Subject: Re: [OMPI docs] openmpi incquiry
From: Jeff Squyres (jsquyres_at_[hidden])
Date: 2012-02-01 14:49:13
I think you're using the wrong email list -- you probably want to send email to the Open MPI user's list, not the documentation list.
You should also send more information about your cluster (e.g., processors, OS, RAM, interconnect, etc.).
You should be able to do some back-of-the-envelope kinds of computations to see what kind of performance you *should* be getting on your cluster (given processor speed, RAM and problem size, etc.), and then compare that to what you *are* getting. If they're wildly different, you probably need to tune.
On Jan 27, 2012, at 8:45 PM, rob allen wrote:
> Hi
>
> i have built a High Performance Computing Cluster, which uses MPI and Octave. i used the PelicanHPC for the setup of my cluster. so am now trying to fine tune my cluster for good/ efficient performance.
>
> i started by doing benchmarks for overall system performance which involved using the standard HPL Linpack.
> now i want to do subsystem performance benchmarking which could involve fine tuning the mpi.
>
> my cluster is using the open mpi package. However am a bit confused and stuck on how to get started to tackle this objective of open-mpi tuning.
>
> may u pliz advise and direct me.
>
> Regards _______________________________________________
> docs mailing list
> docs_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/docs
-- Jeff Squyres jsquyres_at_[hidden] For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/