On Thu, Apr 26, 2018 at 12:23 PM, Salman Zarrini
<[email protected]> wrote:
>
> Dear Christoph,
>
> On Friday, April 20, 2018 at 1:11:03 PM UTC-4, Christoph Junghans wrote:
>>
>> On Fri, Apr 20, 2018 at 9:33 AM, Salman Zarrini
>> <[email protected]> wrote:
>> > Thanks Christoph.
>> >
>> > On Friday, April 20, 2018 at 8:41:41 AM UTC-4, Christoph Junghans wrote:
>> >>
>> >> On Fri, Apr 20, 2018 at 3:35 AM, Salman Zarrini
>> >> <[email protected]> wrote:
>> >> > Dear Votca developers,
>> >> >
>> >> > My Votca calculations are conducted in a cluster which has 82
>> >> > machines
>> >> > each with 16 slots.
>> >> > While the Gromacs can parallel either inter-nodes or intranode, the
>> >> > Votca is only multithreaded so that only 16 slots can be used
>> >> > maximally.
>> >> > Consequently the Gromacs as the most time consuming part of a
>> >> > IBI-Votca job
>> >> > would be limited to 16 slots as well.
>> >> > So, I wonder if there is any possibility to use a higher number of
>> >> > slots
>> >> > for at least the Gromacs part of a IBI calculation in such situation?
>> >> Sure nothing prevents you from running VOTCA and gromacs on a
>> >> different number of cores, e.g. using 4 nodes (4*16 cores) for the
>> >> gromacs part (with mpi) and only 1 node (16 cores) for VOTCA's
>> >> csg_stat.
>> >
>> > Probably doable for the csg_stat by using --nt 16, but how about the
>> > csg_inverse in IBI;
>> > As far as I know, the allocation for running a IBI job (csg_inverse
>> > --options settings.xml) in cluster is addressed by two below commands in
>> > job
>> > submission script and setting.xml, respectively.
>> >
>> > #$ -pe shm n         ## Or other types of parallel environment.
>> >
>> > <mdrun>
>> >       <command> mpirun -np m gmx_mpi mdrun </command>          ### (m =<
>> > n)
>> > </mdrun>
>> >
>> > The n slots indicated in the job submission script would be assigned to
>> > the
>> > both VOTCA and Gromacs commands. Then how can I order to use 4 nodes
>> > (4*16
>> > cores) for the gromacs and 1 node (16 cores) for VOTCA (probably the
>> > other
>> > way  round was much easier in general)? And How VOTCA knows that those
>> > 16
>> > cores are all belong to a single node? because if only 1 core from the
>> > 16
>> > cores is contributed by another node the VOTCA part crashes.
>> The only compute-heavy part in csg_inverse besides mdrun is csg_stat,
>> all the other things take basically no time.
>> And csg_stat when called inside csg_inverse will automatically be
>> given the max. number of threads with the --nt options (check your
>> inverse.log, you can overwrite that with the
>> cg.inverse.simulation.tasks option in settings.xml).
>> On the other hand the behavior of mdrun can be changed by
>> cg.inverse.gromacs.mdrun.command, as you said above.
>>
>> So in your batch script you should set the number in "-pe" option to
>> same number as you want to use for gromacs (and set in
>> cg.inverse.gromacs.mdrun.command, too), but then in body of the
>> script, just call csg_inverse without mpirun/mpiexec. Then csg_inverse
>> will run serially, but call gromacs with mpi options and csg_stat will
>> automatically make use of all threads on the node where csg_inverse
>> got started.
>
>
> The cg.inverse.simulation.tasks helps to allocate different number of slot
> for the "csg_inverse" and "mpirun gmx_mpi mdrun", thanks.
slot for *csg_stat", csg_inverse is never parallel.

>
> Nevertheless, when I call Groamcs via the VOTCA (settings.xml) to be used in
> multiple nodes the speed drastically decreases even for a small system,
> whereas the Groamcs-only job free of VOTCA, works normally fine. After
> discussing the issue with our system administrator, we found out that the
> reason probably is because the Gromacs command is hardcoded into the
> settings.xml file so it cannot dynamically contain the hostlist that was
> decided by the scheduler at run time, whereas in Groamcs-only job the
> environment is readable directly by mpirun which is aware of Grid Engine
> environment variables.
>
> So, I was wondering if there are any other command that should be considered
> in the settings.xml or any solution in general to somehow figure out the
> issue?
You could dump the hostlist to a file in batch script before starting
csg_inverse.
And then use that hostfile by running gmx with "mpirun --hostfile .."
through the setting settings.xml.
E.g. in the slurm batching system, you can use $SLURM_HOSTFILE for that.

Christoph
>
> Sincerely,
>
> Salman Zarrini
>
>>
>> Christoph
>> >
>> > Thank you.
>> > Regards,
>> > Salman Zarrini
>> >>
>> >>
>> >> Christoph
>> >>
>> >> > BTW, Votca launches Gromacs directly with "mpirun” now.
>> >> >
>> >> > Regards,
>> >> >
>> >> > Salman Zarrini
>> >> > Department of Chemical Engineering
>> >> > Drexel University
>> >> > Philadelphia, USA
>> >> >
>> >> > --
>> >> > You received this message because you are subscribed to the Google
>> >> > Groups "votca" group.
>> >> > To unsubscribe from this group and stop receiving emails from it,
>> >> > send
>> >> > an email to [email protected].
>> >> > To post to this group, send email to [email protected].
>> >> > Visit this group at https://groups.google.com/group/votca.
>> >> > For more options, visit https://groups.google.com/d/optout.
>> >>
>> >>
>> >>
>> >> --
>> >> Christoph Junghans
>> >> Web: http://www.compphys.de
>> >
>> > --
>> > You received this message because you are subscribed to the Google
>> > Groups
>> > "votca" group.
>> > To unsubscribe from this group and stop receiving emails from it, send
>> > an
>> > email to [email protected].
>> > To post to this group, send email to [email protected].
>> > Visit this group at https://groups.google.com/group/votca.
>> > For more options, visit https://groups.google.com/d/optout.
>>
>>
>>
>> --
>> Christoph Junghans
>> Web: http://www.compphys.de
>
> --
> You received this message because you are subscribed to the Google Groups
> "votca" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> To post to this group, send email to [email protected].
> Visit this group at https://groups.google.com/group/votca.
> For more options, visit https://groups.google.com/d/optout.



-- 
Christoph Junghans
Web: http://www.compphys.de

-- 
You received this message because you are subscribed to the Google Groups 
"votca" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To post to this group, send email to [email protected].
Visit this group at https://groups.google.com/group/votca.
For more options, visit https://groups.google.com/d/optout.

Reply via email to