[
https://issues.apache.org/jira/browse/IGNITE-20873?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vyacheslav Koptilin updated IGNITE-20873:
-----------------------------------------
Labels: ignite-3 (was: )
> Investigate low performance of IgniteRpcServer executor in sequence put
> workload
> --------------------------------------------------------------------------------
>
> Key: IGNITE-20873
> URL: https://issues.apache.org/jira/browse/IGNITE-20873
> Project: Ignite
> Issue Type: Improvement
> Reporter: Kirill Gusakov
> Priority: Major
> Labels: ignite-3
> Attachments: Performance investigations.pdf
>
>
> During the following workload:
> - 1 node
> - fsync disabled
> - key-value put queries is executing one by one
> We found out, that significant amout of time wasted
> [here|https://github.com/gridgain/apache-ignite-3/blob/00e51e2b2e2e22ac48b64e9dfedacc0f03112543/modules/raft/src/main/java/org/apache/ignite/raft/jraft/rpc/impl/IgniteRpcServer.java#L190]
> - up to 10us (the whole query time 80us). It looks strange, because during
> the simple 1thread put test - there are no other tasks in executor queue.
> This issue is reproducing during the YCSB one-threaded put test or with
> SelectBenchmark.kvGet. Some more details can be found in attached doc.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
