[Gluster-users] Performance question
arnold at arnoldarts.de
Wed Feb 15 22:06:05 UTC 2012
On Monday 13 February 2012 06:33:47 you wrote:
> I would love to see the numbers you get for dbench....as I have been doing
> exstensive testing with iozone and would like to be able to compare numbers
> to see that we are coming to the same conclusions. If you would like to
> see the basics of what I have been testing please look at
> http://community.gluster.com and look for the performance question. You
> can add your information there as well so we have it documented for easy
> look up by others. If there is anything I can to do help out with your
> testing please let me know.
@Bryan and Brian:
Due to time-management reasons I only got to do the tests yesterday. And it
proved to be the same as I have encountered before in physics:
A quick test is done quite fast. But when you look at the preliminary results
and evaluate a real test-program, you have to do it right in the first place.
My results from yesterday evening are incomplete and seem to have a lot of
unreproducible artifacts. What was interesting is that pure-linux-nfs from
node2 to node1 had roughly the same results as glusterfs on node2 to a single-
brick volume on node1...
I will do some real tests on the cluster at work. The same cluster I just
free'd from using gfs2:-)
So the comparisons would be:
1. single local disk
2. pure nfs between the nodes
3.1 glusterfs (aka fuse-mount) with single-brick volumes across network
3.2 glusterfs with dual-brick/dual-node distributed volume
3.3 glusterfs with dual-brick/dual-node replicated volume
4.X same as 3.x but with glusterfs-mount from node3
5.X same as 3.x but with nfs-mount from node3
6. (bonus) 3x2 three nodes, two bricks per node, replicate2, distributed big
volume mounted via glusterfs on one node. + mounted on all nodes. + mounted on
one virtual machine via glusterfs and nfs.
All of that with the same dbench-call and a reasonable test-time. At least
this time my boss won't complain that I introduce a new technology without
first presenting a white-paper. ;-)
Do you have more inputs for the test-regime?
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 198 bytes
Desc: This is a digitally signed message part.
More information about the Gluster-users