Hello!
On 27.02.12 08:10, Venkateswara Rao Dokku wrote:
> Our's is a customized OFED stack[Our own Driver specific library and
> Kernel drivers for the h/w], we use IMB tests for testing the same.
> All the tests [PingPong, Exchange.. etc] stalls after some time with
> no errors.
I found similar issues with a MLNX_OFED_LINUX-1.5.3 on top of RHEL6.2.
So far I found a note in the HP documentation [1] about a buggy mlx4
driver introduced in OFED 1.5.3. The workaround seems to help OpenMPI
1.5.x too, but still not perfect stable.
[1]
http://h10025.www1.hp.com/ewfrf/wc/document?cc=us&lc=en&dlc=en&tmp_geoLoc=true&docname=c03113904
HTH
Beat
--
\|/ Beat Rubischon <beat_at_[hidden]>
( 0-0 ) http://www.0x1b.ch/~beat/
oOO--(_)--OOo---------------------------------------------------
Meine Erlebnisse, Gedanken und Traeume: http://www.0x1b.ch/blog/
|