Darren,<div> Replicate performs a lookup to both servers to ensure the files are in sync. Once cached, it should only perform mtime checks with the servers. Can you check if you are not seeing performance gain even in a non replicated setup? Just as a step in debugging..</div>
<div><br></div><div>Avati<br><br><div class="gmail_quote">On Tue, Jun 28, 2011 at 3:50 PM, Darren Austin <span dir="ltr"><<a href="mailto:darren-lists@widgit.com">darren-lists@widgit.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
[ This didn't seem to come through the first time - sorry if you get two copies :) ]<br>
<br>
----- Forwarded Message -----<br>
From: "Darren Austin" <<a href="mailto:darren-lists@widgit.com">darren-lists@widgit.com</a>><br>
To: <a href="mailto:gluster-users-bounces@gluster.org">gluster-users-bounces@gluster.org</a><br>
Sent: Tuesday, 28 June, 2011 11:05:38 AM<br>
Subject: Performance + Translators<br>
<br>
Hi,<br>
I've been measuring the performance of a GlusterFS backed Apache<br>
instance and noticed some quite concerning performance issues.<br>
<br>
Should the GlusterFS client be hitting BOTH the replicated servers for<br>
every httpd request that comes in?<br>
For every request that is made, the client seems to be requesting the<br>
file from both the replicated servers in the cluster - is this a<br>
situation which is normal?<br>
<br>
Also, even though the requested file is exactly the same every time,<br>
Gluster seems to be re-requesting it from the server(s) for every httpd<br>
request.<br>
<br>
I've read that the quick-read translator is supposed to cache file<br>
access requests to increase performance, but it doesn't seem to have any<br>
effect here - but that may be due to a configuration error on my part.<br>
<br>
I'm not even sure if the quick-read translator is being enabled - and if<br>
it has to be enabled in some way on the client?<br>
I've attached logs from the point of setting up a new volume and<br>
mounting it from the client - even though the vol file being sent to the<br>
client contains references to the quick-read translator, I don't know if<br>
it's actually being used - is there some way to tell?<br>
<br>
Also, as a test, I used 'volume set XXX performance.quick-read off', and<br>
re-mounted from the client. The same performance tests didn't change -<br>
so either the quick-read translator is having no effect, or it's not<br>
being enabled on the client.<br>
<br>
I'm pretty new to GlusterFS, so can anyone offer some insight into how<br>
client side translators get enabled/disabled? My impression from the<br>
documentation is that you set their options using 'volume set' on the<br>
server, and the client picks them up - but that doesn't seem to be the<br>
case here.<br>
<br>
Thanks,<br>
Darren.<br>
<font color="#888888"><br>
--<br>
Darren Austin - Systems Administrator, Widgit Software.<br>
Tel: +44 (0)1926 333680. Web: <a href="http://www.widgit.com/" target="_blank">http://www.widgit.com/</a><br>
26 Queen Street, Cubbington, Warwickshire, CV32 7NA.<br>
<br>
</font><br>_______________________________________________<br>
Gluster-users mailing list<br>
<a href="mailto:Gluster-users@gluster.org">Gluster-users@gluster.org</a><br>
<a href="http://gluster.org/cgi-bin/mailman/listinfo/gluster-users" target="_blank">http://gluster.org/cgi-bin/mailman/listinfo/gluster-users</a><br>
<br></blockquote></div><br></div>