> On a related note: Could somebody confirm that there is no way to
> restrict the resident set size of a process under Linux (since
> ulimit is not implemented)?
ulimit is implemented, it's just that not all portions of the kernel
fully observe it (yet). RSS limiting is on my TODO list, and I'm
currently working on related areas of the swapper, so it may happen
for 2.2.
> I understand that with a good swapping algorithm, such a restriction
> wouldn't be really needed
Even with good swapping, RSS limits are a very good way of letting an
administrator balance the performance of a machine.
> , but my entirely subjective impression is that the responsiveness
> of Linux degrades much more when heaviliy swapping than Solaris.
Much of that is being addressed. I've now got pretty much all of the
infrastructure for proper async swapping mechanisms (both write-ahead
and read-ahead) implemented, and Ingo has got a set of patches for
next-read prediction which would complement the read-ahead well. I'm
also doing a bit more work on kswap to batch up writes better, so we
should be able to improve the throughtput of swap under Linux quite
significantly for 2.2.
Cheers,
Stephen.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu