[molpro-user] memory usage
keythree at gmail.com
Tue May 8 20:04:22 BST 2012
Thank for the help. That clears everything up. My jobs seem to be running
as expected now.
Keith McLaughlin, M.S.
University of South Florida
Dept. of Physics/Chemistry
On Fri, May 4, 2012 at 10:27 AM, Neeraj Rai <neerajrai at gmail.com> wrote:
> Memory is per process so if you are 128 cores then you will need
> 128*7000 mW of memory. If your job needs more memory then you can reduce
> number of cores that you are requesting.
> On Thu, May 3, 2012 at 6:51 PM, Keith McLaughlin <keythree at gmail.com>wrote:
>> Hi all,
>> I'm having a lot of problem completing jobs on blacklight, which is a
>> machine at the Pittsburgh Supercomputing Center.
>> I'm trying to run a job requesting 7000,m. That is, 7000 megawords, which
>> I thought would be 54 gigabytes. My job winds up dying (when run on 128
>> cores = 8 blades @ 128 GIGS per blade, which means I have access to a
>> terabyte of memory. Yet, my job is killed with the following error message:
>> Creating: host=bl0.psc.teragrid.o=>> PBS: job killed: cpuset
>> memory_pressure 139111 reached/exceeded limit 1 (numa memused is 1073599536
>> Is this behavior understood? Why would my job possibly be using that much
>> memory? I've been in contact with PSC support, but they believe this is a
>> Molpro issue.
>> Keith McLaughlin, M.S.
>> University of South Florida
>> Dept. of Physics/Chemistry
>> Molpro-user mailing list
>> Molpro-user at molpro.net
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Molpro-user