Hi,
Have you tried setting maxMem to 2^28 and having the
swap file set at a minimum of 512Mb? I'm not
too sure how multiprocessor machines work, but I know
that Matlab itself doesn't use threads - so won't take
advantage of a second processor. So perhaps this is more
of an OS problem?? I agree that the memory footprint
when SPM is running is excessive, so it may be a
combination of OS and s/w.
Sorry can't be more help.
Yours,
Jon.
_____________________________________________________
Jonathan Brooks Ph.D. (Research Fellow)
Magnetic Resonance and Image Analysis Research Centre
University of Liverpool, Pembroke Place, L69 3BX, UK
tel: +44 151 794 5629 fax: +44 151 794 5635
On Sat, 2 Sep 2000, Robert Welsh wrote:
>Hi,
>
>I have set maxMem to 2^27 (128 meg) on a dual processor linux (PIII 733) box with 512Meg.
>
>I have set up two spm realignment jobs: 5 subjects, 6 sessions each.
>
>Both started out fine with near 100% cpu usage on each of the respective processors. This latested for about 30 minutes.
>
>However, now 8 hours later, only one job is running at top speed. This job is at 357 meg. The stalled job is at 24 meg. Most likely stalled because of massive swapping. 357 is certainly larger than the 128meg I set in maxmem. There is OBVIOUSLY a bad bug in the code.
>
>We also notice this behavior on a dual processor sun ultra 60 (dual 360 mhz, 256meg, maxMem = 2^25).
>
>Apparently there is a big time memory leak in spm. Since this is a realignment job it seems that not many images need to be in memory at any given time.
>
>I am not sure if others with dual processors have noticed this behavior or not. (ie running parallel spm jobs)
>
>Can this be looked into ASAP as the code currently negates the purpose of dual processors? I would like to fix the code but obviously won't know exactly where to look.
>
>Robert Welsh
>
>p.s. (I search mailbase for "memory leak" and even "leak" and found none)
>
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|