From owner-freebsd-hardware Thu Jul 11 11:25:57 1996 Return-Path: owner-hardware Received: (from root@localhost) by freefall.freebsd.org (8.7.5/8.7.3) id LAA23820 for hardware-outgoing; Thu, 11 Jul 1996 11:25:57 -0700 (PDT) Received: from jparnas.cybercom.net (jparnas.cybercom.net [206.28.135.58]) by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id LAA23807 for ; Thu, 11 Jul 1996 11:25:45 -0700 (PDT) Received: from localhost.cybercom.net (localhost.cybercom.net [127.0.0.1]) by jparnas.cybercom.net (8.6.10/8.6.10) with SMTP id OAA03222; Thu, 11 Jul 1996 14:24:30 -0400 Message-Id: <199607111824.OAA03222@jparnas.cybercom.net> X-Authentication-Warning: jparnas.cybercom.net: Host localhost.cybercom.net didn't use HELO protocol To: Henry Spencer cc: hardware@freebsd.org, bsdi-users@bsdi.com X-External-Networks: yes Subject: Re: cable vs. ISDN In-reply-to: Your message of Thu, 11 Jul 1996 10:01:11 EDT. Date: Thu, 11 Jul 1996 14:24:28 -0400 From: "Jacob M. Parnas" Sender: owner-hardware@freebsd.org X-Loop: FreeBSD.org Precedence: bulk In message you write: >> >the hardware used for the Rogers prototype talked to the computers by >> >Ethernet. >> >> As pointed out earlier, isn't ethernet tcp/ip based or some other network >> protocol based... > >The question is phrased poorly, and is ambiguous, so I'll answer both >interpretations. :-) > >Is Ethernet tied to a specific protocol, like TCP/IP? No. Ethernet just >gets a packet from point A to point B, accompanied by a checksum (well, >CRC) and a type indicator. Any other structure is imposed by software. I understand. It can run DECNET a Xerox protocol, etc. I understand the checksum too. Thanks. >Do you need to use a non-trivial protocol of some kind to make use of >Ethernet? In principle, no, but in practice, yes. However, this is not >necessarily a bad thing, because talking to network or a complex device >invariably involves a protocol *anyway*... and better you should use a >well-designed one that your software already supports. The alternative is >not to do without a protocol, but to use some kludged-up mess invented by >the hardware vendor, typically undocumented and buggy. (I've written >device drivers.) I'd much rather have the hardware supplier use a standard >protocol that I have debugging tools for. > > Henry Spencer > henry@zoo.toronto.edu I agree that ethernet has a lot fewer problems than serial line connections. But, I think that's due to poor design and shortsightedness of the designers of most serial stuff. Ethernet was well designed. I don't see why other serial output couldn't be well designed too. For instance, the new TI UART card. I understand it can handle close to a megabit/second sustained input. That's a true improvement. But, many devices don't need a huge input speed. For instance keyboards, printers, terminals, etc don't really have serial problems due to their low inputs. Yes, a lot of serial input have been screwed up, and ethernet would have probably been better from the start than what happened. It just seems like a bit of a, kludge to me, for serial input, twice convert everything to a high speed networking protocol, each way, and pay for it on devices that don't need it, to avoid developing a really good long term protocol for serial devices seems complicated. The cost factor is pretty important too. If there's $100/computer of ethernet serial stuff on it (for all the serial devices at both ends), that would be pretty significant. With all the computers put out per year, this is the pretty significant. My question is why was it decided to use such tiny FIFO's? They went from 1-2 bytes to 16 per port, when the speed isn't so vital and with larger FIFO's, it seems like there would be much fewer interrupts, and writing drivers for it would be much easier. It can't be that expensive should it? 16 bytes isn't much. To have to empty it is a huge load on the bus for moving fairly little data. What about some DMA connection to the modems? Just a question. I don't know the answer. Thanks, Jacob