You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@omid.apache.org by "Daniel Dai (JIRA)" <ji...@apache.org> on 2016/08/01 21:52:20 UTC
[jira] [Comment Edited] (OMID-50) Provide an option to reduce
tso-server CPU usage
[ https://issues.apache.org/jira/browse/OMID-50?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15402886#comment-15402886 ]
Daniel Dai edited comment on OMID-50 at 8/1/16 9:51 PM:
--------------------------------------------------------
I don't want to make every wait strategy configurable since it will increase the test matrix. Actually TimeoutBlockingWaitStrategy in RequestProcessorImpl should not be changed since it will result infinite loop otherwise. I think only two combination makes sense:
High throughput: Use BusySpinWaitStrategy for PersistenceProcessorImpl/ReplyProcessorImpl/RetryProcessorImpl
Low cpu: Use BlockingWaitStrategy for PersistenceProcessorImpl/ReplyProcessorImpl/RetryProcessorImpl
I only open these two options in the patch.
was (Author: daijy):
I don't want to make every wait strategy configurable since it will increase the test matrix. Actually TimeoutBlockingWaitStrategy in RequestProcessorImpl should not be changed since it will result infinite loop otherwise. I think only two combination makes sense:
High throughput: Use BusySpinWaitStrategy for PersistenceProcessorImpl/ReplyProcessorImpl/RetryProcessorImpl
Low cpu: Use BlockingWaitStrategy for PersistenceProcessorImpl/ReplyProcessorImpl/RetryProcessorImpl
I only open these two options in the patch.
> Provide an option to reduce tso-server CPU usage
> ------------------------------------------------
>
> Key: OMID-50
> URL: https://issues.apache.org/jira/browse/OMID-50
> Project: Apache Omid
> Issue Type: Improvement
> Reporter: Daniel Dai
> Assignee: Daniel Dai
> Fix For: 0.8.3.0
>
> Attachments: OMID-50-1.patch
>
>
> Currently tso-server use 300% cpu even at idle time. The reason for this 300% cpu usage is due to the BusyWaitStrategy used in disruptor. While this is good for throughput, it wastes resources in many use cases. In this ticket, I provide a config to use a different wait strategy and reduce the cpu usage with the option on.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)