Hi,

Just to add to Romain's email, that preproc and postproc jobs in bepostx are 
the most demanding memory-wise. So you should be able to get away by simply 
increasing the amount of memory requested for the preproc and postproc, which 
normally do not take more than a few minutes to run.

Cheers
Stam


On 21 Oct 2013, at 13:24, romain valabregue <[email protected]> wrote:

> Hello
> 
> I also get similar problem : the diffusion data are quite to big. I had to 
> repeat the bedpostx 2 or 3 times in order to make it complete.
> So I guess it depend of the load of each node. It was not killed by the sge 
> for any ressource limit, but it was kill by the processor because of memory 
> issue I guess (so I could not see it on the cluster log file)
> 
> I had a similar problem when trying to do probtrackx. (the jobs were kill by 
> the cluster) The solution was just to ask a big amount of memory (qsub -l 
> mem=12gb)  so that the cluster will only run one or two probtrack process on 
> the same node.
> 
> I do not know how to change fslsub so that bedpostx add the extra argument to 
> qsub (ie : -l mem=...)
> You can also use qalter as suggest by Colin, but then you have to be quick to 
> make the qalter during the preproc step, before each element of the array 
> start.
> 
> 
> Cheers
> 
> Romain
> 
> 
> Le 17/10/2013 00:24, Huang Su-Chun a écrit :
>> You are right, the missing slices mainly located in the middle of the brain. 
>> So I guess the problem comes from our grid engine....
>> 
>> I tried use -nf=3, but the process fails all the time. I guess it too 
>> demanding for the computer
>> 
>> Thank you,
>> 
>> 
>> Su
>> 
>> 
>> 2013/10/16 Glasser, Matthew <[email protected]>
>> Do you know that your cluster is not killing some of the jobs?  Which slices 
>> are missing?  If it is mainly the middle ones, I'd suspect that they are 
>> exceeding some limit for the cluster.  
>> 
>> For HCP data you can successfully model 3 fibers (--nf=3) and should use the 
>> -g option with FSL 5.0.5.  I'd leave fudge and bi at their defaults unless 
>> you have a reason to change them.  --model=2 is correct. 
>> 
>> Peace,
>> 
>> Matt.
>> 
>> From: Huang Su-Chun <[email protected]>
>> Date: Wednesday, October 16, 2013 4:27 PM
>> To: Matt Glasser <[email protected]>
>> Cc: Stamatios Sotiropoulos <[email protected]>, 
>> "[email protected]" <[email protected]>
>> 
>> Subject: Re: [HCP-Users] BedpostX failed
>> 
>> It happened on subject 100307 and 105115. I am running the job in a grid 
>> engine. 
>> The command line I use is listed below:
>> 
>> bedpostx 
>> /NAS_II/Public-Data/Parcellation/Q2_released_data/105115/105115_Diffusion_preproc/T1w/Diffusion
>>  --nf=2 --fudge=1  --bi=1000 --model=2
>> 
>> Thank you,
>> 
>> Su
>> 
>> 
>> 
>> 2013/10/15 Glasser, Matthew <[email protected]>
>> What subject is this?
>> 
>> What command line did you call bedpostx with?
>> 
>> Are you running this on your local machine or in some cluster environment?
>> 
>> Note that a new version of FSL was just released with improvements to 
>> bedpostx that you need to correctly process HCP data (not fixing missing 
>> slices, but correcting for the effects of gradient nonlinearities on the 
>> bmatrix), so I highly recommend you update.   
>> 
>> Peace,
>> 
>> Matt.
>> 
>> From: Huang Su-Chun <[email protected]>
>> Date: Tuesday, October 15, 2013 6:12 PM
>> To: Stamatios Sotiropoulos <[email protected]>
>> Cc: "[email protected]" <[email protected]>
>> Subject: Re: [HCP-Users] BedpostX failed
>> 
>> Hi Stam,
>> 
>> I checked the bedpostX results and it didn't have the same number of slices 
>> as the input data. It caused problem when I want to to probabilistic 
>> tracking. And this missing slices problem happens quite a lot! 
>> 
>> Thank you,
>> 
>> Su
>> 
>> 
>> 2013/10/15 Stamatios Sotiropoulos <[email protected]>
>> Hi 
>> 
>> Is it just the message that worries you? Have you checked whether bedpostx 
>> results have the same number of slices as the input data? Notice that the 
>> diffusion data are transferred to T1 space and lots of inferior/superior 
>> slices may be empty. 
>> 
>> Cheers
>> Stam
>> 
>> 
>> On 15 Oct 2013, at 04:47, Huang Su-Chun <[email protected]> wrote:
>> 
>>> Hi dear experts,
>>> 
>>> I am running bedpostX on the HCP diffusion data, however, it seems that 
>>> sometimes (60% of chance) the bedpostX ended before finishing all the 
>>> slices. Take the HCP data for example, it should have 145 slices, but the 
>>> betpostX stopped at slice 103 then show a message said "All slice 
>>> processed."
>>> 
>>> Does anyone know where the problem might be? Thank you!
>>> 
>>> Cheers,
>>> 
>>> Su
>>> 
>>> -- 
>>> Su-Chun Huang
>>> Research Scientist
>>> Integrated Brain Imaging Center
>>> Department of Radiology
>>> University of Washington
>>> Seattle, WA 98195
>>> phone: 206-687-0113
>>> _______________________________________________
>>> HCP-Users mailing list
>>> [email protected]
>>> http://lists.humanconnectome.org/mailman/listinfo/hcp-users
>>> 
>> 
>> 
>> 
>> 
>> -- 
>> Su-Chun Huang
>> Research Scientist
>> Integrated Brain Imaging Center
>> Department of Radiology
>> University of Washington
>> Seattle, WA 98195
>> phone: 206-687-0113
>> _______________________________________________
>> HCP-Users mailing list
>> [email protected]
>> http://lists.humanconnectome.org/mailman/listinfo/hcp-users
>> 
>>  
>> The materials in this message are private and may contain Protected 
>> Healthcare Information or other information of a sensitive nature. If you 
>> are not the intended recipient, be advised that any unauthorized use, 
>> disclosure, copying or the taking of any action in reliance on the contents 
>> of this information is strictly prohibited. If you have received this email 
>> in error, please immediately notify the sender via telephone or return mail.
>> 
>> 
>> 
>> 
>> -- 
>> Su-Chun Huang
>> Research Scientist
>> Integrated Brain Imaging Center
>> Department of Radiology
>> University of Washington
>> Seattle, WA 98195
>> phone: 206-687-0113
>>  
>> The materials in this message are private and may contain Protected 
>> Healthcare Information or other information of a sensitive nature. If you 
>> are not the intended recipient, be advised that any unauthorized use, 
>> disclosure, copying or the taking of any action in reliance on the contents 
>> of this information is strictly prohibited. If you have received this email 
>> in error, please immediately notify the sender via telephone or return mail.
>> 
>> 
>> 
>> 
>> -- 
>> Su-Chun Huang
>> Research Scientist
>> Integrated Brain Imaging Center
>> Department of Radiology
>> University of Washington
>> Seattle, WA 98195
>> phone: 206-687-0113
>> _______________________________________________
>> HCP-Users mailing list
>> [email protected]
>> http://lists.humanconnectome.org/mailman/listinfo/hcp-users
>> 
> 
> _______________________________________________
> HCP-Users mailing list
> [email protected]
> http://lists.humanconnectome.org/mailman/listinfo/hcp-users
> 


_______________________________________________
HCP-Users mailing list
[email protected]
http://lists.humanconnectome.org/mailman/listinfo/hcp-users

Reply via email to