cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Brandon Williams (JIRA)" <j...@apache.org>
Subject [jira] Commented: (CASSANDRA-1453) stress.java
Date Thu, 23 Dec 2010 21:07:46 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-1453?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12974740#action_12974740
] 

Brandon Williams commented on CASSANDRA-1453:
---------------------------------------------

Building this requires building Cassandra first because it depends on build/classes, which
doesn't seem like a necessary requirement to me, but I'm not sure.  I will make bin/stress
executable at commit, and I'm going to remove CompileThreshold.  It takes 80s for the JVM
to warmup with 200 threads and CompileThreshold at 1, but only 5-6s with it removed (though
a scad more CPU is used afterwards.)

> stress.java
> -----------
>
>                 Key: CASSANDRA-1453
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1453
>             Project: Cassandra
>          Issue Type: New Feature
>          Components: Tools
>            Reporter: Jonathan Ellis
>            Assignee: Pavel Yaskevich
>            Priority: Minor
>             Fix For: 0.7.1
>
>         Attachments: stress-v2.patch, stress.patch
>
>
> stress.py seems to scale poorly past a fairly small number of threads/processes.  (against
a 3-node, RF=1 cluster, I got 3x as much throughput with 4 machines running stress.py -t 32,
as running 1 with -t 128.  these were 8-core client machines, and -t 128 reported only 50%
cpu used.)
> since we ship with the thrift java api pre-built, this would also mean not making people
build thrift before using the stress test, which is at best a pain for newcomers and on Windows
a major hurdle.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message