[
https://issues.apache.org/jira/browse/ARTEMIS-1162?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16010156#comment-16010156
]
ASF GitHub Bot commented on ARTEMIS-1162:
-----------------------------------------
Github user michaelandrepearce commented on the issue:
https://github.com/apache/activemq-artemis/pull/1268
@franz1981 and here in lies the issue, that the project doesn't currently
have a set of agreed performance metrics for different end 2 end scenarios,
that are run automatically and reported each release.
I like micro benchmarks as they help isolate a problem, and quickly test if
the changes look promising and the best we have today.
As we move to look at other perf improvements without the end 2 end
scenarios reporting perf metrics version to version, some changes may start to
even step on each other, e.g. one fix may actually regress another, e.g. the
classical, batching can improve throughput very easily but trades it for
latency.
> Make new Adapting TimedBuffer and old Fixed TimedBuffer configurable
> --------------------------------------------------------------------
>
> Key: ARTEMIS-1162
> URL: https://issues.apache.org/jira/browse/ARTEMIS-1162
> Project: ActiveMQ Artemis
> Issue Type: Improvement
> Affects Versions: 2.1.0
> Reporter: Michael Andre Pearce
> Fix For: 2.2.0
>
>
> New Adapting TimedBuffer was added and replaced existing Fixed TimedBuffer.
> This could cause behaviour/performance change of journal IO (and hopefully
> for the better).
> As this hasn't been run significantly in a real world production environment,
> to avoid causing any production deployment issues having unexpected behaviour
> changes, or to allow managed rollout and rollback.
> We would like to add back the fixed timed buffer (default), and make it
> configurable to the new improved timed buffer.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
