[
https://issues.apache.org/jira/browse/CASSANDRA-8160?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14329875#comment-14329875
]
Matt Stump commented on CASSANDRA-8160:
---------------------------------------
For several large low-latency deployments the OS wasn't aggressively moving
SSTables into the cache despite available memory. In these instances the data
set was smaller than available memory and the cluster was sized for the read
ops/second. Forcing the SSTables into the buffer cache by cating them out to
/dev/null had an immediate impact on the 95% latency metric. This was an effort
to force the OS to aggressively pre-cache. I was experimenting with
FADV_SEQUENTIAL vs WILLNEED because the documentation and implementation
differences between releases is unclear. I'm leaving the patch here in it's
current state because I get pulled away often and hadn't had the time to finish
the necessary testing to justify it's inclusion. I'll try to finish the work if
I can.
> CF level option to call posix_fadvise for sstables on creation and startup
> --------------------------------------------------------------------------
>
> Key: CASSANDRA-8160
> URL: https://issues.apache.org/jira/browse/CASSANDRA-8160
> Project: Cassandra
> Issue Type: New Feature
> Components: Core
> Reporter: Matt Stump
> Assignee: Branimir Lambov
> Priority: Minor
> Fix For: 2.1.4
>
> Attachments: trunk-8160.txt
>
>
> We should have a CF level configuration with will result in posix_fadvise
> being called for sstables for that CF. It should be called on node startup
> and for new sstables. This should be configurable per CF to allow for some
> CFs to be prioritized above others. Not sure if we should use
> POSIX_FADV_SEQUENTIAL or POSIX_FADV_WILLNEED.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
