Perhaps you could try a server built with --without-bmi-tcp, just to get
a feel for the impact of the TCP polling. That would help us understand
how close you're getting to what you expect from the rest of the system.
Regards,
Rob
Salmon, Rene wrote:
Hi,
Searched through the list archives and it found some posts that show IB
performance around 400MB/s. Also found the post that states in order to
get best performance over IB to configure with "--without-bmi-tcp". I
need tpc as well as IB so I this is not an option for me.
This is a simple setup. Single pvfs server with a single pvfs client.
DDR IB connect the two as well as 1Gb ethernet. On the server there is
an xfs filesystem that can write at about 500MB/sec that is what the
pvfs server will use.
I would expect that my bottle neck would be the xfs filesystem at
500MB/sec. When I run pvfs-cp on the client I only get around 260
MB/sec bandwidth:
hpcxe005(salmr0)37:/usr/local/opt/pvfs-2.6.3/bin/pvfs2-cp
-t /test/pvfs2/testfile1 /test/pvfs2/testfile5
Wrote 655360000 bytes in 2.355496 seconds. 265.336905 MB/seconds
Any ideas on how to get this to perform a bit better?
client mount command:
---------------------
mount -t pvfs2 ib://hpcxe007:3335/pvfs2-fs /test/pvfs2
server pvfs2-fs.conf:
----------------------
<Aliases>
Alias hpcxe007 ib://hpcxe007:3335,tcp://hpcxe007:3334
</Aliases>
pvfs2-server.conf-hpcxe007:
---------------------------
StorageSpace /pvfs2-storage-space
HostID "ib://hpcxe007:3335,tcp://hpcxe007:3334"
Thanks
Rene
_______________________________________________
Pvfs2-users mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
_______________________________________________
Pvfs2-users mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
