[Gluster-users] Performance loss from 3.4.2 to 3.6.2

[ML mail]

Thanks Pranith for the details. So with that option one would be trading data consistency with performance. I am now interested to hear about Nico's new tests with this option disabled... 

     On Thursday, February 12, 2015 9:23 AM, Pranith Kumar Karampuri <pkarampu at redhat.com> wrote:
   

  
 On 02/12/2015 01:17 PM, ML mail wrote:
  
  Dear Pranith 
  I would be interested to know what the cluster.ensure-durability off option does, could you explain or point to the documentation?  
 By default replication translator does fsyncs on the files at certain times so that it doesn't lose data when the filesystem crashes and changelogs that afr stores are corrupted. This is introduced from 3.5.x on wards.
 
 Pranith
 
  
  Regards ML 
 
       On Thursday, February 12, 2015 8:24 AM, Pranith Kumar Karampuri <pkarampu at redhat.com> wrote:
   
 
 
 On 02/12/2015 04:37 AM, Nico Schottelius wrote:
 > Hello,
 >
 > switching from 3.4.2 to 3.6.2 reduces our average test performance dramatically.
 >
 > Our test setup: directly connected 1 GBit/s hosts setup with:
 >
 >      rm -rf /home/gluster/.glusterfs/
 >      rm /home/gluster/*
 >      setfattr -x trusted.glusterfs.volume-id  /home/gluster/
 >      setfattr -x trusted.gfid /home/gluster/
 >      gluster volume create xfs-plain replica 2 transport tcp vmhost1-cluster1:/home/gluster vmhost2-cluster1:/home/gluster
 >      gluster volume start xfs-plain
 >
 > Afterwards we run our near real world test
 >
 >      mount -t glusterfs vmhost1-cluster1:/xfs-plain /mnt/gluster/
 >      while true; do dd if=redmine-from-ceph-20150204 of=/mnt/gluster/testvm bs=1M; rm /mnt/gluster/testvm; done
 >
 > The results are
 >
 >      3.4.2: ~71-72 MiB/s    [ubuntu 14.04]
 >      3.6.2: ~59-64 MiB/s    [gluster-3.6 ppa]
 >
 > We have removed 3.6.2 and re-installed 3.4.2 and can consistently reproduce
 > these numbers over hours of testing.
 Could you repeat the test on 3.6.2 with 'volume set <volname> 
 cluster.ensure-durability off' option and post the results.
 
 Pranith 
 >
 > Is there any configuration change that we need to incorporate to run
 > 3.6.2 faster or is this a known problem?
 >
 > Cheers,
 >
 > Nico
 >
 
 _______________________________________________
 Gluster-users mailing list
 Gluster-users at gluster.org
 http://www.gluster.org/mailman/listinfo/gluster-users
  
 
      
  
 _______________________________________________
Gluster-users mailing list
Gluster-users at gluster.org
http://www.gluster.org/mailman/listinfo/gluster-users 
 
 

   
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.gluster.org/pipermail/gluster-users/attachments/20150212/99fcdac5/attachment.html>