Performance of MTL login node

Performance of MTL login node

Portrait de dotcsw
dotcsw
dim, 08/05/2011 - 18:52

Newcomer to MTL here. My program runs fine on the Manycore Testing
Lab as long as it's singlethreaded. Whenever I enable multithreading the
performance gets worse on the same input. Starting more threads only
makes it slower.

At first I thought it was just poor design on my
part - too much contention for mutexes or something. However, now I'm
testing a program with NO synchronization between threads and
it's still happening. The performance is what one might expect from a single
core processor.

This is on the login node (acana01). Does it have
access to the 40 cores? Do I have to do anything special to enable all
of them?

- Rick

RSS Haut
4 posts / 0 nouveau(x)
Dernière contribution
Reportez-vous à notre Notice d'optimisation pour plus d'informations sur les choix et l'optimisation des performances dans les produits logiciels Intel.
Portrait de Mike Pearce (Intel)
Mike Pearce (Intel)
lun, 09/05/2011 - 09:47

Rick,
It's a little difficult diagnosingyour complete lack of multi-thread performance, without some more data points, like the programming language and libraries you are using and how your data is handled/partitioned to each thread. Since the MTL started up over a year ago, no one has reported a complete lack of performance improvement moving from serial to parallel when using threads. Can you post a simple test code that exhibits these problems?

In addition you asked about access to the 40-core system - the 40-core system is ONLY available as a batch job. You have to specify the number of CPUs in the qsub command to be equal to 40, to execute your job on the 40-core system. The login node does not contain 40 physical cores, but does reflect 64 logical cores, with the use of hyperthreading enabled. Ether way there is nothing special you need to do to specifically enable them.

Haut
Portrait de dotcsw
dotcsw
lun, 09/05/2011 - 13:51

Never mind. It must be my program that's doing something wrong. I am indeed an "apprentice" when it comes to multithreaded programming.

Thanks for the info about the 64 logical cores on the login node. I ran VTune Amplifier XE (from the command line) but only managed to produce a data file. Do I need to setup VNC in order to actually view the results?

- Rick

Haut
Portrait de Mike Pearce (Intel)
Mike Pearce (Intel)
lun, 09/05/2011 - 14:47

There are a couple of options to view your results; the first being using VNC or Xming to view the data file on the MTL, the second is toinstall a trial version of VTune Amplifier XE on your local machine and then download the data file from the MTL and view it locally.

Haut

Connectez-vous pour laisser un commentaire.