From owner-freebsd-hackers Thu Feb 3 11:26:32 2000 Delivered-To: freebsd-hackers@freebsd.org Received: from panzer.kdm.org (panzer.kdm.org [216.160.178.169]) by builder.freebsd.org (Postfix) with ESMTP id 8072C400D for ; Thu, 3 Feb 2000 11:26:29 -0800 (PST) Received: (from ken@localhost) by panzer.kdm.org (8.9.3/8.9.1) id MAA53763; Thu, 3 Feb 2000 12:21:14 -0700 (MST) (envelope-from ken) Date: Thu, 3 Feb 2000 12:21:14 -0700 From: "Kenneth D. Merry" To: Matthew Jacob Cc: Thomas Stromberg , freebsd-hackers@FreeBSD.ORG Subject: Re: Suggestions for Gigabit cards for -CURRENT Message-ID: <20000203122114.A53673@panzer.kdm.org> References: <20000202113259.A43505@panzer.kdm.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Mailer: Mutt 1.0i In-Reply-To: ; from mjacob@feral.com on Thu, Feb 03, 2000 at 10:50:32AM -0800 Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG On Thu, Feb 03, 2000 at 10:50:32 -0800, Matthew Jacob wrote: > > > The FreeBSD driver (written by Matt Jacob) is based on the Linux driver, > > which Intel wrote, and he hasn't yet managed to get decent throughput > > through the cards. (Maybe Matt will comment.) They also only have 64K of > > memory on board, which is insufficient for a heavily loaded server, IMO. > > That's not memory- that's FIFO- there are two of them, I believe, one for > receive, the other for xmit. You can devote 64k to ring descriptors for > receive- that's 4096 descriptors- each able to manage a 2k buffer. And > you can have two receive rings. You can have the same size for xmit. > > So, the receive performance bottleneck for this chip/board will be in how good > your PCI implementation is at first followed by how low an amount of > interrupt latency for reinstruct. If your PCI implementation can keep up with > Gigabit speeds, you're fine. If not, I'm not sure that 512K or 1MB buffering > buys you much. I think the memory would come in handy on a heavily loaded system, since you would gain a little extra time in case you were a little late servicing interrupts. i.e. it would smooth out the bumps a little bit. If your PCI implementation won't keep up with gigabit speeds, you'll just go slower. :) Most newer systems (e.g. 440BX) shouldn't have any trouble doing a reasonable amount of speed over gigabit ethernet, though. Ken -- Kenneth Merry ken@kdm.org To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message