[Gluster-users] Way to improve GlusterFS performance for serving rails applications?

Horacio Sanson hsanson at gmail.com
Sat Oct 30 16:11:26 UTC 2010


On Saturday 30 October 2010 01:27:16 Joe Landman wrote:
> On 10/29/2010 12:23 PM, Horacio Sanson wrote:
> > The problem is that if I run any rails application when stored in the
> > gluster volume it take minutes to start and then each page load take
> > 5~10 secs. If instead I run the rails app from the local hard disk in
> > the same machine it goes fast as thunder.
> 
> This sounds like a locking or a stat issue ... though there could be
> something else going on.
> 
> > To give an idea here is some output from rails log file:
> > 
> > # Using gluster to store the rails application takes 5.6 secs to render a
> > page Completed in 5645ms (View: 2948, DB: 46) | 200 OK
> > 
> > # Using local storage it take 400ms to render the same page
> > Completed in 401ms (View: 296, DB: 34) | 200 OK
> > 
> > All machines (both storage and rails app) are connected to the same
> > switch via 1GB ethernet and there is no other traffic in that switch (is
> > isolated).
> > 
> > The questions:
> > 
> > Am I doing something wrong or missing something or is this expected?
> > 
> > Are there any options I can set when mounting the partition to improve
> > performance?
> 
> What do your client configs look like?

I use gluster 3.1 so I did not write any configuration files. But if I am 
correct the vol files inside he glusterd folder are the client config files:

volume opt-client-0
    type protocol/client
    option remote-host 192.168.4.86
    option remote-subvolume /opt
    option transport-type tcp
end-volume

volume opt-client-1
    type protocol/client
    option remote-host 192.168.4.87
    option remote-subvolume /opt
    option transport-type tcp
end-volume

volume opt-replicate-0
    type cluster/replicate
    subvolumes opt-client-0 opt-client-1
end-volume

volume opt-write-behind
    type performance/write-behind
    subvolumes opt-replicate-0
end-volume

volume opt-read-ahead
    type performance/read-ahead
    subvolumes opt-write-behind
end-volume

volume opt-io-cache
    type performance/io-cache
    subvolumes opt-read-ahead
end-volume

volume opt-quick-read
    type performance/quick-read
    subvolumes opt-io-cache
end-volume

volume opt
    type debug/io-stats
    subvolumes opt-quick-read
end-volume

-- 
regards,                                                                                                                                                                                                       
Horacio Sanson



More information about the Gluster-users mailing list