From owner-freebsd-hardware Fri Apr 24 10:59:07 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id KAA14466 for freebsd-hardware-outgoing; Fri, 24 Apr 1998 10:59:07 -0700 (PDT) (envelope-from owner-freebsd-hardware@FreeBSD.ORG) Received: from scanner.worldgate.com (scanner.worldgate.com [198.161.84.3]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id KAA14452; Fri, 24 Apr 1998 10:59:01 -0700 (PDT) (envelope-from marcs@znep.com) Received: from znep.com (uucp@localhost) by scanner.worldgate.com (8.8.7/8.8.7) with UUCP id LAA22100; Fri, 24 Apr 1998 11:58:55 -0600 (MDT) Received: from localhost (marcs@localhost) by alive.znep.com (8.7.5/8.7.3) with SMTP id LAA27836; Fri, 24 Apr 1998 11:57:54 -0600 (MDT) Date: Fri, 24 Apr 1998 11:57:53 -0600 (MDT) From: Marc Slemko To: Open Systems Networking cc: freebsd-advocacy@FreeBSD.ORG, freebsd-hardware@FreeBSD.ORG Subject: Re: *** Real Action Item: SPECweb In-Reply-To: Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-freebsd-hardware@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org On Fri, 24 Apr 1998, Open Systems Networking wrote: > On Fri, 24 Apr 1998, Don Wilde wrote: > > > With that in mind, I'd like to solicit comments on what the best hardware > > (single CPU) we can get to give us the highest TCPIP network throughput. If > > you'll look at http://www.SPECbench.org, you'll see the disclosures for the > > systems Novell and others have used to achieve their performance. The key to > > this thing is maximum bandwidth, disk performance, and maximum intelligence on > > the IO channels. Even if we do it by using a gigabit card set and fiber, we're > > after exposure for FreeBSD and Apache, not necessarily long-lasting benchmark > > numbers. Whatever you do, your results had better be limited by something other than the network. The drivers are important, the card design is important, etc. but whatever you do the point of the benchmark is to not be network limited. > > I have no doubt in my mind that the intel card david is using in WCARCHIVE > is THE card to pump out data in insane quantities. I used to love the DEC > based cards and still do, and the new tx0 SMC driver is nice to, but I > think the Intel Pro/100B witht he fxp driver is THE card to use. get a MB > with 4 PCI slots and slam 4 of those bad boys in there. I dont know if > there are multiport versions of them but even if not these cards spit > fire. Thats for PCI if you want anything else im not sure. I dont know > about the ATM driver never used it. We should see what david says. I think > even just 4 single port 100B's could outdo by a nice margin whatever > novell used. IMO anyway. David? Generally TYAN MB's I think are the best. > But it all depends on the CPU/Chipset combo. > So i'm going to go look at the spechweb benchmark place before hunting > down good combo's. For web benchmarks jacking up the MTU beyond what you can get with Ethernet can give significant wins. FDDI can help a bit, ATM can help even more. Can't comment on driver support in FreeBSD, and using a NIC with a bad driver or design can hurt things a lot. See ftp://ftp.cup.hp.com/dist/networking/briefs/WebMTU.pdf for some example numbers on the impact this can have. This is the sort of thing you are dealing with when competing against SPECweb96 numbers. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hardware" in the body of the message