On 10/13/2015 09:48 AM, Ben Turner wrote:
----- Original Message -----
From: "Lindsay Mathieson" <lindsay.mathieson@xxxxxxxxx>
To: "Paul Cuzner" <pcuzner@xxxxxxxxxx>
Cc: "Ben Turner" <bturner@xxxxxxxxxx>, "gluster-users" <gluster-users@xxxxxxxxxxx>
Sent: Monday, October 12, 2015 10:44:49 PM
Subject: Re: Test results and Performance Tuning efforts ...
Thanks, good to know.
On 13 October 2015 at 11:57, Paul Cuzner <pcuzner@xxxxxxxxxx> wrote:
I don't think so. The workaround affects how glusterd is working. The
performance fix benefit with epoll is on the glusterfsd daemons - AFAIK.
Perhaps one of the devs can chime in to confirm the impact.
Hmm I think it will negate the MT epoll perf improvement. Try setting what Paul suggested and do gluster v info, if you see:
server.event-threads: 1
client.event-threads: 1
In your gluster v info output then I think it will negate things. @Shyam - can you confirm / deny this?
The change is to the *glusterd* vol file (by editing the same), so the
client and brick vol files should retain the event-threads as is.
-b
On Tue, Oct 13, 2015 at 2:53 PM, Lindsay Mathieson <
lindsay.mathieson@xxxxxxxxx> wrote:
On 13 October 2015 at 11:51, Paul Cuzner <pcuzner@xxxxxxxxxx> wrote:
add
# for epoll issue glusterd crash fix
option ping-timeout 0
option event-threads 1
to your glusterd.vol files (/etc/glusterfs/glusterd.vol)
Thanks, yah I saw that.
Won't that remove the performance benefits though?
--
Lindsay
--
Lindsay
_______________________________________________
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
http://www.gluster.org/mailman/listinfo/gluster-users
