[Neurodebian-users] fsl_sub error

[Lewis, Dave]

> -----Original Message-----
> From: Michael Hanke [mailto:michael.hanke at gmail.com] On Behalf Of
> Michael Hanke
> Sent: Sunday, March 11, 2012 3:17 PM
> To: Lewis, Dave
> Cc: neurodebian-users at lists.alioth.debian.org
> Subject: Re: [Neurodebian-users] fsl_sub error
> 
> Hi,
> 
> On Sat, Mar 10, 2012 at 09:48:15PM -0500, Lewis, Dave wrote:
> > I believe that I found an error (a bug waiting to happen) in
fsl_sub,
> > and the fix is simple.  This error isn't in the original version of
> > fsl_sub from the FSL website.  Our cluster has Debian 6.0 and fsl
> > 4.1.9-2~nd60+1.
> >
> > The person who installed FSL on our cluster (via NeuroDebian)
changed
> > the line
> >     #queue=long.q
> > to
> >     queue=all.q
> > (Our cluster was installed with all.q and debug.q.)
> 
> I'm not exactly sure what is happening with your cluster, but the
> solution you are proposing doesn't scale enough. There is not
guarantee
> that any particular SGE installation will have an 'all.q'. Meaning
> always submitting to all.q will cause an error on most clusters.
> Moreover, even if there is an all.q, most installations will have more
> fine-grained queue configurations.
> 
> SGE is capable of selecting an appropriate queue automatically, based
> on
> other information about a job. In the case of FSL this is mostly the
> expected runtime. Removing the queue specification from fsl_sub was
> necessary to make it compatible with most (apparently not all) SGE
> installations.
> 
> I'd be curious to know what particular configuration aspect prevents
> this from working on your cluster.
> 
> Michael
> 
> 
> --
> Michael Hanke
> http://mih.voxindeserto.de

Thanks Michael.

I see now that I wasn't clear enough about what I was suggesting.  I
don't mean that users should use all.q.  What I meant was that if people
are going to specify the queue via the queue= line in the NeuroDebian
version of fsl_sub, they should include the -q switch.

The version of fsl_sub from the FSL site has the -q switch before $queue
in the call to qsub:

    sge_command="qsub -V -cwd -shell n -b y -r y -q $queue -M $mailto -N
$JobName -m $MailOpts $LogOpts $sge_arch $sge_hold"

So with that version of fsl_sub, the $queue variable should just be the
queue name, like long.q (or in the case of our cluster, all.q).

The version of fsl_sub from NeuroDebian doesn't have the -q switch:

    sge_command="$qsub_cmd -V -cwd -shell n -b y -r y $job_timelimit
$queue -M $mailto -N $JobName -m $MailOpts $LogOpts $sge_arch $sge_hold"

So with that version of fsl_sub, the $queue variable should include the
-q switch.  That's all fine, and it has the advantage that the queue
variable doesn't have to be defined in the NeuroDebian version of
fsl_sub.

The person who installed our cluster saw the line
    #queue=long.q
and changed it to 
    queue=all.q
because our cluster has all.q rather than long.q.  That would have
worked for the version of fsl_sub from the FSL website, which we were
used to using.  But it doesn't work for the NeuroDebian version, because
it doesn't include the -q switch.

One way to help people avoid the mistake that we made is to change the
queue= line to 
    #queue="-q long.q"

Then if someone decides to uncomment the line and change the queue name,
they'll see that they need to include the -q switch.

I believe that it would have worked if we had also kept that line
commented out.

Thanks,
Dave


Conserve Resources. Print only when necessary.

IMPORTANT NOTICE: This e-mail is meant only for the use of the intended recipient. It may contain confidential information which is legally privilegedor otherwise protected by law. If you received this e-mail in error or from someone who is not authorized to send it to you, you are strictly prohibited from reviewing, using, disseminating, distributing or copying the e-mail. PLEASE NOTIFY US IMMEDIATELY OF THE ERROR BY RETURN E-MAIL AND DELETE THIS MESSAGE FROM YOUR SYSTEM. Thank you for your cooperation.