Dave Hall wrote:
> Hi Raghavendra,
>
> Thanks for the response. I played with it a bit today, but I don't
> have any good news - see below.
>
> On Tue, 2010-01-05 at 19:18 +0400, Raghavendra G wrote:
>> * Do you run into same problems without performance translators
>> (write-behind, io-cache)?
>
> Unfortunately it does it with both disabled. I ran tcpdump the
> machine which was playing up and there didn't appear to be any
> traffic, either locally or to the other servers.
>
>> If you don't face any problem after removing performance
>> translators, Is it possible to pin point which among write-behind
>> and io-cache is causing the problem (Again by retaining only one
>> of write-behind/io-cache)? Also is it possible to find out
>> whether the client or server process that is using too much of
>> cpu ("top" may help here)?
>
> Sorry, I should have been clearer, it is the glusterfs client.
> glusterfsd never seems to go above 10% even under load. it is only
> 1 glusterfs process and it max's out a whole core.
>
Is it always the same client that is showing this problem
or do the clients on all machines have CPU usage shoot-ups?
If it is always the same machine, will it be possible to run the
"glusterfs" client process under valgrind on this system? It'll slow
it down but should help us figure out where the CPU usage is going.
You can do this by using the following command to start up the
glusterfs client process:
$ valgrind --tool=callgrind <path-to-glusterfs> -f
<path-to-client-volfile> -L NONE --no-daemon <client-mount-point>
This will run the glusterfs process in the foreground while you
can access the mount point from a different shell session.
You can then monitor the CPU while this is running and press
CTRL+C for valgrind once the high CPU usage phase has passed.
Then mail us the *.out file generated by valgrind.
-Shehjar
> Cheers
>
> Dave
>
> _______________________________________________ Gluster-users
> mailing list Gluster-users at gluster.org
> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
