[
https://issues.apache.org/jira/browse/COUCHDB-2097?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14052879#comment-14052879
]
Paul Joseph Davis commented on COUCHDB-2097:
--------------------------------------------
Benoit,
In general you're correct that having the two couch_file processes would be
faster than a single couch_file. But this was concerned with the entire stack
change which includes the change to reference counting which is required for
the changes to couch_server which makes the entire system significantly faster
and more stable under load.
As to figures, Mike Wallace already did an analysis of this that I linked to
earlier in this ticket. The new version is much better for reads with the two
fd version slightly better for writes. As we've discussed previously, we can
look into revamping the reference counting post-merge and then thinking about
getting this two fd version back for the minor increase in write speeds.
Given that there isn't a direct regression from this work I'd consider this
ticket closed. If you like we can open a new ticket to investigate rewriting
the reference counting code so that we can get the two fd split back in eke out
the extra bit of write performance. But given that's new development work I
don't see why it should block the merge.
> Avoid performance regression with a single fd
> ---------------------------------------------
>
> Key: COUCHDB-2097
> URL: https://issues.apache.org/jira/browse/COUCHDB-2097
> Project: CouchDB
> Issue Type: Improvement
> Security Level: public(Regular issues)
> Components: BigCouch
> Reporter: Paul Joseph Davis
> Labels: release
> Attachments: mike-wallace-fd-benchmarks.txt
>
>
> Part of one of our large enhancements required that we remove a CouchDB
> performance optimization on having two file handles to each .couch file. We
> need to make sure that this doesn't negatively impact performance.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
