[
https://issues.apache.org/jira/browse/TEZ-2849?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14960176#comment-14960176
]
Gopal V commented on TEZ-2849:
------------------------------
[~karams]: Can you test this patch with terasort on JDK7 and JDK8?
[~ozawa]: This patch will definitely break if you've got any changes from
porting the sort buffer to direct byte buffers (or on 0.5.x branch).
> Implement Specific Workaround for JDK-8026049 & JDK-8073093
> ------------------------------------------------------------
>
> Key: TEZ-2849
> URL: https://issues.apache.org/jira/browse/TEZ-2849
> Project: Apache Tez
> Issue Type: Sub-task
> Affects Versions: 0.8.1-alpha
> Reporter: Gopal V
> Assignee: Gopal V
> Attachments: TEZ-2849.1.patch, TEZ-2849.2.patch
>
>
> The CPU performance overhead of the sort operation is tied to the JIT
> handling of the aliased byte buffers without an explicit alignment (i.e a
> byte[] buffer referenced as an IntBuffer).
> https://bugs.openjdk.java.net/browse/JDK-8026049
> +
> https://bugs.openjdk.java.net/browse/JDK-8073093
> Workaround that JIT limitation while keeping in mind the modern CPU
> micro-code optimizations around alignment.
> http://lemire.me/blog/archives/2012/05/31/data-alignment-for-speed-myth-or-reality/
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
