Print

Print
Hi Dara,

Apologies for the late reply to this one. It was assigned to myself and got lost in my inbox.

The most memory intensive component of contrast_mgr will be the correlation matrix for the parameter estimates. This will be Num of voxels * (Num of parameter estimates)^2. Clearly, for designs with large numbers of EVs this could be very large. This is loaded in at the start and is needed by all of the subsequent contrasts and so is maintained in memory throughout. There is no straightforward way of avoiding that.

I don't think contrast_mgr should be much more memory intensive for increased numbers of contrasts, as they are processed and saved to disk one at a time.  

Cheers, Mark.

----
Dr Mark Woolrich
EPSRC Advanced Research Fellow University Research Lecturer

Oxford University Centre for Functional MRI of the Brain (FMRIB),
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK.

Tel: (+44)1865-222782 Homepage: http://www.fmrib.ox.ac.uk/~woolrich

From: Dara Ghahremani <[log in to unmask]>
Date: 30 August 2006 20:45:28 BDT
To: [log in to unmask]
Subject: Re: [FSL] 32-bit vs. 64-bit FSL 3.3.7
Reply-To: FSL - FMRIB's Software Library <[log in to unmask]>

Hi Steve,

The 64-bit did indeed finish :). Forgive me for being unfamiliar with the way that contrast_mgr works, but  is there a reason why it needs to keep information about all of the contrasts in memory at the same time? Would it be possible to compute each contrast independent of the others?

Cheers,
Dara

On Aug 30, 2006, at 12:05 AM, Steve Smith wrote:

Hi - with such a huge number of EVs and contrasts it is possible that such a large amount of RAM is required, yes, which would explain why contrast_mgr cannot complete on a 32-bit system.  You didn't say whether the 64-bit run did finish in the end?  It may have been "idle" because of a large amount of swapping - you may want to increase your RAM and SWAP.


Cheers, Steve.



On 30 Aug 2006, at 07:58, Dara Ghahremani wrote:


Hi,

In using Feat to estimate a large model (31 EVs, 37 contrasts), I get an "uncaught exception!" error when running the contrast_mgr command on the 32-bit version of 3.3.7, but not when using the 64-bit version on the same computer (running SUSE Linux 2.6.16.21-0.13-smp).

We've noticed that the 64-bit version uses a lot of virtual memory (5GBs) while the CPU is idle. here is the "top" output:

  PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
21214 dara 18 0 5147m 2.7g 756 D     0          75.9     0:34.53         contrast_mgr

Is the allocation of 5GBs of RAM an artifact of the 64 bit code or does it just reflect the size of the analysis?

Cheers,
Dara

----------------------------------------------------------------
Dara Ghahremani, Ph.D.
Poldrack Lab
Psychology Department
Franz Hall, UCLA
Los Angeles, CA 90095-1563
----------------------------------------------------------------



---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director,  Oxford University FMRIB Centre

FMRIB, JR Hospital, Headington, Oxford  OX3 9DU, UK
+44 (0) 1865 222726  (fax 222717)
[log in to unmask]    http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------