[torquedev] memory limit enforcement by pbs_mom - REQUEST FOR FEEDBACK

Åke Sandgren ake.sandgren at hpc2n.umu.se
Tue Feb 7 04:45:39 MST 2006


On Tue, 2006-02-07 at 15:12 +1100, Chris Samuel wrote:
> On Wednesday 01 February 2006 07:18, Dave Jackson wrote:
> 
> >  Currently, the pbs_mom enforces memory limits specified with '-l
> > pmem=X' but does not enforce memory limits specified with '-l mem=X'
> > This is confusing for some users.  I propose that we modify
> > mom_set_limits() to enforce stack and data segment limits if pmem is
> > specified or mem is specified and the job is serial.
> 
> Can I ask for a compile time option to disable the enforcement of memory 
> limits ?
> 
> I want to be able to set defaults for memory limits for jobs in the queues 
> (the vast number of our users have no idea how much RAM they're going to use) 
> and then that lets people set different limits if they need them and we can 
> have the scheduler work things out sensibly.
> 
> Otherwise Moab treats no limit specified as a job wanting 0 RAM (not 
> unreasonably) and so if someone has requested (total RAM - 1MB) it'll shove 
> another job with no limit specified onto that node, potentially causing 
> chaos.
> 
> But with the current enforcement I can't begin to contemplate doing this as 
> dog knows how many users jobs would get massacred by it..

If you have a memory and smp homogeneous system you could probably get
away with setting default limits as
default_vmem = (total_mem + total_swap) * k / ncpu
(where k ~= .8 to leave space for kernel and system processes)
and default_mem = total_mem * k / ncpu


More information about the torquedev mailing list