Message: Re: MT in hpcc problem Not Logged In (login)
 Next-in-Thread Next-in-Thread
 Next-in-Forum Next-in-Forum

Question Re: MT in hpcc problem 

Forum: Multithreading
Re: None MT migration problem (Alex Kav)
Date: 27 Aug, 2015
From: Alex Kav <Alex Kav>

Dear Geant4 experts!
First of all great thanks for your comments on my previously questions. Its help and now I run my simulation of gamma ray spectra under 14 MeV neutron in multi thread mode on conventional laptop (I used Panasonic CF53, intel core I5-3320M CPU@2. 60GHz x 4). I have speed up of simulation approximately in 4 times by comparison with sequential mode.
I’ve try to run my MT code on High Performance Compute Cluster (http://eng.auburn.edu/admin/ens/hpcc/hpcc-description.html). Next problems appear at carrying out of simulation:
 In PBS script 10 cores was ordered. In code “G4int nThreads = 10; “. Simulation was gone and I have approximately 10 times speed-up. But:
-	simulation was gone on one core only ;
-	cpu load is too high, ~1000% for 10 thread; 
-	the jobs of other users are blocking (another user of hpcc complain that their job gone, for instance, 13 hours instead 10 min when my simulation run).
And when in PBS script the 10 cores was ordered, but in code “G4int nThreads = G4Threading::G4GetNumberOfCores(); “ the job was running in 128 thread (but also on one core) what probably correlate with number of cores in one node (see hpcc-description)”.
Could you say what is the reason such behavior of hpcc (all thread goes in one core) and what I need to do in code or PBS-script to fix this problems (main of them – affecting on the job time other users). Note, when simulation is running in sequential mode no effect on other jobs.

Thanks in advance, Alex Kav

Inline Depth:
 1 1
 All All
Outline Depth:
 1 1
 2 2
 All All
Add message: (add)

1 None: Re: MT in hpcc problem   (Alex Kav - 15 Sep, 2015)
(_ Warning: Re: MT in hpcc problem   (Andrea Dotti - 15 Sep, 2015)
 Add Message Add Message
to: "Re: MT in hpcc problem"

 Subscribe Subscribe

This site runs SLAC HyperNews version 1.11-slac-98, derived from the original HyperNews