[
https://issues.apache.org/jira/browse/TS-1674?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13570133#comment-13570133
]
Uri Shachar commented on TS-1674:
---------------------------------
I did some quick benchmarks with a sample program -- At worse, it seems like
using atomic long longs is ~6-7 times slower then regular long longs (This is
obviously dependant on arch/compiler -- I used GCC over 64bit x86).
My guess would be that the impact on ATS wouldn't be noticeable (since there
will be some improvement from the code simplification). I'll try to cobble
together a working patch and check it out....
> TSStatIntDecrement is broken
> ----------------------------
>
> Key: TS-1674
> URL: https://issues.apache.org/jira/browse/TS-1674
> Project: Traffic Server
> Issue Type: Bug
> Components: TS API
> Reporter: Brian Geffon
> Assignee: Brian Geffon
> Fix For: 3.3.1
>
>
> ATS keeps a thread-local copy of the stat value and aggregates it
> periodically and on demand when stats are being collected. However the
> decrement function doesn't let a value go negative. So in a thread that has
> not incremented this stat previously, the decrement call "fails". This
> negative value check is logically flawed as it doesn't account for the
> per-thread distribution of the stats.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
