[torquedev] Linux kernel/glibc ulimit strangeness

Garrick Staples garrick at usc.edu
Fri Nov 30 09:59:37 MST 2007


On Fri, Nov 30, 2007 at 04:48:23PM +0100, Bogdan Costescu alleged:
> On Fri, 30 Nov 2007, Chris Samuel wrote:
> 
> >1) In the Linux pbs_mom should we be setting RLIMIT_AS in addition 
> >to the the others so that these limits are enforced regardless of 
> >which allocation strategy is followed by the application ?
> 
> I think that this is a good idea. To go even further, I personally 
> don't find any value in setting something else than virtual memory, 
> because it's only this parameter which reflects truely how much memory 
> is used by a process. Setting f.e. data segment size says nothing 
> about the other memory usage aspects of the program (real code, shared 
> libs code, stack, etc.) which means that any scheduling based on it is 
> unreliable.

Fine by me.  Someone work up a patch?


> >2)  Would it be possible to have a configuration option to disable
> >setting ulimits for those who want to use them as guidelines but not
> >enforced (for non-expert users) ?
> 
> I don't quite get your meaning: how do you intend for a limit to be 
> used as a guideline if it's not set ?

Node allocation decision, but not a process limit.  In effect, it becames a
minimum memory, not a maximum memory.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://www.supercluster.org/pipermail/torquedev/attachments/20071130/620b0fcc/attachment.bin


More information about the torquedev mailing list