0

I have a shared memory server with PBSpro installed as the job manager. There are 320 cores total, and pbs is configured so there is 1 job queue having 30 of the 32 cpu's so 300 physical cores to be divided up among users, with 2 cpu's or 20 cores left for head node operation.

some software will run on or parallelize over 100+ cores, and i'd like to verify N processes from user A are on unique and separate core numbers than the M processes from user B.

If I use top I can do

hit F to get into sort menu
hit J to sort by P = Last CPU used (SMP)

this gives me a P column in the top output having core number, but on a 300 core system I can only get around 70-80 rows before I run out of screen, on a 1920x1200 monitor. I can start shrinking font size in the terminal window but in the end I won't be able to see up to 300 rows, not that I can even visually process all that with top updating every 1..3 seconds.

My goal is to quickly and easily

  • verify users are running stuff in the job queue within the correct core numbers, and not on the head node
  • verify for a given cpu core that is at 100%, or anything over 50%, that only one process from one user is running on it. I want to make sure that if user A with programA.x is on core #234, user B with anything.x is NOT on core #234.

what's the best way to do this, when a single image shared memory server has many cores?

1 Answer 1

0

You could use ps with the custom option and sort flags. I'm not 100% sure what sort of filtering you want from your question.

So, perhaps you mean show me all processes. I want to see the CPU they are on, the username and process name. Sort this by CPU

ps ax -o c,user,comm k c

The "ax" means all processes, "-o c,user,comm" is what columns to display and "k c" means sort by CPU.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .