martin,
thanks for the quick reply.
that makes sense, but it would still be very useful to dole the jobs
out to whatever node is least loaded, or to keep the job in queue
until a node becomes available. in fact this is what happens with
FEAT as far as i can tell.
so, i'm still hoping that there's a way to get this to go through the
queue system.
thanks
lokke
On Oct 12, 2007, at 1:41 PM, Martin Kavec wrote:
> Hi Lokke,
>
> I think that FAST can not be parallelized (as simply) as other FSL
> tools.
> Segmentation is performed on the whole image at once and not slice
> by slice.
> So I assume that the FSL guys didn't configure it for execution on the
> cluster, as it would not give us the results faster anyway.
>
> Martin
>
> On Friday 12 October 2007 19:14:31 Lokke Highstein wrote:
>> hello list,
>>
>> i'm sure you are all sick of my constant posts about the xserve
>> cluster, but i have another question.
>>
>> we are successfully getting FEAT, bedpost, and probtrack jobs to run
>> through the SGE, and get distributed across the cluster, but when we
>> run FAST, it doesn't get submitted to the queue, and only runs on the
>> head node.
>>
>> i tried reading the fast.tcl file and searching for the terms long.q
>> all.q fsl_sub and SGE, didn't find any of them there.
>>
>> is there something i need to edit to make FAST jobs get submitted to
>> my SGE queue?
>>
>> thanks,
>>
>> lokke
|