On Sat, 10 Nov 2001 at 09:11, Anuradha Ratnaweera wrote:
> How about using preemptive patch on a firewall, running tranparent
> proxy (squid/iptables) and also a little bit of qos/fair queuing stuff
> (includes XFS on software raid 1)?
I really don't think it will help much, but then I don't know enough
kernel internals (and theory) to be able to give authoritative feedback. I
was reading an article about how kernel preemption works, though, and
bottomline is with kernel preemption you sacrifice overall kernel
throughput for responsiveness that comes from being able to put certain
things on hold, switching between kernel tasks very quickly (like
So on a pure server (ie: one that is not used as a console/workstation at
the same time) I wouldn't use the kernel preemption patch. On everything
else, I would. :)
> Will try it today and send some feedback.
That'll be great. I'm interested in finding out how it actually performs
on a "pure server". :)
I wonder: there is a component of preemption that is XFS-centric, and the
XFS code is already preemptible. Would the XFS developers have
authoritative information on how the performance of XFS varies with
preemption enabled on an IO-bound system?
Federico Sevilla III :: jijo@xxxxxxxxxxxxxxxxxxxx
Network Administrator :: The Leather Collection, Inc.
GnuPG Key: <http://jijo.leathercollection.ph/jijo.gpg>