John,
Yeah, I had a feeling it was due to sampling errors option (what used to be called spin history ??). Though I could imagine a circular buffer for the memory of images as you stated "serveral images" in memory at once. Maybe I will look at that code in my spare time.
I think the problem with matlab and multi-processors is that matlab itself is not all that great on memory management. If left to itself matlab trys to eat up memory that then slows other memory hungry matlabs to a halt - (I have a student that has seen this with other processes not SPM related). I have a feeling that having spm do the "paging" might end up being better than having the system do the paging especially when matlab tries to get all available memory.
Thanks for the comments. I hope you were not offended by my comments...I was pretty frustrated this weekend as I had hoped to have the systems running unattended over the Labor Day weekend (here in the US). Luckily I did drag myself in to check on things and restarted them.
More importantly I take your comment on "fudge" a bit more seriously on whether we should utilize this correction or not.
I will look into the code on the order of the loops and see if I can make the change locally. I highly suspected something like this. I also suspect that others experimental designs probably don't exceed a few hundred images, maybe 500 as that is long being on task. Personally my longest session is 400 images, but that is at a TR of 1000ms. And with that experiment there is only 1 or 2 sessions. The experiment that I ran into memory issues was 200 images per session.
Thanks,
Robert
-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Robert C. Welsh, PhD
Research Investigator
Department of Radiology
University of Michigan
(734) - 764 - 2412 (fax)
[log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|