well - i've performed some tests through natd on a test net to see what
causes the CPU load to shoot up on out production boxes.
i found the following uinteresting:
* using natd with the -nonat option still has the "context switch" problem.
the throughput is not much better than with nat on.
killing nat and bypassing diverts increases throughput significanctly.
* the above tests (netperf, netperf with a a 50MB binary ftp going) does
not cause natd to load the CPU significantly.
i'll try the following to try to see if the load the CPU:
* i'll try multiple tcp sessions...
* try multiple NAT sources...
ps - I can't use ipf/ipnat as it would mean a rewrite of lots of production
software. stuck with a bad situation i know.
Thanks for the suggestiosn so far.
t
-----Original Message-----
From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED]]On Behalf Of Matthew Emmerton
Sent: 07 February 2002 21:29
To: Tariq Rashid; [EMAIL PROTECTED]
Subject: Re: squeeze more performance out of natd?
> i've spent a good number of hours RTFMs, trying to make the best of a bad
> situtaion: userland natd instead of kernel-space nat.
I've been told that if you use ipf and ipnat, then you get the benefit of
kernel-space NAT. Have you investigated this to see how it compares to
natd/ipfw for your purposes?
--
Matt Emmerton
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-net" in the body of the message
intY has automatically scanned this email with Sophos Anti-Virus
(www.inty.net)
intY has automatically scanned this email with Sophos Anti-Virus (www.inty.net)
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-net" in the body of the message