[torqueusers] how to interpret resources_used.mem?

David Singleton David.Singleton at anu.edu.au
Fri Aug 29 06:44:51 MDT 2008


I didn't see a response to this.  resources_used.mem should be
the maximum memory used by the job during execution.  There are
numerous ways in which this can be inaccurate.  For a start, the
MOM only samples the resource usage of processes at an interval
(which I believe would usually be of the order of a minute) so
fluctuations may be missed.  Also in the presence of shared memory
segments (or other shared pages), PBS will usually overestimate
the actual memory use.  But the most common problem is that unless
you use a PBS-aware MPI, PBS will not know anything about the MPI
tasks and in particular will not evaluate their memory usage.

You should note that job physical memory usage is not likely to be
repeatable (the same job may have a different resources_used.mem
on a subsequent run) because it depends on what else is running on
the system and how much the processes are paged.  vmem should be
more repeatable but may not be as meaningful or useful.

David

Roger Moye wrote:
> 
> All,
> 
> We are using Torque 2.3.0.  I have a question about information for 
> completed jobs such as resources_used.mem.   Is this showing me the 
> maximum amount of memory used by the job, an average, or something 
> else?  I have looked through the documentation and can not find this 
> defined anywhere.
> 
> The specific issue is that we have a user on our cluster who would like 
> to monitor the cpu time and maximum memory used by his jobs.  The best I 
> could come up with for memory is resources_used.mem but it seems that 
> this value is understated on a test job that I ran today.
> Mailing list threads on this topic are old so I thought I'd pose the 
> question.
> 
> Thanks in advance.
> -Roger
> 




More information about the torqueusers mailing list