[Gluster-users] Introduction and request for help

[Kevin Carpenter]

Hello community.

I'm Kevin Carpenter, an infrastructure architect at my company.  I'm
currently looking for a scalable parallel file system for usage with our
High Performance Computing environment.  Like most geeks, I rather like to
try things out rather than just do paper reviews so have built a two node
cluster on my home system.

I am using the config files from the sample Replicated system at Gluster.org:

http://www.gluster.org/docs/index.php/Automatic_File_Replication_%28Mirror%29_across_Two_Storage_Servers

Only modifications:  Different directory statements in the glusterfsd.vol
file to map to locally available space on each server, and of course,
different server names in the glusterfs.vol file.

I'm using the gluster 2.0.4 distribution.  Both nodes are source file
built and run on Gentoo distributions.  Node A is running a 2.6.29 kernel,
Node B is running a 2.6.30 kernel.  Both nodes are running clients locally
mounted:

mount -t glusterfs localhost /glusterfs

Both Node A and Node b use JFS for there underlying file systems.

Everything works!

As a test case, I'm running my weekly rsync based backup pointing the
target at /glusterfs/weekly-backup.  All files being backed up show up on
both systems under /glusterfs - all appears functionally fine.

Alas... performance is 1/10th normal.  I timed having Node B do the same
rsync backup of Node A (yes, to an empty directory) across the network and
it runs in about 55 minutes real time.  When I have Node A perform the
rsync to its local /glusterfs, allowing gluster to replicate the files to
node B, it takes over 550 minutes (551 to be more exact) with virtually
all the time being ate by the glusterfsd and glusterfs processes (~10
hours of CPU time for each, on both systems).

I suspect this isn't normal, so wanted to ask this forum for advice.

Thanks,

Please let me know if more information is needed.

Kevin