>>> [some unspecified kind of memory leak]

>>	- What kind of system activity is going when you notice the
>>	  most lossage?

> One thing that kills it for some reason is named.  We are running a
> dedicated name server machine, and all it is running is named.
> Sometimes there is something like 40MB of swap claimed, and the
> aparent sum total of the processes is something like 15MB.

> Yes.  On my machine here at school, I frequently see the swap
> consumption jump to 25 megabytes, and the X server is only claiming
> about 6MB, and other clients have a total of 8.  However, when the X
> server exits, it drops to about 3MB of swap (never back to zero).

Both of these remind me of the shadow VM object leak.  Both named and
the X server are long-running processes that fork children.

Seems to me we should get something, anything, in to fix this, even if
it's a scavenger that's run once a minute to find these extra shadow
objects and nuke them.

But then, given how core has been resisting bounce buffers in the i386
port, we may have to either live with (yet another) brokenness in the
VM subsystem or keep private patches for it.  Not that private patches
bother _me_ particularly; I've got over 100K of private patches
already.  But I know they bother some people.

					der Mouse

			    mouse@collatz.mcrcim.mcgill.edu