[Neurodebian-users] FSL and Condor

[Michael Hanke]

On Sat, Mar 31, 2012 at 01:47:05PM +0200, Bertram Walter wrote:
> - I started time fsl-selftest feat without FSLPARALLEL and got
> ...
> real	3m46.880s
vs.
> real	5m13.042s

> Oooh, much slower than without Condor

That is not surprising. None of the FEEDS tests submits any
jobs that _can_ actually run in parallel. The 90s slow down comes from
scheduling and monitoring delays -- but they are a fixed overhead -- one
could do the math when parallelization would pay off ;-)

Anyway, if the full FEEDS test passes for you, chance are good that
parallelization works for you. You could try an actual FEAT analysis,
where you should see parallel jobs 9and speed up) in the preprocessing
stage.

> - I changed in /etc/condor/config.d/00debconf the lines
> CONDOR_HOST = 127.0.0.1
> ALLOW_WRITE = 127.0.0.1
> as described in Neurodebian-users Digest, Vol 19, Issue 10
> and restarted ubuntu for restarting condor completely.
> 
> - I started time fsl-selftest feat again.
> After 15 minutes I checked condor_q:
> -- Submitter:  : <127.0.0.1:45824> : bion05.uni-giessen.de
>  ID      OWNER            SUBMITTED     RUN_TIME ST PRI SIZE CMD
>   23.0   walter          3/30 17:05   0+00:00:00 I  0   1.0  bash
> 
> 11 jobs; 0 completed, 0 removed, 1 idle, 5 running, 5 held, 0 suspended
> 
> - Quite disappointed I killed fsl-selftest.

Don't be disappointed. It worked before on your machine, so why change
the config in the first place. There seem to be machines where the
default configuration doesn't do the job -- it does for you -- don't
modify it.

If you could repeat the last step again, and post the output of

  condor_q -analyze CLUSTERID

where CLUSTERID is the ID of that ideling job (23.0 in your message).
That would tell us why it is not running. The output if

  sudo ifconfig -a

might be necessary as well.

Michael

-- 
Michael Hanke
http://mih.voxindeserto.de