Walt Mankowski on 25 Nov 2003 13:20:03 -0500 |
On Mon, Nov 24, 2003 at 11:39:56AM -0500, Reilly, Kris wrote: > The 2.4 kernels have a 2G memory limitation per pid. I ran into this same > problem performance tuning a mysql database. Our solution was to break the > process into multiple processes. Parent/Child isn't good enough either, as > the parent pid will reach the 2G limit and then hang. That's what I'm finding empiracally, too. I wrote a little test program yesterday that just malloc'ed huge chunks of memory to see how big a process could get until malloc() started returning ENOMEM. I found I could get to around 2 GB - 32 MB. And, perhaps incidentally, the kernel running on this box is right around 32 MB in size. We've been talking anyway about how to split up the process so that we can test more genes at once, and this finding will likely speed up that process. As someone else pointed out in this thread, this is already a cpu-bound process, so for performance reasons we really need to try to avoid paging anyway. Walt Attachment:
pgpCB3UzcoQ85.pgp
|
|