thanks for the suggestions, it seems like i am getting somewhere...
at least it's trying to use the SGE now...
i followed suggestions from both sheyna and martin as follows:
i had installed SGE in /usr/local/gridengine which seemed to be the
default location. i went ahead and edited /usr/local/fsl/bin/fsl_sub
to point to /usr/local/gridengine in all the places that it was
referenced, rather than creating symbolic links.
then i ran the qconf -aq command on veryshort.q short.q long.q and
verylong.q changing the values for both processors and slots to equal
2 (i am running a dual 1ghz ppc mac) - one thing i noticed was that i
had to be root to do this, not sure if that is causing my later
problems, but i installed the SGE as root.
since i didn'tt play with the original set of queues i didn't have to
edit the bedpostx files as they all seemed to refer to the queues
that i had set up.
i also edited the bash and tcsh configuration files (we have users
that use both) to have the FSLDIR, SGE_ROOT, and FSLMACHINELIST
environment variables set to their respective places, and to have
them added to the path.
however, after all this, i tried to run bedpost, and got the
following error:
Errors: Unable to run job: Job was rejected because job requests
uknown queue "short.q".
Exiting
Unable to run job: Job was rejected because job requests uknown queue
"long.q".
Exiting
Unable to run job: Job was rejected because job requests uknown queue
"long.q".
Exiting
after that i decided to check that the queues were set up right and
ran the following:
bash-2.05b$ qconf -sql
all.q
long.q
short.q
verylong.q
veryshort.q
bash-2.05b$
any suggestions where i went wrong with this?
thanks for any help, i feel like i am so close!
lokke
On Sep 17, 2007, at 4:21 PM, Sheyna Gifford wrote:
> Lokke-
>
> You question is neither dumb nor trivial. I have a few degrees and
> it still
> took me a while of hacking and playing to figure it out. ;)
>
> Yes, you need to edit fsl_sub in several key places, but you ALSO
> need to
> edit bedpostx_postproc.sh and bedpostx_preproc.sh. Follow these
> simple
> steps:
>
> Step 1: Edit fsl_sub in these key places
>
> A) Tell it where you settings.sh file is
>
> METHOD=SGE
> if [ "x$SGE_ROOT" = "x" ] ; then
> if [ -f /bsp/cluster/sge/default/common/settings.sh ] ; then
> . /bsp/cluster/sge/default/common/settings.sh
> elif [ -f /bsp/cluster/sge/default/common/settings.csh ] ; then
> . /bsp/cluster/sge/default/common/settings.csh
> else
> METHOD=NONE
> fi
> fi
>
>
> B) Tell it who to email
> queue=all.q
> #queue=short.q
> [log in to unmask]
> [log in to unmask]
> MailOpts="n"
>
>
> C) Tell it what the name of your SGE queue or queues is (or are).
> Ours is,
> standardly, all.q. I commented everything else out.
>
> map_qname ()
> {
> queue=all.q
>
> # if [ $1 -le 20 ] ; then
> # queue=all.q
> # elif [ $1 -le 120 ] ; then
> # queue=all.q
> # elif [ $1 -le 1440 ] ; then
> # queue=all.q
> # else
> # queue=all.q
> # fi
> # echo "Estimated time was $1 mins: queue name is $queue"
> }
>
>
> STEP TWO: Now change to reflect the queue or queues that you want
> to use:
>
> node02:/bsp/cluster/local/fsl/bin sgifford$ more bedpostx_preproc.sh
> #!/bin/sh
>
> #$ -cwd
> #$ -q all.q
> #$ -S /bin/sh
> #$ -V
> #$ -N bpx_preproc
> #$ -m as
>
>
>
> STEP THREE: Now change bedpostx_postproc.sh to reflect the queue or
> queues
> that you want to use:
>
> node02:/bsp/cluster/local/fsl/bin sgifford$ more bedpostx_postproc.sh
> #!/bin/sh
>
> #$ -cwd
> #$ -q all.q
> #$ -S /bin/sh
> #$ -V
> #$ -N bpx_postproc
> #$ -m ae
>
>
> Since doing this, I am in the land of happy bedpostxing. Please note:
> paralleling only takes place AFTER the preprocessing stage is
> finished. So,
> bedpostx will sit on one processor on one node, until all of the
> slices are
> processed. Then, xfibres takes off on every node designated by
> FSLMACHINELIST.
>
> Best of luck, Lokke!
>
> - seg
|