 |
 |
Hi, On Thu, Nov 01, 2007 at 02:08:22PM +0200, Antti Korvenoja wrote: > Hi! > This error (see previous postings) seems to be related to the file size. > I truncated the timeseries to various sizes and managed to execute the > fslstats command for timeseries containing up to 600 volumes. With > longer timeseries the system started swapping and came near to lock-down > while processing (I have 2 GB RAM and 3 GB swap on my laptop) and > finally produced the error. There seems to be drastic change in the > system behaviour when the critical file size is exceeded. The size of > the file I'm trying to process is slightly over 150 MB. I can confirm this behaviour on an AMD64 system with 12GB RAM running Debian etch. I uses a 4d image with 902 volumes with (uncompressed) slightly more than 2GB to test fslstats. Calling 'fslstats -R' is quick and uses only slightly more memory than the actual file size. However adding the '-k' option with a mask image changes the situation completely. fslstats consumes all physical memory (12GB!) and some swap space. Therefore the calculation take a lot longer, but doesn't abort (the machine has another 10GB of swap). This would be different of course, if all swap space is used as well. At the moment I haven't had the chance to look into fslstats code, but there was no change to it from 4.0.0 to 4.0.1. I'd be glad, if someone could try a similar thing using the binaries provided by FMRIB to confirm that this bug is not Debian/Ubuntu-specific. Currently I don't see, what could cause this bug on Debian/Ubuntu, but not on other systems. Thanks, Michael -- GPG key: 1024D/3144BE0F Michael Hanke http://apsy.gse.uni-magdeburg.de/hanke ICQ: 48230050