[petsc-dev] ugly shit

Satish Balay balay at mcs.anl.gov
Fri Nov 11 06:59:41 CST 2011


On Fri, 11 Nov 2011, Kai Germaschewski wrote:

> One was NASA's CCMC (community coordinated modeling center) running
> mvapich, the other, IIRC, was a local cluster with mpich-gm (Myrinet).

I guess sysadmins on some of the older machines that are chugging
along are reluctant to upgrade. [If its wokred fine so far why bother
changing? And change usually means tons of user e-mail - extra effor
on a machine that can potentially be replaced soon..]

However I believe - in these cases - the user can install a newer MPI
with same functionality [mvapich or mpich2-gm] - and be able to use
such clusters. Should we rely on this approach?

Its not clear to me if there are older non-cluster machines still in
use - with vendor MPI thats not easy for the user to circumvent with
his/her own newer MPI install..

Satish



More information about the petsc-dev mailing list