FAQ  •  Login

ThreadCount PPN Relationship

Moderators: Dan Negrut, ME964 Spring 2012

<<

S12jrregner

Newbie
Newbie

Posts: 29

Joined: Mon Jan 23, 2012 9:07 pm

Unread post Wed Apr 11, 2012 10:17 pm

ThreadCount PPN Relationship

When designing and subsequently running my program I feel that I would be better served by understanding the relationship between some of the MPI built-in variables and the CLI programs.

MPI_Comm_size - This is determined when you call mpiexec -n ThreadCount
MPI_Comm_rank - This is the current "thread"

qsub -I -l nodes=1:ppn=4 - Requests 1 compute node and will use 4 of the compute nodes processors.

Question - Is there any relation between Nodes, PPN, and ThreadCount? Is it necessary that ThreadCount=Nodes*PPN?
Is there a limit on ThreadCount? (my program seems to hang when launched with mpiexec -n 1000000 Problem1)

Thanks,
John
<<

Dan Negrut

Global Moderator
Global Moderator

Posts: 833

Joined: Wed Sep 03, 2008 12:24 pm

Unread post Thu Apr 12, 2012 5:58 pm

Re: ThreadCount PPN Relationship

John - there is no relation between them.
First, i'd call them "processes", not ThreadCount.
You can have one node, four processors per node, and launch your program with eight processes. Then, internally, you'll most likely have each processor being over-committed as it has to run two processes.
And yes, there is a limit on how many processes you can launch and that is capped by the administrator and by the number of processors that you have at your disposal.
Post follow up questions if any.
Dan
<<

S12hatt

Newbie
Newbie

Posts: 47

Joined: Mon Jan 23, 2012 9:07 pm

Unread post Wed Apr 18, 2012 3:02 pm

Re: ThreadCount PPN Relationship

How many processes can run on a node? How would you prefer use to configure qsub so that the system doesn't get bogged down the night that the HW is due?
<<

Andrew Seidl

Administrator
Administrator

Posts: 193

Joined: Thu Oct 28, 2010 11:54 am

Unread post Wed Apr 18, 2012 3:52 pm

Re: ThreadCount PPN Relationship

Each of the GPU nodes (:intel) have 8 physical cores = 16 logical cores with HyperThreading (our code in the lab typically benefits from HT; yours might not). Each of the new AMD nodes (:amd) has 64 cores (not exactly, but close enough).

The best thing to do for the homework would be to only submit a batch job, not interactive. That way your job will eventually run, but it might not happen immediately.

Return to ME964 Spring 2012: High Performance Computing

Who is online

Users browsing this forum: No registered users and 1 guest

cron
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group.
Designed by ST Software.