From owner-freebsd-hackers Sun Jun 28 12:29:35 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id MAA29344 for freebsd-hackers-outgoing; Sun, 28 Jun 1998 12:29:35 -0700 (PDT) (envelope-from owner-freebsd-hackers@FreeBSD.ORG) Received: from duey.hs.wolves.k12.mo.us (root@duey.hs.wolves.k12.mo.us [207.160.214.9]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id MAA29338 for ; Sun, 28 Jun 1998 12:29:29 -0700 (PDT) (envelope-from cdillon@wolves.k12.mo.us) Received: from duey.hs.wolves.k12.mo.us (cdillon@duey.hs.wolves.k12.mo.us [207.160.214.9]) by duey.hs.wolves.k12.mo.us (8.8.7/8.8.7) with SMTP id OAA22942; Sun, 28 Jun 1998 14:29:29 -0500 (CDT) (envelope-from cdillon@wolves.k12.mo.us) Date: Sun, 28 Jun 1998 14:29:28 -0500 (CDT) From: Chris Dillon X-Sender: cdillon@duey.hs.wolves.k12.mo.us Reply-To: Chris Dillon To: Mike Tancsa cc: hackers@FreeBSD.ORG Subject: Re: Will 8 Intel EtherExpress PRO 10/100's be a problem? In-Reply-To: <3595e10d.1180692826@mail.sentex.net> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG On Sun, 28 Jun 1998, Mike Tancsa wrote: > On Thu, 25 Jun 1998 23:32:35 -0500 (CDT), in sentex.lists.freebsd.misc > you wrote: > >Within the next few months, i will be needing to set up a router for our > >internal network, tying together 7 networks, with some room to grow. I > >plan on buying a rather expensive chassis from Industrial Computer source. > > You could probably buy 3 plain-jane pentiums for the price of the > fancy one you are talking about and not have to worry about cramming > so many cards into one box and overloading the PCI bus. If the boxes > are going to act as routers and NAT machines, you really dont need > that much horse power/RAM/HD space. Where did I mention how much RAM and HD space I was going to use? :-) I do know that those are not something you need an excess of for a router (RAM, sometimes). I only planned on putting 64MB ECC SDRAM in it. It could probably do with 16 or 32, but RAM is cheap these days. I would even do without an HD (without moving parts, that is), if I could. I'm thinking about using these flash based drives I've been seeing lately, if the price is right. Even PicoBSD on a floppy (or bootable ZIP disk?) might be an option. As for the horsepower, the more the better, since it would increase my peak bandwidth capacity (assuming the bus isn't already saturated, which it shouldn't be) and reduce latency. It is true that going with something ultra-fast means a lot of heat, and reduced reliability and life. I do agree that multiple cheaper boxes might be better for removing a single point of failure (i.e., if one box failed, the networks on that box would be separated, but the rest would be communicating), but that would require a few more NICs total (not a big deal cost-wise), a switch to interconnect all of them (now we're talking money, and if not a switch, then a hub, or then even more NICs are required for point-to-point links between boxes). I'd be shoving each of these machines totally full of NICs for interconnection along with the NICs required to service the our networks, filling up their relatively small PCI busses. The total cost is back up to where it was with the single high-capacity high-reliability machine, and complexity is increased by several orders of magnitude. I'm going to take a quick guess at how I could use multiple boxes, just to see if it would work for me. Assuming I used common motherboards with 4 PCI slots, I would need one NIC for interconnection, leaving three slots free for our internal networks. So, I'd need three of these boxes and a switch for the interconnection. Four boxes if I want to grow. The switch just became a single point of failure, not to mention a 100Mbit/sec bottleneck between any two boxes. So, instead, I put two NICs in each machine for interconnection, creating some kind of simple star or ring. This still creates a sort of bottleneck between boxes, but increases availability significantly. With two slots free, I'd now need four boxes to meet my needs, five for growth. Using boards with 5 PCI slots would change this scenario a little bit, but would still pretty much stick me in the same boat, especially if I grow. Basically, I'm just buying a box with 9 PCI slots. If I grow past 8 networks (which I seriously doubt), I'll buy another and I'll have "multiple" boxes to share the job. :-) I do believe in splitting up work between multiple boxes when possible, but in _this particular scenario_, it just doesn't seem ultimately feasable. I do appreciate all of the comments people are giving me. That was the secondary reason why I posted the question to the list (first being wether what I had originally planned would actually work), rather than go blindly. :-) All of the responses to this point have made me think twice about one thing or another, or altered my direction slightly. Your response, Mike, along with Kevin's from Atipa, made me think about how I actually _could_ use multiple boxes, which I basically just laid out above. -- Chris Dillon - cdillon@wolves.k12.mo.us - cdillon@inter-linc.net /* FreeBSD: The fastest and most stable server OS on the planet. For Intel x86 and compatibles (SPARC and Alpha under development) (http://www.freebsd.org) */ To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message