[Beowulf] Should I go for diskless or not?

Gus Correa gus at ldeo.columbia.edu
Tue May 26 09:40:08 PDT 2009

Hi Dr. Cool Santa

Dr Cool Santa wrote:
> @Gus: According to the documentation, Schrodinger software should only 
> run good on MPICH1, otherwise I'd have preferred MPICH2.

First off, I know nothing about computational chemistry.
Hence, consider my suggestions at your own risk.

Did you check with Schroedinger.com if they really require MPICH1,
or would have executables compiled with an MPI
that is alive and kicking, unlike Schroedinger's cat and MPICH1?
I presume they distribute the executables, not the source code, right?
Otherwise you could link to the MPI of your choice.

Incidentally, way back when I was a grad student,
Schroedinger (com, not Erwin) was a startup sharing an
IBM SP2 with us.
I.e., sharing when the sys admin didn't kill
my modest 1-processor jobs. :)

There have been reports of MPICH1 failing even with the simplest
"cpi" programs on current Linux kernels:

The fix?  Use another MPI, say MPICH2.

The same problem was reported with another computational
chemistry program (PCGAMESS), with the same proposed solution
(don't use MPICH1):

Apparently, it is all about the old P4 library beneath MPICH1,
which was phased out in MPICH2.

The MPICH developers don't support MPICH1 anymore,
and recommend that users migrate to MPICH2 w/ nemesis channel (Ethernet):

OpenMPI is yet another alternative (Ethernet, Infiniband, Myrinet)

And MVAPICH2 if you have Infiniband:

Gus Correa
Gustavo Correa
Lamont-Doherty Earth Observatory - Columbia University
Palisades, NY, 10964-8000 - USA

> @Tom: Yeah I know I could do that, I could assign the IP address, that 
> worked, but the hostname just didn't work, so I didn't bother. I did 
> check the option for it to configure hostname via DHCP, and I also 
> configured my server to give out hostnames, it just didn't work. I made 
> the startup scripts assign the hostname according to the IP address, and 
> it seems the server is giving out the IP addresses according to the MAC 
> address already. Its not going on in the order.
> @John: I knew that, it is not the issue I am having problems with.
> ------------------------------------------------------------------------
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

More information about the Beowulf mailing list