The homework requires students to parallelize a straightforward (but
computationally demanding) problem and prove that their implementation
is scalable and efficient using averages and 95% CI from at least 5
different runs using 3 different data sets. The sequential runtime is
about 3 hours. So students are throwing more compute power to get
scalability data faster. Most students seemed to have figured out the
use of qsub and parameters to mpiexec. Hopefully "dir" will also figure
it out after this homework.

On Wed, 2013-04-24 at 22:40 -0400, Robin Robin wrote:
> Interestingly, the student issues mpiexec -n 40 on a 8-core login
> nodes, thinking that he 
> could be done for the homework earlier.
> 
> 
> 
> I asked dir to do qsub -IV -l nodes=1:ppn=8 that will get him entire
> node for him to work on. So, he got several compute nodes to work on.
> 
> 
> Within the compute node, he still does -n X where X > 8. 
> 
> 
> I suspect dir does not know that X is typically number of nodes X PPN
> requested.
> 
> 
> I wonder if your homework is asking them to compare results with
> different number of CPUs involved. If that's the case, dir needs to
> know the batch system.
> 
> 
> Say, if you want to run -n 48, then a way to request that is
> nodes=6:ppn=8. I wonder if he knows about it.
> 
> 
> Robin
> 
> 
> 
> 
> On Wed, Apr 24, 2013 at 10:05 PM, Dhananjai Rao <[log in to unmask]>
> wrote:
>         Yes, I think this user is a student in my course. I will talk
>         to the
>         student tomorrow. There is a HPC homework due tomorrow. So,
>         there is
>         going to be some traffic tonight.
>         
>         On Wed, 2013-04-24 at 22:03 -0400, Mueller, Jens Dr. wrote:
>         > DJ,
>         >
>         >
>         > The user with the many processes is 'dir'. Is the user a
>         student in
>         > your class by any chance? The user is launching 'mpiexec -n
>         40' jobs
>         > on the head node without allocating proper resources.
>         >
>         >
>         > Thanks,
>         >
>         > Jens
>         >
>         >
>         > On Wed, Apr 24, 2013 at 4:55 PM, Dhananjai Rao
>         <[log in to unmask]>
>         > wrote:
>         >         Hello,
>         >
>         >         There is a user running 100s of processes on the
>         head node and
>         >         other
>         >         users are running heavy weight jobs on the head node
>         that is
>         >         making the
>         >         head node very sluggish.
>         >
>         >         thanks
>         >
>         >         with regards
>         >         D. M. Rao
>         >
>         >
>         
> 
>