Re: help needed on saga adapter change

7 views
Skip to first unread message

Andre Merzky

unread,
Feb 7, 2017, 3:54:06 PM2/7/17
to Andrey Khmelevskiy, Anna Kostikova, saga-...@googlegroups.com
Hey Andrey, Anna,

[with Cc to the SAGA list, hope that's ok]


On Tue, Feb 7, 2017 at 2:07 PM, Andrey Khmelevskiy <kmele...@gmail.com> wrote:
> Hi Andre,
> Long time not talk to you. Hope you are well and sound!

Thanks, doing well :-D I hope the same for you!


> I have a question for you. I wonder does python SAGA
> (https://github.com/radical-cybertools/saga-python) support MPI and if not
> can it be done?

It does! You should be able to submit MPI jobs in SAGA, by specifying
the required number of CPU cores and setting the 'SPMD_VARIATION'
attribute in the job description to 'MPI', like this:

jd = saga.job.Description()
jd.wall_time_limit = 60
jd.executable = 'gromacs'
jd.arguments = [...]
jd.queue = "development"
jd.project = "TG-MCB090174"
jd.total_cpu_count = "16"
jd.spmd_variation = "MPI"

If that for some reason does not work as expected we would consider
that an error - please do open a ticket in that case!


> I have noticed that BigJob project
> (https://github.com/saga-project/BigJob) has MPI capability, but we use
> python version and it will take a lot of work for us to switch to BigJob.

Yes, indeed, as does the successor 'radical.pilot'
(https://github.com/radical-cybertools/radical.pilot/) - but I should
mention that neither relies on SAGA's ability to submit MPI jobs.

Please keep us posted if the above works for you!

Best regards, Andre.


> Your thoughts and suggestions would be very helpful for us.
>
> Thank you in advance and looking forward to hearing from you.
>
> Kind regards,
> Andrey



--
99 little bugs in the code.
99 little bugs in the code.
Take one down, patch it around.

127 little bugs in the code...
Reply all
Reply to author
Forward
0 new messages