From owner-freebsd-hackers Mon Mar 30 09:48:53 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id JAA10662 for freebsd-hackers-outgoing; Mon, 30 Mar 1998 09:48:53 -0800 (PST) (envelope-from owner-freebsd-hackers@FreeBSD.ORG) Received: from terra.Sarnoff.COM (terra.sarnoff.com [130.33.11.203]) by hub.freebsd.org (8.8.8/8.8.8) with SMTP id JAA10543 for ; Mon, 30 Mar 1998 09:48:43 -0800 (PST) (envelope-from rminnich@Sarnoff.COM) Received: (from rminnich@localhost) by terra.Sarnoff.COM (8.6.12/8.6.12) id MAA04630; Mon, 30 Mar 1998 12:47:24 -0500 Date: Mon, 30 Mar 1998 12:47:24 -0500 (EST) From: "Ron G. Minnich" X-Sender: rminnich@terra To: Chris Csanady cc: freebsd-hackers@FreeBSD.ORG Subject: Re: Virtual Interface Architecture In-Reply-To: <199803301722.LAA03539@friley585.res.iastate.edu> 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 well, there is a via emulation written for linux by bozeman at LBL. You might see if he would let us have a look. app-app latency is 20 microseconds or so: good but not earth-shattering, UNET is already about there anyways. The via spec defines an API that if you conform to you can pretty much put anything underneath. For real low latency you need a hardware via interface. myrinet will be releasing one of these this summer. For whatever network you use, there have to be tags in the packets that allow hardware demux directly to an application or at least a memory area. As it happens, HIPPI is not that great this way: you have to put host memory addresses to tell the destination interface where to put the data. [IMHO, virtual memory addresses in packets are a really bad way to support hardware demux]. Believe it or not, one thing that is good is ATM, which we've shown in practice here by building a via-like interface for ATM (it's not via-compatible because we designed it about three years before intel, microsoft et. al. thought of via). If you think about it the ATM VCs can in fact define an application as an endpoint, not a host, unlike Ethernet. Of course, since the most common use of atm interfaces is to emulate ethernet (i.e. LANE), this application-endpoint use is not common :=) Unfortunately the ongoing ATM disaster (quick: what's the difference between the Titanic and ATM? your favorite punchline here. [[[ forget it. it's too easy to make the titanic look good compared to ATM]]]) will probably mean nobody ever builds via's based on ATM. But Myrinet will build good expensive VIAs; and there are other good expensive cards coming. It sure would be nice if we got to some good cheap cards at some point too: say, a VIA for 100bt that costs about 100 bucks. ron Ron Minnich |Java: an operating-system-independent, rminnich@sarnoff.com |architecture-independent programming language (609)-734-3120 |for Windows/95 and Windows/NT on the Pentium ftp://ftp.sarnoff.com/pub/mnfs/www/docs/cluster.html To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message