case.run

[rsheppar@iu_edu]

Hi,  I have a bit of a bug and I am not sure where it is set.  It is not really a big deal but when the case directory is built, The PBS file that is created to submit has the line:#PBS -l nodes=64
Not knowing the total required memory and being cautious (each node is 16 cpu's and 32 GB of memory) I have been having them just replsace the line with:#PBS -l nodes=8:ppn=8But, it would be really great to find the script where this comes from so it (or possibly "nodes=4:ppn=16") could be inserted auto-magically.  Any help would be apprciated.Thanks!Ray P.S.  There are many, many posts in this forum.  looked to try to find if this has been previously discussed.  Is there some sort of tool that searches al the forum posts for key words or somethiong similar?Ray again  

 

[jedwards]

Hi Ray,There is a search tool at the top of the forum web page.   In most cases 2GB of memory per cpu is plenty and using 16ppn should work fine.   The number of tasks per node is set in file config_machines.xml (variable MAX_TASKS_PER_NODE) and the computation is done in a perl script called taskmaker.pl .   You can modify the behavior in the mkbatch.{machine} for the machine you are running on.

 

[rsheppar@iu_edu]

Thanks!  I am not sure how my eyes missed the yellow tab.  In the config_machines.xml file, the MACH="their_machine"  stanza already has MAX_TASKS_PER_NODE set as 16.  It isn't translating into ppn=16.  I have never touched taskmaster.pl, but I guess I should try tracing it out to see why it never got picked up.  Thanks again.           Ray