[
https://issues.apache.org/jira/browse/COUCHDB-1009?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12976762#action_12976762
]
Robert Newson commented on COUCHDB-1009:
----------------------------------------
"it will be better to use it with even larger sizes (512Kb, 1Mb for example). "
Except that this will block concurrent writers from making progress (I need
*you* to transmit 1mb before I can transmit my 1mb), etc.
Increasing from 4k seems a no-brainer to me though when I experimented with
larger buffer sizes a while ago (a year?) I found that things were little
improved, matching your numbers.
I would be interested in knowing if it helps with compaction, though, since
reading and writing larger attachment blocks should really help.
> Make couch_stream buffer configurable
> -------------------------------------
>
> Key: COUCHDB-1009
> URL: https://issues.apache.org/jira/browse/COUCHDB-1009
> Project: CouchDB
> Issue Type: Improvement
> Components: Database Core
> Environment: trunk
> Reporter: Filipe Manana
> Assignee: Filipe Manana
> Priority: Trivial
> Attachments: COUCHDB-1009.patch
>
>
> The couch_stream buffer is hardcoded to 4Kb.
> This value should be configurable. Larger values can improve write and
> specially read performance (if we write larger chunks to disk, we have higher
> chances of reading more contiguous disk blocks afterwards).
> I also think it's a good idea to change the default value from 4Kb to
> something higher (64Kb for e.g.).
> Patch attached
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
