Date: Fri, 14 Nov 1997 20:52:34 -0800 From: David Greenman <dg@root.com> To: Eric Anderson <keric@websidestory.com> Cc: freebsd-questions@FreeBSD.ORG Subject: Re: Ethernet problem with 2.2.5 and 3C509B Message-ID: <199711150452.UAA14789@implode.root.com> In-Reply-To: Your message of "Fri, 14 Nov 1997 16:43:44 PST." <Pine.BSI.3.96.971114164134.18807G-100000@mail.websidestory.com>
next in thread | previous in thread | raw e-mail | index | archive | help
>> I'm not sure I understand what the problem is. Please explain the nature >> of the problem you are having - other than the collision rate being a bit >> high, I don't see anything wrong with the above numbers. > >Why are the 3Com cards so much more efficient? Why are the collision >rates on the Intel cards sky high, and the collision rates on the 3Com's >so low? They aren't - exactly the opposite. The Intel cards are extremely aggressive with packet transmissions and insert no delays between packets. This can result in higher collision rates, but the numbers you're seeing want affect the performance of your network unless you are near 100% capacity all of the time (a 50% collision rate would reduce the ceiling by about 5%). FWIW, the DEC chip based cards are similarly aggressive. If you're really stressing over this, you can change the "linear_priority" configuration option in the if_fxp.c driver source to a "1" - this will change the collision backoff algorithm to add additional delay and reduce the collisions. >> You mentioned that you tried setting full duplex...what is the machine >> connected to? It must be connected to a switch for full duplex to >> work. Moreover, if the switch is a Cisco Catalyst (or Cisco router), >> you have to configure the switch port explicitly to full duplex - it >> will not autonegotiate. > >Ahhh.. funny you should say that - we have Cisco Catalyst switches. >RTFM eh? (forgive me, I'm new to Cisco switches) Ahhh...Catalyst switches. Yes, I've seen high collision rates with Cisco stuff - both with 'de' and 'fxp' cards. It's possible that their collision backoff algorithm is actually broken, but that's just a guess. Yeah, full duplex is the best way to fix that. -DG David Greenman Core-team/Principal Architect, The FreeBSD Project
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199711150452.UAA14789>