[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: wolverine kernel leaks

The main problem you're probably seeing is that swap space is not
currently reclaimed for running processes in the 2.4 model.  Swap
space is only reclaimed on process exit.  So if there is a leaky
program like Netscape that gets swapped out, chances are that the
amount of swap that it consumed will not get reclaimed until the
program exits.  This isn't really that unexpected under several Unix
VM systems.

The problem about your machine locking up is more serious, IMHO.
There are areas of the kernel that have very bad behavior in OOM (out
of memory) conditions.  This can cause your machine to lock up before
the kernel can take measures to recover from the out of memory
situation, which usually means killing processes.  The killer seemed
to have a bit of a slow trigger finger when actually doing the deed of
killing.  I think this may be fixed in the very latest ac kernel (but
I'm not sure).



On Wed, Mar 14, 2001 at 12:46:53AM -0500, Tom Diehl wrote:
> I think it is a 2.4 thing. Linus has stated several times that 2.4 makes
> more use of swap than previous kernels do. I run generic 2.4 and 2.4-ac kernels
> here and see over the course of several days large qtys of swap used.
> Part of this I attribute to netscape leaking. Shutting netscape down
> frees things up. Leaving netscape running will eventaully lock the machine.
> Without netscape running the machine happly uses various large qtys of
> memory and swap but appears to free things up as needed. FWIW this machine
> has 128 Megs in it with 256 Megs of swap. Memory * 2 per Linus's
> recommendation.
> Just my $.02,

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]