Frequently Asked Questions
Q:
My code runs on a single processor.
How can I run multiple copies,each
with different input parameters, with a single batch submission?
Also, since the beowulf hosts each have two processors, if I allocatethe
whole system to one process, one CPU is just sitting idle. So howdo
I run two processes on each host?
A:
Ernie Valeo has created a script that allows
you do do this.
Q: Do you have pbs reporting tools/packages that provides more info than simple pbs-report?
A: You don't have any other pbs reporting tools/packages that provide more info.
Q: What is Muda?
A: Muda is the Japanese term for waste and comes from Lean Enterprise/Manufacturing. In this context we refer to 'waiting/idle muda'.
It's a ratio of how much of a jobs'time is spent suspended, as a ratio of
suspend time to CPU time. Our goal is to drive muda to zero.
Q: UserA runs large parallel jobs 8/16 node jobs all the time, and he got 0.128 Efcy. UserB runs single process jobs all the time, and he got 2.624 Efcy. Can you explain this?
A: The explanation about the differences for Efcy of userA and userB is that
MPI job spawn processes that are not being tracked by PBS. Therefore that
would explain the difference.
Q: Where is ghostview on portal?
A: Please try ggv, which stands for Gnome Ghostview.
Q: What is the maximum value that I can set for P4_GLOBMEMSIZE?
A: Currently, on all PPPL cluster nodes, the maximum value for P4_GLOBMEMSIZE is 1073741824.