Date: Wed, 5 Mar 2003 10:33:45 -0500 (EST) From: Andrew Gallatin <gallatin@cs.duke.edu> To: David Leimbach <leimy2k@mac.com> Cc: Christopher Fowler <cfowler@outpostsentinel.com>, freebsd-current@freebsd.org Subject: Re: IP over IEEE1394? Message-ID: <15974.6361.471584.917266@grasshopper.cs.duke.edu> In-Reply-To: <43A441A1-4F17-11D7-A732-0003937E39E0@mac.com> References: <1046874658.5666.27.camel@cfowler.outpostsentinel.com> <43A441A1-4F17-11D7-A732-0003937E39E0@mac.com>
next in thread | previous in thread | raw e-mail | index | archive | help
David Leimbach writes: > True... I guess I didn't state my case clearly enough that I think IP > over firewire > is in itself a good thing for clusters. From my experience with the Apple IP over Firewire, it seems slow, and very high overhead. A dual 800MHz G4 host which can transmit at well over 1 Gb/sec (using ethernet-over-Myrinet) maxes its CPU out at 200Mb/sec, or less with IP over Firewire. (I'd report GigE numbers, but I don't have a GigE switch or a decent enough cable to get a host-host link at 1Gb/s). Its possible that the Apple code just sucked, I don't know. It used a 1500 byte mtu, for example. I'd have thought you'd be able to have much large mtus w/firewire. If the Apple code is typical, then I think that unless you've got some sort of alternate zero-copy protocol running over raw firewire, you'd be better off running GigE, or even multiple 100Mb links. Drew To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?15974.6361.471584.917266>