Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

Subject: Re: [OMPI users] openMPI installation on front-end and worker nodes
From: Salvatore Podda (salvatore.podda_at_[hidden])
Date: 2012-02-24 11:44:16


Apologize for the delay, but I missed your post.

> Hi,
>
> Am 22.02.2012 um 14:21 schrieb Salvatore Podda:
>
>> we have a computational infrastructure composed from front end and worker nodes
>> which differ slightly differ in the architectures on board (I mean same processor different socket).
>
> can you clarify this - something like an AM3 CPU in an AM3+ socket?

The processors on front-end and backend nodes are the same (from /proc/cpuinfo):

Intel Xeon CPU E5345

The two platforms are definitely different: front-end IBM x3650 and backend IBM bladecenter HS21,
but probably I am wrong and the socket are equal (LGA 771).

>
>
>> OFED and openMPI have been compiled and built on both systems at the same version level.
>> As users have access to the system only through front-end nodes (they compile on front-end
>> and submit job to worker nodes) the eventual difference in the OFED/openMPI installation
>> depending from the architecture can jeopardize the operation and the performance of the
>> user application?
>>
>> More generally, in the case of a front-end nodes with a processors definitively different from
>> worker nodes (same firm i.e.Intel) can openMPI applications compiled on one run correctly
>> on the others?
>
> This depends on the used compiler options. If you compile just for a CPU type in the machine you compile on (and usually this is taken as the target architecture by the compiler), it might fail on older CPUs which are in the nodes. Like compiling on a machine with Shanghai-64 CPU might fail on older Opterons, unless you use compiler switches to target the least common instruction.

Maybe I have not been enough clear. The first part of the question concern the compillation of openMPI in different
compilers flavours (gcc, intel pgi). The "config" phase will guess (if not imposed) flags/options from the architecture
on which is running. At this point having obtained a "succesfull" openmpi-compiler version on two different architectures
the user, in our case, will compile his own application on the front-end, and will run on the backend nodes with a "different"
version of openmpi. I am not speaking of great differences in the "age" of processors but of architectures definitely different.
However the conclusions of your comment are completely coherent with the advices of other openMPI users.

Thanks for your support,

Salvatore

>
> -- Reuti
>
>
>> Many thanks
>>
>> Salvatore
>> _______________________________________________
>> users mailing list
>> users_at_[hidden]
>> http://www.open-mpi.org/mailman/listinfo.cgi/users
>