Hi all
I'm looking for some basic help with a fresh open-mpi installation. Main question first and background to follow:
Is open-mpi as tolerant of heterogeneous low spec hardware as lam was?
I sysadmin a small cluster for a research company. The project for which this cluster is required drifts in and out of importance and as such, our last install (back in 2002) was lam-6.5.9 - which worked fine on an 8 node cluster running Slackware 7.1. Some were as old as Pentium 1-120s with 12 Mb of RAM ranging up to P3-666 and 128Mb of RAM. As I say, it all hung together well and did the job.
The programmers tell me they now need open-mpi as subsequent releases of lam refused to compile on this hardware - not sure why (lack of RAM I guess) but as a humble sysadmin, I don't want to get too far into it - as long as I did:
./configure
make all
make install
and setup appropriate permissions on each node - it worked!
So if we decide to use openmpi instead of lam, is it worth bothering with the existing hardware? To be fair, the Pentium 1's have now been dropped but all of the "new" hardware is around the P3 spec with 64-128Mb of RAM.
Any constructive help gratefully received - but before you ask - no - the company isn't about to buy any new high-end workstations to replace this old kit! They would rather stick with lam on old kit than buy new kit to allow openmpi to run.
Cheers
Rob
|
