You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by "Chris Burroughs (JIRA)" <ji...@apache.org> on 2013/08/26 16:25:51 UTC

[jira] [Created] (CASSANDRA-5937) check read performance between 1.2.5 and 1.2.8

Chris Burroughs created CASSANDRA-5937:
------------------------------------------

             Summary: check read performance between 1.2.5 and 1.2.8
                 Key: CASSANDRA-5937
                 URL: https://issues.apache.org/jira/browse/CASSANDRA-5937
             Project: Cassandra
          Issue Type: Test
            Reporter: Chris Burroughs
            Assignee: Ryan McGuire
            Priority: Minor


We upgraded from 1.2.5 to 1.2.8 on 08-02 and saw a significant increase in read latency:
 * median: < 1ms to > 1ms
 * 75p: ~ 1 ms to > 3.5 ms
 * 95p: ~ 3 ms to ~ 8 ms

Cluster is a 2 DC cluster with about 20 nodes per DC (using GossipingPropertyFileSnitch).  All queries are for one Keyspace filled with skinny rows so we have Row cache enabled and the Key cache disabled.  The row cache hit is close to 70%, which makes the magnitude of the new median/75th hard to understand.

I have not been able to demonstrate a regression with stress on either a single  or dozen node cluster.  For internal reasons invovling client side problem we have not tried rolling back so I'm not positive it's a cassandra code problem (as opposed to something else we did/configured).

Thanks for checking it out!

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira