Ralph,
I'll be happy to test whatever you send. I'll also setup the same
revision on a few other systems and see if any patterns emerge wrt
reproducibility.
Greg
On 7/2/14 10:50 PM, Ralph Castain wrote:
Rats - this is one tough race condition. I've been cycling it since your note,
without
Rats - this is one tough race condition. I've been cycling it since your note,
without hitting one problem.
I'm running Centos6 on a dual processor, 6-core Xeon, with a dedicated disk
installed on the node. Apparently, that seems to be adequate to mitigate the
problem.
I'll keep trying to find
Ralph,
I finally got some time to look into this again. I can reproduce the
issue when running against the latest nightlies for 1.8.2 (r32119) and
1.9 (r32113).
While verifying that, I noticed that testing on a different system
required a much larger number of iterations (~2000), while on t
Hi Greg
I've been running your script over and over again with the current 1.8.2 and
svn developer's trunk, and I cannot get a failure. It just merrily runs.
Could you tell me how you configured OMPI to get this behavior?
Thanks
Ralph
On Jun 10, 2014, at 11:59 AM, Ralph Castain wrote:
> Ouch
Ouch - I'll try to chase it down.I'm unaware of anyone passing a significant
amount of data via stdin before, so it's quite possible this has been around
for awhile. Normally one avoids that practice.
On Jun 10, 2014, at 11:36 AM, Greg Thomsen wrote:
> All,
>
> I believe I've found a bug in
