For more tips on using ithaca, but not specifically Matlab, try this link in the wiki:

https://www.aoe.vt.edu/~stedwar1/Steve/doku/dokuwiki-2009-02-14/doku.php?id=aoe:hpc#batch_system

Setup instructions:

https://portal.arc.vt.edu/matlab

The validation is buggy. If a distributed job works, try a matlab pool anyway. My thought is the validation gets hung in queue because it is asking for too many nodes.

On machines without a DNS record,

pctconfig hostname 128.173.xxx.xxx

where xxx has the ipaddress

To check the queue

showq
showq -p ITHACA

To monitor the queue on ithaca:

watch 'qstat -u stedwar1'

If Matlab closes and there are still jobs in queue, try this

qdel xxxxx

where xxxxx is the job number from qstat

Matlab test commands

>> matlabpool (2)
>> parfor i=1:2,a=rand(4,4);eig(a),end
>> parfor i=1:9,a=rand(4,4);eig(a),end
>> parfor i=1:2,a=rand(4,4);eig(a);pctconfig,end
>> matlabpool close

To check job stats in Matlab (Eugene Cliff)

>> sched = findResource();
>> jobs = get(sched, 'jobs')

To delete stuck jobs: (Eugene Cliff)

>> destroy(jobs(i))  ; % I think you have to destroy these one-at-a-time, so i = 1, 2, etc

R2009b gets upset if R2010a has tried to run jobs. Try deleting all the files and Matlab crashdumps on both ithaca and the client host at the locations specified in the parallel configuration.

Another simple test:

parfor i = 1 : 1024
    d(i) = sin ( i * 2 * pi / 1024 )
end