[Wien] -machinefile _HOSTS_

[Peter Blaha]

If you do not mix k-parallel and mpi-parallel runs, it is probably save. (i.e. for VERY big cases)

The question is: why are you using   mpiexec_mpt  ??

Usually it is better to use the more general mpirun (or the less general mpiexec) instead
of the "special mpiexec_mpt comand.

Besides that, for some mpi-versions/commands it might be   -host or -hostfile instead of -machinefile

checkout, which   mpiXXX commands do you have, and with man/help pages, what are the available options.

To my understanding  PBS_NODEFILE  is a variable of some queuing-systems (like PBS, but maybe some others
too. It is not really "mpi-specific".

PS: In "mixed" parallelizations you want to request eg. 256 cores, but you may have 4 k-points, thus doing
4 k-parallel lapw1-jobs on 64 cores each. In such a situation you need to tell mpi, on which nodes it should run
a particular process. And I'm pretty sure, that EVERY (meaningful) installation can do this.

Am 08.04.2013 19:43, schrieb Luis Ogando:
> Dear all,
>
>     Dr. Gavin Abo pointed me out that I should have mentioned that I am using mpiexec_mpt instead of mpiexec, so, in my case, the mpi execution is controlled by the
> PBS_NODEFILE variable and not by a machines file (thanks Dr. Abo for this).
>     Anyway, I would like to know if it is safe to run MPI Wien2k with mpiexec_mpt.
>     All the best,
>                     Luis
>
> PS: I have no problem on generating the .machines file "on the fly" (queuing system). Despite the fact that it will not be used by mpiexec_mpt, I know that it is required
> by Wien2k.
>
>
>
>
> ---------- Forwarded message ----------
> From: *Gavin Abo* <gsabo at crimson.ua.edu <mailto:gsabo at crimson.ua.edu>>
> Date: 2013/4/8
> Subject: Re: [Wien] -machinefile _HOSTS_
> To: Luis Ogando <lcodacal at gmail.com <mailto:lcodacal at gmail.com>>
>
>
> Dear Luis,
>
> You probably should have mentioned that you are using 'mpiexec_mpt' not 'mpiexec'.
>
> -machinefile is an option for mpiexec [http://linux.die.net/man/1/mpiexec], but it doesn't seem to be a option for mpiexec_mpt
> [http://techpubs.sgi.com/library/tpl/cgi-bin/getdoc.cgi?coll=linux&db=man&fname=/usr/share/catman/man1/mpiexec_mpt.1.html].
>
> mpiexec_mpt seems to use the PBS_NODEFILE variable instead of -machinefile [http://www.arl.hpc.mil/docs/pbsUserGuide.html].  So your parallel_options are probably fine as
> long as the PBS_NODEFILE variable is set automatically by your system or you.
>
> Kind Regards,
>
> Gavin
>
>
> On 4/8/2013 10:18 AM, Luis Ogando wrote:
>> Hi Gavin,
>>
>>    Thank you for your answer.
>>    Actually, I generated the .machines files for the queuing system without problems. The issue is that the " setenv WIEN_MPIRUN " in the "parallel_options" file has to
>> be " setenv WIEN_MPIRUN "mpiexec_mpt -np _NP_  _EXEC_" ", without the "-machinefile" option because it is not defined in the system.
>>    All the best,
>>            Luis
>>
>>
>>
>>
>> 2013/4/8 Gavin Abo <gsabo at crimson.ua.edu <mailto:gsabo at crimson.ua.edu>>
>>
>>     Dear Luis,
>>
>>     I'm sending this email off the mailing list as Prof. Marks or Blaha might respond with a better answer.
>>
>>     I think the answer is yes, it can impact the Wien2k performance. This is because the machinefile variable contains the list of hostnames for 'multiple' nodes.
>>      Without it, the calculation will likely run on only 'one' node.
>>
>>     Even if you don't have admin privileges, you can likely still define the machinefile variable as a user by creating a .machines file in your case directory.  However,
>>     the creation of the .machines file may depend on whether or not you are required to use a queuing system.  If you are not required to use a queuing system, you should
>>     be able to copy the .machines file in SRC_templates to your case directory and then edit it a text editor (note: you might not see the .machines file unless you do a
>>     directory listing that includes hidden files).  The .machines file is should be described in the Wien2k userguide or you can search the internet for some examples
>>     [https://www.xsede.org/documents/10157/305826/ecss_hliu_051012.pdf]. If you are required to use a queuing system, you likely need to setup a script that will create
>>     the .machines file as described at the link:
>>
>>     http://www.wien2k.at/reg_user/faq/pbs.html
>>
>>     If you have problems creating the script, your administrator(s) or support person(s) should know the most about your computer system, so they can likely help you
>>     create a script that will work on your system.
>>
>>     Kind Regards,
>>
>>     Gavin
>>
>>
>>     On 4/8/2013 6:19 AM, Luis Ogando wrote:
>>
>>         Dear Prof. Marks, Blaha and Wien2k community,
>>
>>            I want to do calculations in a computer where the machinefile variable is not defined for the mpiexec (I am not the administrator). I would like to know if
>>         this will have some impact on the MPI Wien2k performance.
>>            Thanks in advance,
>>                                         Luis
>>
>>
>>
>
>
>
>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> SEARCH the MAILING-LIST at:  http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
>

-- 
-----------------------------------------
Peter Blaha
Inst. Materials Chemistry, TU Vienna
Getreidemarkt 9, A-1060 Vienna, Austria
Tel: +43-1-5880115671
Fax: +43-1-5880115698
email: pblaha at theochem.tuwien.ac.at
-----------------------------------------