[Wien] parallel under sge environment

[Laurence Marks]

Several comments:

1) Your sysadmin is expecting you to run single mpi jobs, rather than
the Wien2k method of splitting up k-points to run on different
machines. So long as you are setting up a machines file correctly,
using mpi and are using 9.3 I think you can set MPI_REMOTE=0 and
USE_REMOTE=0 and avoid rsh/ssh completely. If you have no rsh/ssh you
are not going to be able to do anything except use mpi.

2) More serious. In the submit.sh script you attached there is the line
module load compilers/gcc/4.1.2 mpi/openmpi/1.3.4_gcc

Obviously you don't want gcc for the compilers. Very serious --
openmpi 1.3.4 has several bugs in it which will prevent running large
Wien2k jobs but are OK for small ones. Use version 1.4.X if available,
otherwise request that it be compiled.

N.B., I have a small utility somewhere (it is too early in the morning
at the moment in the US for me to find it) which will create a Wien2k
machines file for qsub/msub and a few other cases which I can send --
you can probably customize it for the particular format you have in
the PE_HOSTS file.

2010/4/20 zhaoyh <yhzhao.mail at gmail.com>:
> Dear Prof. Blaha,
>
> Thanks for your reply.
>
> "I cannot use ssh" means that this supercomputer doesn't allow users to
> log in to the compute node directly. I have consulted the admin already.
> He just ask me to use sge script to submit job. The attachment is the
> sge script used to submit a PWSCF job.
>
> I also have tried "ssh-keygen -t rsa", and copy the public key to
> authorized-key. However, it still prompts me to input the password of
> the compute node when using "ssh node1.........".
>
> In fact, I just want to use wien2k under the k-points paralleling, not
> the mpi-version, because the cell contains only about forty atoms.
>
> Any suggestion is grateful.
>
> Regards,
>
> yonghong
> On Tue, 2010-04-20 at 08:12 +0200, Peter Blaha wrote:
>> What does it mean: "I cannot use ssh".
>> You need to describe your procedure and the error message in much more
>> detail.
>>
>> -----------------------------------------------------------
>> It depends on your cases, if you want to use k-parallel or mpi-parallel
>> versions.
>> For small cases (less than 20-40 atoms) the mpi-version is completely
>> useless!!! It may even run faster on one node than on 16.
>>
>> You have to follow the previous advise:
>>
>> Ask your computing center if ssh is possible and how one can do it.
>>
>> You may need to generate ssh-keys  (ssh-keygen -t rsa)
>> and copy the public key into "authorized_keys".
>>
>> if ssh is really not possible (I doubt that!), install WIEN2k with the
>> option "shared memory" in siteconfig.
>> Then you can at least use all 4-8 cores of one node (if your hardware
>> has dual quadcore nodes) for k-parallel runs.
>>
>> Just to be sure: Most likely you cannot ssh from the fronend (where you
>> login) to the compute nodes, so you need to put the run_lapw -p into
>> your script.
>>
>> Maybe you need some additional switches in your job script to get a
>> proper environment.....
>>
>>
>> Am 19.04.2010 21:39, schrieb zhaoyh:
>> > Dear Blaha,
>> >
>> > I want to use wien2k in a supercomputer which uses sge batch submission
>> > system. I can use sge variable $PE_HOSTFILE to get the name of the
>> > computing nodes, and then write the correct .machines file. However, I
>> > cannot use ssh or rsh to log in the computing nodes then I cannot
>> > execute run_lapw which include "rsh/ssh node...." command.
>> >
>> > Would you please help me?
>> >
>> > Regards,
>> >
>> > Yonghong Zhao
>> >
>> >
>> > _______________________________________________
>> > Wien mailing list
>> > Wien at zeus.theochem.tuwien.ac.at
>> > http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>>
>
>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>
>



-- 
Laurence Marks
Department of Materials Science and Engineering
MSE Rm 2036 Cook Hall
2220 N Campus Drive
Northwestern University
Evanston, IL 60208, USA
Tel: (847) 491-3996 Fax: (847) 491-7820
email: L-marks at northwestern dot edu
Web: www.numis.northwestern.edu
Chair, Commission on Electron Crystallography of IUCR
www.numis.northwestern.edu/
Electron crystallography is the branch of science that uses electron
scattering and imaging to study the structure of matter.