hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Nicolas Liochon (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-9775) Client write path perf issues
Date Sat, 19 Oct 2013 15:41:44 GMT

    [ https://issues.apache.org/jira/browse/HBASE-9775?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13799926#comment-13799926

Nicolas Liochon commented on HBASE-9775:

bq. Sure I can! Do you have a patch for 0.94? Or only trunk so far?
I would love to see how 96 RC 5 behaves on the test suite you ran a few months ago on HBASE-6295,
as well as 0.94.12: when you ran the tests back them, we were seeing a 50% perf improvement
on the write path. I had something similar on a ~10 nodes cluster. Doing the test again would
help to see if we have a perf problem or a scalability problem.

bq. Nicolas Liochon One of the things you asked for earlier was running this with CMS. Just
did it with a much smaller heap (19g vs 29.9g) it still fails at about the same place, though
it takes longer to get to the same spot.
It was actually Nick who asked, but the result is very interesting. As well, in the logs you
have ~1s stop-the-world if I understand correctly? Our defaults sockets  timeouts are usually
around 30 th seconds, but I wonder if we don't have here the impact of hbase.rpc.timeout (default
to 60s): we lower the timeout during the retries iirc. If it's possible, I would love to see
the full logs of one RS. The responseTooSlow is strange too compared to the stop-the-world

> Client write path perf issues
> -----------------------------
>                 Key: HBASE-9775
>                 URL: https://issues.apache.org/jira/browse/HBASE-9775
>             Project: HBase
>          Issue Type: Bug
>          Components: Client
>    Affects Versions: 0.96.0
>            Reporter: Elliott Clark
>            Priority: Critical
>         Attachments: Charts Search   Cloudera Manager - ITBLL.png, Charts Search   Cloudera
Manager.png, hbase-9775.patch, job_run.log, short_ycsb.png, ycsb_insert_94_vs_96.png, ycsb.png
> Testing on larger clusters has not had the desired throughput increases.

This message was sent by Atlassian JIRA

View raw message