From owner-freebsd-hackers Sat Jun 24 12:14:46 1995 Return-Path: hackers-owner Received: (from majordom@localhost) by freefall.cdrom.com (8.6.10/8.6.6) id MAA21751 for hackers-outgoing; Sat, 24 Jun 1995 12:14:46 -0700 Received: from gndrsh.aac.dev.com (gndrsh.aac.dev.com [198.145.92.241]) by freefall.cdrom.com (8.6.10/8.6.6) with ESMTP id MAA21743 for ; Sat, 24 Jun 1995 12:14:41 -0700 Received: (from rgrimes@localhost) by gndrsh.aac.dev.com (8.6.11/8.6.9) id MAA13409; Sat, 24 Jun 1995 12:14:37 -0700 From: "Rodney W. Grimes" Message-Id: <199506241914.MAA13409@gndrsh.aac.dev.com> Subject: Re: FreeBSD as a router To: dennis@et.htp.com (dennis) Date: Sat, 24 Jun 1995 12:14:36 -0700 (PDT) Cc: freebsd-hackers@freebsd.org In-Reply-To: <199506241710.NAA19368@mail.htp.com> from "dennis" at Jun 24, 95 01:10:00 pm X-Mailer: ELM [version 2.4 PL24] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 1603 Sender: hackers-owner@freebsd.org Precedence: bulk > > >> > >> > >> On Thu, 22 Jun 1995, Jordan K. Hubbard wrote: > >> > >> > That said, be aware that any kind of UN*X box doesn't exactly compete > >> > with a Cisco in terms of performance. They throw raw hardware at the > >> > problem whereas we have to do it the hard way, in software. > >> > >> The bottleneck certainly can't be in the CPU can it? Where is the > >> bottleneck with PCI and a good 486 motherboard? > > > >The bottleneck is that you have to wait for full frame reception > >before you get an interrupt to tell you to go look at the header > >to decide what to do with the packet. > > > >In dedicated router hardware they use the trick of interrupting > >the CPU after N bytes have been recieved (N is programmable) so > >they can actually decide what to do with the packet before it is > >even completly received. > > > > > This is not necessary to get good throughput, although it wouldn't hurt. You > can still get 5mbs without this, which is plenty. Not when you start looking at 100MB/sec ethernet it isn't!!! Sure 400 to 500 KByte/sec for 10MB/sec ethernet routing is just fine by me, but as soon as I reproduce the numbers for 100MB/sec routing you will see what I mean about we need to make some improvements. We need to get that routing performance into the 50MB/sec range and we are not even close. (I seem to recall about 20MB/sec, but am not sure right now, too many numbers floating around in my head). -- Rod Grimes rgrimes@gndrsh.aac.dev.com Accurate Automation Company Reliable computers for FreeBSD