[ 
https://issues.apache.org/jira/browse/HDDS-9232?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17769029#comment-17769029
 ] 

Ivan Andika commented on HDDS-9232:
-----------------------------------

[~szetszwo] From what I saw in the Ratis community the reason of the 
configuration is to throttle the amount of AppendEntries sent to the followers 
in case . 

[~ckj] suggested in RATIS-1562 that we can introduce "follower progress state" 
in Ratis to prevent sending AE during InstallSnapshot (similar to ones 
implemented in etcd https://github.com/etcd-io/raft/blob/main/design.md). Maybe 
this would help to prevent unnecessary AE sent to the followers, while at the 
same time not affect the overall write performance?

> Write performance degradation
> -----------------------------
>
>                 Key: HDDS-9232
>                 URL: https://issues.apache.org/jira/browse/HDDS-9232
>             Project: Apache Ozone
>          Issue Type: Bug
>    Affects Versions: 1.3.0
>            Reporter: Mladjan Gadzic
>            Assignee: Tsz-wo Sze
>            Priority: Blocker
>              Labels: ozone-performance, pull-request-available
>
> It is observed on real cluster using unsecure Ozone HA (3 OMs, 3 SCMs, 3DNs) 
> that write performance degraded significantly after Ratis upgrade to 2.5.1 
> (https://issues.apache.org/jira/browse/HDDS-8382). At the time of writing 
> this ticket, it takes master branch ~25mins for 100k keys to be written, 
> while it took 1.3.0 release ~2mins for the same number of keys.
> Test was done using single freon client on OM node. Command 
> {code:java}
> ozone freon rk --bufferSize 4096 --factor THREE --keySize 10240 
> --numOfVolumes 1 --numOfBuckets 1 --numOfKeys 100000 --numOfThreads 10 
> --replication-type RATIS {code}
> was used for that purpose.
> Commit before HDDS-8382, HDDS-8550 (sha 
> 290e5f2ba1826a2084c1745e458872dc76c7ae91) did not exhibit degradation in 
> write performance.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to