[
https://issues.apache.org/jira/browse/OAK-6710?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16210806#comment-16210806
]
Tommaso Teofili commented on OAK-6710:
--------------------------------------
in r1812608 I've committed a first auto tuning algorithm which works like this:
* the commit rate (docs and MB per second) time series is described using
[single exponential
smoothing|http://www.itl.nist.gov/div898/handbook/pmc/section4/pmc431.htm]
* when the no. of segments exceeds the maximum (currently set to 20) the
averaged and maximum commit rates are compared, if the average exceeds the
maximum the latter is increased again using SES
Currently each merge policy resets the averages (commit rates and no. of
segments) every 50 merge attempts (hence the time series has no more than 50
points).
> CommitMitigated merge policy should not reduce performance significantly
> ------------------------------------------------------------------------
>
> Key: OAK-6710
> URL: https://issues.apache.org/jira/browse/OAK-6710
> Project: Jackrabbit Oak
> Issue Type: Improvement
> Components: lucene
> Affects Versions: 1.7.7
> Reporter: Vikas Saurabh
> Assignee: Tommaso Teofili
> Labels: performance, scalability
> Fix For: 1.8
>
>
> While running performance tests (internally) using latest unstable releases
> having {{CommitMitigatedTieredMergePolicy}} we observed significant drop in
> performance (OAK-6704).
> The policy, although bad for performance, showed dramatic drop in churn
> created in data store size. So, clearly, OAK-5192 did what it intended to do.
> Opening this issue to factor in drop in performance and then set the default
> back to {{CommitMitigatedTieredMergePolicy}}.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
