[Gate-users] Validation and job splitting

[Mikell,Justin K.]

This problem may be related to the thread on "Sudden drop in recorded coincidences."

I see localized dips in the decay curve for O-15 and F-18 when trying to validate benchPET using the job splitter.

Here are some more results showing a dip in the first time bin of 12 seconds:
'gjs -n 20 -c openPBS -openPBSscript myscript benchPET.mac'
http://tinypic.com/r/2wqu26p/5

And here are results from a nearly identical simulation, the only difference being I've switched seeds on the 1st and 10th job. You can see that the dip in decays moved from the 1st to the 10th time bin.
http://tinypic.com/r/oa6hpe/5

Is there a fix for this or is identifying outlier jobs and rerunning them the only current work around?


Here are two seeds that were bad for me: (just in case someone wants to track this down)
1793480520928942512
931533650595167598

And here is my setup:
Gate 6.1.
Geant 4.9.4.p01
CLHEP 2.1.0.1
Root 5.28
gcc v4.1.2
Linux cnode339 2.6.18-194.el5 #1 SMP Fri Apr 2 14:58:14 EDT 2010 x86_64 x86_64 x86_64 GNU/Linux

-Justin
From: Mikell,Justin K.
Sent: Friday, March 23, 2012 4:03 PM
To: 'gate-users at lists.opengatecollaboration.org'
Subject: Validation and job splitting

Has anyone else had a problem with job splitting on the benchmark data?
I noticed some artifacts when I use splitting, but they disappear without.


Results of benchPET with gjs -n 10
http://tinypic.com/r/2wbydfn/5


Results of benchPET without splitting.
http://tinypic.com/r/1ekkno/5


For benchSPECT, results without job splitting match the available data well, but I see deviations again when I use gjs -n 16


Justin Mikell
Graduate Research Assistant
Radiation Physics
The University of Texas M. D. Anderson Cancer Center
ACBP1.2920
(713)-563-9128
jmikell at mdanderson.org