[torqueusers] guassian cput vs walltime problem.

Brock Palen brockp at umich.edu
Fri Jul 6 10:15:37 MDT 2007


We have a 8 core sun x4600 managed by torque.
We have a user running gaussian on it (a threaded chemistry code)

There is problems between the resorces_used.cput and walltime  For  
example:

[root at nyx dta_ipas-pancreas]# qstat -f 504714 | grep res
     resources_used.cput = 303:45:44
     resources_used.mem = 62955004kb
     resources_used.vmem = 63221752kb
     resources_used.walltime = 21:58:56

Its a 8 core machine but the cput  vs the wall time  is 13.8x  the  
wall time.  How can that be?  It should never be more than 8x's  the  
walltime.

  Any insight into what could cause this would help.  We use the cpu  
times reported by pbs to do some basic performance monitoring (bad  
pbs scripts swapping etc).  And these scripts make the assumption  
that cputime  can not be more than ncpu*walltime.

Thanks!


Brock Palen
Center for Advanced Computing
brockp at umich.edu
(734)936-1985




More information about the torqueusers mailing list