On Tue, May 2, 2017 at 9:48 PM, Roland Haas <[email protected]> wrote:

> Hello Robert,
>
> > My problem is similar to the one discussed in a thread here in April,
> > though I have not solved it. My system is based on the AMD FX-8350
> > CPU, which should have either 4 or eight cores depending on whether
> > they are physical or virtual. The OS is Ubuntu 17.04 Zesty Zapus
> > (x64), released a couple of weeks ago. I followed the Simplified
> > Tutorial, with everything pasted in without alteration. (The
> > exception is that 'mpich' is apparently required instead of
> > 'mpich2.')
> Good to know. I'll update the tutorial, just have to understand how
> to make it work with the LTS and current releases and ideally both 17
> and 16.
>
> > The graphical output is exactly the same as that presented
> > on the Website, as near as I can tell. When computation starts, it
> > says it's using one thread, instead of the same number as the number
> > of processors, as I was led to believe. I then set the environment
> > variable: export OMP_NUM_THREADS=4 and verified it with printenv, but
> > it's still using only one thread. (It took about five hours to
> Hmm, so given that printenv worked and showed the OMP_NUM_THREADS is
> the expected value it should also show up in Carpet's output (and you
> have eliminated ~75% of the usual causes). Can you attach the *.out
> file, please?
>
> Given that you said you followed the simplified tutorial which uses
> simfactory, OMP_NUM_THREADS is likely not respected since the
> debian.run file overwrites it. Instead when using simfactory to
> start the simulation did you specify a --num-threads option? Like so:
>
> simfactory/bin/sim submit static_tov --parfile par/static_tov.par
> --procs 1 --num-threads 4 ?
>
> This may actually fail stating that there are not enough cores on you
> laptop since simfactory (I think) defaults to assuming only a single
> core. You may have to add "--mdbentry max-num-threads 4 --mdb-entry ppn
> 4" as well. This should likely be changed in simfactory.
>
> Please do not use --procs 4 --num-threads 4 or similar since the number
> of cores used with be --procs (number of processes) times --num-threads
> (numer of threads per process) though that would give you a different
> output in Carpet.
>

I'm not sure this is correct. I think that "procs" specifies the total
number of cores that Cactus uses, and "num-threads" specifies the number of
OpenMP threads per MPI process. The number of MPI processes is then the
quotient of "procs" and "num-threads".

Apart from this I agree with Roland that the *.out file will likely contain
the information needed to debug this.

-erik

-- 
Erik Schnetter <[email protected]>
http://www.perimeterinstitute.ca/personal/eschnetter/
_______________________________________________
Users mailing list
[email protected]
http://cactuscode.org/mailman/listinfo/users

Reply via email to