[Mauiusers] mprof problem and questions (maui-3.2.6p11)

[Franco M. Bladilo]

How does mprof calculate the PSrun value when profiling stats ?
We are seeing large differences between mprof results and  custom 
scripts that we wrote for accounting purposes.
Our scripts follow the formula indicated in the "workload trace format" 
webpage to calculate PSrun :
(completion time - start time) * num_tasks_allocated
Example, mprof -D 5 -t tracefile (extracted example job):

MUGetIndex([DEFAULT],ValList,0)
INFO:     job '9935' (1075:1075:1076) FQT: 1096384969  QT: 1096658466  
ST: 1096658468
INFO:     job '9935'(  4/   900) dist 1075 :   0.11 (           7.63)
INFO:     job '9935'(  4/   900) dist 1076 :   0.90 (           1.42)
MStatUpdateCompletedJobUsage(9935,2,0)
INFO:     job '              9935' completed.  QueueTime:      2  
RunTime:    900  Accuracy: 100.00  XFactor:  1.00
INFO:     start: 1096658468  complete: 1096659378  SystemQueueTime: 
1096658466
INFO:     overall statistics.  Accuracy:  0.08  XFactor:  2.57
INFO:     updating statistics for Grid[time: 3][proc: 2]
INFO:     job '9935' completed  X: 1.002222  T: 900  PS: 3600  A: 1.000000

Total PS=3600 is calculated using 900 PSrun when 1096659378 - 1096658468 
= 910 ??

 The trace record seems to be correct :
9935                   0   4 user1 compilers     900 Completed   
[short:1] 1096658466 1096658468 1096658468 1096659378    [NONE] [NONE] 
[NONE] >=    0M >=      0M     [NONE] 1096658466   4    0 [NONE]:DEFAULT 
[RESTARTABLE]    [NONE]          [NONE] [NONE]   0 3639.80   
DEFAULT      1      0M      0M      0M
      0 2140000000 n49:n48:n47:n46 base [NONE] [NONE] [DEFAULT] [NONE] 
[NONE]

910 x 4 = 3640 (matches ps_secs_run)
Are we missing something?

These little discrepancies make a great  impact when  processing 
thousands of jobs.


Thanks in advance,

-- 
Franco Bladilo
Linux/HPCC Administrator
Research Computing Group
Rice University
bladilo at rice.edu