[Beowulf] NASTRAN on cluster

[Lombard, David N]

>From Currit, Dennis on Monday, April 11, 2005 10:01 AM
> 
> We just installed a small cluster and are running NASTRAN 2005 on
it...

The answer with Nastran is always "it depends."

Nastran doesn't really want to run more than one job (MPI rank) per
node.

Nastran is generally B/W oriented.  Whether memory or disk is more
significant depends on the job; latency isn't generally an issue.

Network bandwidth and latency usually aren't an issue on DMP jobs, at
least not on Linux.

The distro can/will have a significant impact on allocatable memory.
Nastran uses brk(2) to allocate memory, so the TASK_UNMAPPED_BASE is
significant.

I can't comment on SATA, but PATA disks are a really bad choice, as they
require too much effort from the CPU to drive them--SCSI is MUCH
preferred in that case.

If disk I/O is an issue, Linux can do a good job caching in memory if
the nodes have sufficient physical memory, unless, of course, you're
disk files are >> physical memory size--in this case, it's all about the
RAID-O config.  Maximizing disk B/W is all about # of PCI channels, # of
controllers, # of drives, don't even think about a RAID controller.

As for CPU v. I/O.  The factors are (in no order):

fp performance
memory b/w
disk b/w
memory size

Which of the above dominates the analysis depends on the analysis.

--
David N. Lombard

My comments represent my opinions, not those of Intel Corporation.
________________________________________