[
https://issues.apache.org/jira/browse/HADOOP-14031?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15843599#comment-15843599
]
Daryn Sharp commented on HADOOP-14031:
--------------------------------------
This will be a port of internal production enhancements that significantly
improved performance (up to 4X throughput) under heavy and/or abusive loads.
Ex. Heavy file creation or globbing from wide multi-thousand task jobs. These
changes tamed client latencies from multi-hundred ms into generally
single-digit ms.
> Reduce fair call queue performance impact
> -----------------------------------------
>
> Key: HADOOP-14031
> URL: https://issues.apache.org/jira/browse/HADOOP-14031
> Project: Hadoop Common
> Issue Type: Bug
> Components: ipc
> Affects Versions: 2.7.0
> Reporter: Daryn Sharp
> Assignee: Daryn Sharp
>
> The fair call queue has performance deficits that create an illusion of
> reasonable performance under heavy load. However, there is excessive lock
> contention, priority inversion, and pushback/reconnect issues that combine to
> create an artificial rate-limiting on the ingress.
> The result is server metrics look good, call queue looks low, yet clients
> experience dismal latencies.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
