> after 22719 17501
> {noformat}
> Configuration: 1 server node, 1 client, 1 loading thread on client side.
> Batch size is 1K per operation.
> h3. Definition of done
> * Find the root cause of the issue.
> * The degradation will have to disappear.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
