From owner-freebsd-current Thu Sep 19 06:53:36 1996 Return-Path: owner-current Received: (from root@localhost) by freefall.freebsd.org (8.7.5/8.7.3) id GAA15859 for current-outgoing; Thu, 19 Sep 1996 06:53:36 -0700 (PDT) Received: from halloran-eldar.lcs.mit.edu (halloran-eldar.lcs.mit.edu [18.26.0.159]) by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id GAA15832 for ; Thu, 19 Sep 1996 06:53:33 -0700 (PDT) Received: by halloran-eldar.lcs.mit.edu; (5.65v3.2/1.1.8.2/19Aug95-0530PM) id AA08715; Thu, 19 Sep 1996 09:53:24 -0400 Date: Thu, 19 Sep 1996 09:53:24 -0400 From: Garrett Wollman Message-Id: <9609191353.AA08715@halloran-eldar.lcs.mit.edu> To: dg@root.com Cc: Michael Smith , danj@netcom.com (Dan Janowski), current@FreeBSD.org Subject: Re: Various drivers... In-Reply-To: <199609190325.UAA01124@root.com> References: <199609190206.LAA07511@genesis.atrad.adelaide.edu.au> <199609190325.UAA01124@root.com> Sender: owner-current@FreeBSD.org X-Loop: FreeBSD.org Precedence: bulk < said: > The Pro/100B actually performs better in some benchmarks for raw speed, and > the overhead of the fxp driver is significantly lower (10-20%). Well, hmmm. It is certainly true that the overhead is lower (I have graphs which demonstrate this). It's also true that it's a dog at sending small packets (more graphs). For every NIC, the amount of time to send a packet is determined: C = A + size*B + w(size) For large sizes, the B term dominates. For small sizes, the A term dominates. Based on my experience, the Intel NICs have a very low B factor but a rather hefty A term; the DEC NICs have a much higher B factor, but a much lower A term. The point where the two curves cross is about size==250 bytes. Whether this matters to you depends a great deal on your application; if you are running a Web server, it probably doesn't, but if you are trying to do videoconferencing, the DEC chip is much better. (Having said that, the socket interface is so slow that you are unlikely to notice the effect either way. Our long-term plans here involve rewriting the socket layer to eliminate much of this problem.) >From our results, I would surmise that the Intel chip has a more efficient PCI DMA engine than the DEC chip. -GAWollman -- Garrett A. Wollman | O Siem / We are all family / O Siem / We're all the same wollman@lcs.mit.edu | O Siem / The fires of freedom Opinions not those of| Dance in the burning flame MIT, LCS, ANA, or NSA| - Susan Aglukark and Chad Irschick