[
https://issues.apache.org/jira/browse/HDDS-3745?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
runzhiwang updated HDDS-3745:
-----------------------------
Description:
I start a ozone cluster with 1000 datanodes and 10 s3gateway, and run two weeks
with heavy workload, and perf om and scm.
1. From om perf, getServiceList cost 63.75% cpu.
!screenshot-1.png!
2. From scm perf, queryNode come from om::getServiceList cost 33.20% cpu
!screenshot-2.png!
Now s3g create a client for each request. when create each RpcClient, s3g will
call ServiceInfoEx serviceInfoEx = ozoneManagerClient.getServiceInfo(),
getServiceInfo will call getServiceList. Then om and scm are busy with
getServiceList.
was:
I start a ozone cluster with 1000 datanodes and 10 s3gateway, and run two weeks
with heavy workload, and perf om and scm.
1. From om perf, getServiceList cost 63.75% cpu.
!screenshot-1.png!
2. From scm perf,
Now s3g create a client for each request. when create each RpcClient, s3g will
call ServiceInfoEx serviceInfoEx = ozoneManagerClient.getServiceInfo(),
getServiceInfo will call getServiceList. Then om and scm are busy with
getServiceList.
> Improve OM and SCM performance with 50% by avoid call getServiceList
> --------------------------------------------------------------------
>
> Key: HDDS-3745
> URL: https://issues.apache.org/jira/browse/HDDS-3745
> Project: Hadoop Distributed Data Store
> Issue Type: Sub-task
> Reporter: runzhiwang
> Assignee: runzhiwang
> Priority: Major
> Attachments: screenshot-1.png, screenshot-2.png
>
>
> I start a ozone cluster with 1000 datanodes and 10 s3gateway, and run two
> weeks with heavy workload, and perf om and scm.
> 1. From om perf, getServiceList cost 63.75% cpu.
> !screenshot-1.png!
> 2. From scm perf, queryNode come from om::getServiceList cost 33.20% cpu
> !screenshot-2.png!
> Now s3g create a client for each request. when create each RpcClient, s3g
> will call ServiceInfoEx serviceInfoEx = ozoneManagerClient.getServiceInfo(),
> getServiceInfo will call getServiceList. Then om and scm are busy with
> getServiceList.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
