[
https://issues.apache.org/jira/browse/CASSANDRA-6843?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13947806#comment-13947806
]
Benedict commented on CASSANDRA-6843:
-------------------------------------
I had no intention of getting into a proper discussion about the results, as
they are not remotely rigorous, nor are they a main justification for this
patch; more extensive testing is necessary, and this patch mostly enables other
adaptations, so I didn't see the point of wasting time on either count;
however, since you ask, I was testing a single node with:
cassandra-stress write n=1000000 -key dist=uniform\(1..10000\) -col
n=fixed\(1\) size=fixed\(10000\) -mode thrift -rate threads=50
Which yielded rates of 22Kop/s without vs 50Kop/s with; with rates dropping to
40Kop/s throughput for this patch when number of records inserted was bumped to
5M, as the CL became a bottleneck.
> Reclaim overwritten off-heap memtable space early
> -------------------------------------------------
>
> Key: CASSANDRA-6843
> URL: https://issues.apache.org/jira/browse/CASSANDRA-6843
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Benedict
> Assignee: Benedict
> Labels: performance
> Fix For: 3.0
>
>
> Perform GC on the off-heap memtable arenas
--
This message was sent by Atlassian JIRA
(v6.2#6252)
