I upgraded my workstation from rhel4 to rehl5 (both x86_64) and then the matlab simulations consume much more memory than before. It used to be only 3 Gig and now more than 4 Gig. I traced down (maybe just partially) the problem to difference in loading mat files (saved variable files).

In rhel4, I load several mat files (total ~800M) and then clear them from memory, the memory used by matlab (as told by top) will be the same as a fresh start (~30M).

But in rhel5, I do the same thing, load then clear, the memory used by matlab is much higher than a fresh start, about 600M is still used by matlab. This is strange because the workspace in matlab is empty.

I think this must be due to some changes in memory management in the kernel from 2.6.9 to 2.6.18. So I am seeking help here from your kernel experts. I know you may not be matlab experts, but an insight into kernel may help a lot.

Thanks a lot

Lianqi

I use Matlab R2007a on Debian. Did you reinstall Matlab after upgrading? Try doing that. This may be due some of the configurations by Matlab. Again, I am just guessing. It is definitely not a problem of the kernel. Just reinstall, and if the problem persists, post the details.

I did a reinstallation, but it seems doesn't help. I posted a file which I am having problem with at http://download.yousendit.com/31ACB2492BFE7695 . If you have time, can you have a test, just load the file and then clear, and compare memory consumption.
Thanks a lot

Lianqi

Please give me sometime. I have work and will put it for download. I shall reply by tonight or next morning.

I have found the problem. It has nothing to do with the linux kernel, but has to do with the glibc. I use gentoo 2007.0 with glibc 2.4 and the problem is gone. The problem seems to be in the memory allocation system call changes in glibc 2.5 or 2.6.