[Gluster-users] Very bad performance /w glusterfs. Am I missing something?

[Joe Landman]

On 08/11/2011 10:36 AM, Jean-Francois Chevrette wrote:
> Hello everyone,
>
> I have just began playing with GlusterFS 3.2 on a debian squeeze
> system. This system is a powerful quad-core xeon with 12GB of RAM and
> two 300GB SAS 15k drives configured as a RAID-1 on an Adaptec 5405
> controller. Both servers are connected through a crossover cable on
> gigabit ethernet ports.
>
> I installed the latest GlusterFS 3.2.2 release from the provided
> debian package.
>
> As an initial test, I've created a simple brick on my first node:
>
> gluster volume create brick transport tcp node1.internal:/brick
>
> I started the volume and mounted it locally
>
> mount -t glusterfs 127.0.0.1:/brick /mnt/brick
>
> I can an iozone test on both the underlying partition and the
> glusterfs mountpoint. Here are my results for the random write test
> (results are in ops/sec):

[...]

> (sorry if the formatting is messed)
>
>
> Any ideas why I am getting such bad results? My volume is not even
> replicated or distributed yet!

You are not getting "bad" results.  The results from the local fs w/o 
gluster are likely completely cached.  This is a very small test, and 
chances are you'r IOs aren't even making it out to the device before the 
test completes.

The only test in your results which is likely generating any sort of 
realistic IO is that very last row and last column data size.

A 15k RPM disk will do ~300 IOPs, which is about what you should see per 
unit.  For a RAID1 across 2 such disks, you should get (depending upon 
how you built the RAID1 and what the underlying RAID system is), from 
150-600 IOPs in most cases.


-- 
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics, Inc.
email: landman at scalableinformatics.com
web  : http://scalableinformatics.com
        http://scalableinformatics.com/sicluster
phone: +1 734 786 8423 x121
fax  : +1 866 888 3112
cell : +1 734 612 4615