[Gluster-users] GlusterFS Performance gigE

[Chris Layton]

I agree something is wrong. 

Here is what I am getting on a more ISP grade setup (but no SAN sadly)
also with no Ethernet tuning :


dd if=/dev/zero of=/usr/local/sites/test.test bs=1M count=64
64+0 records in
64+0 records out
67108864 bytes (67 MB) copied, 0.754847 seconds, 88.9 MB/s
dd if=/dev/zero of=/usr/local/sites/test.test bs=1M count=1024
1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB) copied, 13.9273 seconds, 77.1 MB/s

I did some other testing using tiobench, iozone, and dt with similar
(albeit MUCH more detailed) results. 

The setup above is using cluster/replicate between 2 remote bricks
(raid10 SAS 15K). Also this was run during a low load window while there
was still some light load on the webservers so this would be slightly
higher if I could stop our webservers but that isn't a option for a few
more days at least. Also we use lsycnd (inotify based) to sync changed
files off to remote backup servers to overcome the limits on wirespeed
with gluster (our remote nodes are 100Mb) so that has some brick/disk
overhead as well.

One thing you could do to help troubleshoot is run a speed test and use
atop or sar tools to debug and also get a snapshot of the test for
review.

 
-- 
Chris Layton 
Senior Systems Engineer
WorldSpice Technologies 
chrisl at worldspice.net
901-843-9300

On Fri, 2010-09-10 at 13:16 -0300, Henrique Haas wrote:
> Hello people,
> 
> For those have *GlusterFS running over GigE, what is the throughput you have
> for write data?*
> I have 3 nodes, running GNU/Linux Ubuntu Server 10.04, GlusterFS 3.0.2.
> The network is GigE, dedicated switch.
> 
> I am experiencing maximum 2 MB/sec running a "dd" over a client mount point.
> 
> Thank you very much
> 
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users