On 27-Apr-98 David Edelsohn wrote:

>> The memory usage under discussion here is not in the server's space.
>
>       Actually, it probably is.
>
>       You are describing an X Windows client and server running on a
> local system.  X Windows shared memory extension probably is enabled --
> especially for transferring pixmaps.  That's just one possibility for a
> memory leak which starts in the client and is not released in the server.

Don't I wish it were that simple.

First, shared memory would be visible in the server's vm space.  This is not.

Second, the problem is the same without sysvshm even in the kernel.

I make no claim to knowing yet what causes this, but I've attacked it from a
number of angles over a fair space of time, and I've ruled out a lot.  The
problem has proven remarkably resistant to any number of "cures", short of
simply restarting the session.

Any number of things -- different mallocs, different window managers, different
file managers, different X servers, different kernels, different clients and
data under test -- change the numbers, but not the basic problem.

There remains this memory, piling up on the active list, which is NOT reflected
in the vm space (according to ps) of any program.  Eventually it migrates to
the inactive list, and from there to swap, and eventually it grinds the machine
to a thrashing halt.

Chuck Cranor just suggested a way to use ddb with X, and if I can do that I
hope to learn a lot more real soon now....


--Jim Howard  <jiho@mail.c-zone.net>


----------------------------------
E-Mail: jiho@mail.c-zone.net
Date: 27-Apr-98
Time: 12:04:08

This message was sent by XFMail
----------------------------------