AW: NAMD slows at startup phase 1 smp problem

From: Norman Geist (norman.geist_at_uni-greifswald.de)
Date: Tue Jan 13 2015 - 05:50:39 CST

Guess I've understood the problem wrong. You seem to apply for more cores
than physically exist on your machines. Can we see the output of

 

/cat/proc/cpuinfo | grep -E "processor|core id|physical id"

 

and your call to namd?

 

Norman Geist.

 

Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im Auftrag
von Norman Geist
Gesendet: Dienstag, 13. Januar 2015 08:22
An: namd-l_at_ks.uiuc.edu; 'Ryan Gordon'
Betreff: AW: namd-l: NAMD slows at startup phase 1 smp problem

 

This indicate that you may have Hyper-Threading enabled. As it is not
supposed to bring any improvement for namd (and IMHO actually nowhere) you
can disable it in your BIOS. Anyway one would have to control which cores
NAMD is using to get reliably the best performance using f.i. taskset (not
sharing physical cores). As this is a unnecessary effort, the better and
easier option is to disable HT.

 

If you want to keep it enabled, you should use the commands offered in your
message to prevent the threads from sleeping while being idle as this
usually comes with a performance decrease aswell.

 

Norman Geist.

 

Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im Auftrag
von Ryan Gordon
Gesendet: Montag, 12. Januar 2015 19:25
An: namd-l_at_ks.uiuc.edu
Betreff: namd-l: NAMD slows at startup phase 1 smp problem

 

I am having some trouble running NAMD 2.10 for Linux-x86_64-MPI-smp-CUDA. I
am running on 128 processors, 128 nodes, and 8 physical nodes. The warning I
am getting at the beginning is as follows:

Charm++> Warning: the number of SMP threads (32) is greater than the number
of physical cores (16), so threads will sleep while idling. Use
+CmiSpinOnIdle or +CmiSleepOnIdle to control this directly.

I am not sure how to address this issue, and it seems to take a long time
for "startup phase 1" to run compared to the other startup phases. Has
anyone else had similar problems?

--
Ryan Gordon
Ph.D. Candidate
Chemical and Biological Engineering
Drexel University
CBEGSA Vice President

This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:20:48 CST