[Users] max number of processes on TACC Stampede login nodes

[Erik Schnetter]

10 processes is quite easy to reach. There are:

- make
- bash
- perl (for Cactus preprocessing)
- cc (driver)
- cc1 (actual compiler)
- as

and then there might be:

- another make (there's one per thorn)
- another bash

etc. etc.

-erik



On Fri, Dec 16, 2016 at 10:55 AM, Frank Loeffler <knarf at cct.lsu.edu> wrote:

> On Tue, Dec 13, 2016 at 02:30:32PM -0600, Roland Haas wrote:
>
>> rhaas at login3:~/tmp$ ulimit -a | grep processes
>> max user processes              (-u) 150
>>
>> which is reached (in my case) when compiling with the default -j
>> settings in simfactory on stampede (somewhere in PeTSc).
>>
>
> Since I had to look it up and others might want the same: stampede is
> currently configured to use
>  make -j8
> which is reasonable, I believe.
>
> And: 150 is indeed quite low. Maybe it would make sense to ask them to
> increase this a little if you can show that it does hamper parallel
> compilation.
>
> Nevertheless, I am surprised that you reach 150 processes with just 8
> parallel make sessions. Assuming you got some 'background' processes just
> by logging in, that would probably still mean more than 10 processes per
> make, which sounds high. Do you know why that number is so high?
>
> I also compiled a lot on stampede lately. I've never seen that problem.
> Could it be that your limit of 150 is taken by something else than the
> processes spawned by make?
> My "problem" on stampede is the limit of 5GB home. Despite $WORK being
> said to be optimized for large files, they advised me to compile there if
> $HOME isn't big enough. 5GB is already too small for two configurations...
>
> Frank
>
>
> _______________________________________________
> Users mailing list
> Users at einsteintoolkit.org
> http://lists.einsteintoolkit.org/mailman/listinfo/users
>
>


-- 
Erik Schnetter <schnetter at cct.lsu.edu>
http://www.perimeterinstitute.ca/personal/eschnetter/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.einsteintoolkit.org/pipermail/users/attachments/20161216/1de4ee26/attachment.html