From owner-freebsd-hackers Tue Jul 10 5:15:58 2001 Delivered-To: freebsd-hackers@freebsd.org Received: from ns5.pacific.net.au (ns5.pacific.net.au [203.143.252.30]) by hub.freebsd.org (Postfix) with ESMTP id 34F1E37B403 for ; Tue, 10 Jul 2001 05:15:55 -0700 (PDT) (envelope-from mckay@thehub.com.au) Received: from dungeon.home (ppp203.dyn249.pacific.net.au [203.143.249.203]) by ns5.pacific.net.au (8.9.0/8.9.1) with ESMTP id WAA20042; Tue, 10 Jul 2001 22:15:44 +1000 (EST) Received: from dungeon.home (localhost [127.0.0.1]) by dungeon.home (8.11.3/8.11.1) with ESMTP id f6ACGDF25604; Tue, 10 Jul 2001 22:16:13 +1000 (EST) (envelope-from mckay) Message-Id: <200107101216.f6ACGDF25604@dungeon.home> To: freebsd-hackers@freebsd.org Cc: mckay@thehub.com.au, Kenneth Wayne Culver , Terry Lambert Subject: Re: more on latency References: <3B4A0F74.672D7B27@mindspring.com> In-Reply-To: <3B4A0F74.672D7B27@mindspring.com> from Terry Lambert at "Mon, 09 Jul 2001 20:09:24 +0000" Date: Tue, 10 Jul 2001 22:16:13 +1000 From: Stephen McKay Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.ORG On Monday, 9th July 2001, Terry Lambert wrote: >Kenneth Wayne Culver wrote: >> >> I think I found the reason that my FreeBSD box is performing >> so poorly as a NATing router. When I do an ipnat -l to see >> what "active connections" are there on the router, a list >> about 3 pages long (using ipnat -l | more) appears. I think >> maybe it's having trouble because for every packet coming in >> and out of the router, it's got to look at that list of >> active connections for the right one to send to and from. Is >> there any way to make connections that aren't being used go >> away from the NAT faster? Thanks a lot. > >Don't run unnecessary daemons. > >The pcb lookups are a linear traversal, as well, and for >a large number of connections, the calllout wheel for >timers sucks. I can't imagine even the most inefficiently coded linear traversal causing this problem given the beefy machine being used. I set up a cable sharing system for friends of mine and it is a Pentium 100 with 2 ISA NICs! That system adds no more than 2 or so ms to the latency with 3 simultaneous counterstrike players. I used ipfw and natd in a trivial configuration on 4.3-R. I wonder if the problem is a lack of mbufs or some similar misconfiguration tragedy. "netstat -m" and "top" output might be helpful. Stephen. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message