Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Path: utzoo!linus!vaxine!wjh12!genrad!decvax!cca!ima!haddock!johnl From: johnl@haddock.UUCP Newsgroups: net.lan Subject: Re: Looking for comments on 3COM - (nf) Message-ID: <145@haddock.UUCP> Date: Mon, 2-Apr-84 23:38:27 EST Article-I.D.: haddock.145 Posted: Mon Apr 2 23:38:27 1984 Date-Received: Wed, 4-Apr-84 07:57:18 EST Lines: 33 #R:utcsrgv:-365500:haddock:19000001:000:1720 haddock!johnl Apr 2 15:07:00 1984 We currently have an Ethernet on which there's a Vax 780 with an Interlan interface and six IBM PCs with 3COM cards. We use a locally hacked up protocol to do file transfer, mail, remote printing, etc. (It gateways transparently to uucp, by the way, so that usenet sites albacore, anemone, chub, alewife, eel, and wahoo are actually these PCs on Ethernet. But I digress.) I'm reasonably happy with the 3COM cards. They do work, and cramming the entire interface and the transceiver onto one card was no mean feat. We use their thin Ethernet cable to hook up all of the PCs, spliced to a fairly short piece of regular fat cable for the Vax hookup. This all works fine except when somebody steps on the thin cable, since it's not very rugged. We should have it all off the floor and into the walls and ceiling, at which time that problem should go away. The documentation is pretty good and we were able to write our own drivers without undue trouble. We haven't tried 3COM's software, since we use PC/IX rather than PC-DOS. The biggest complaint I have to make about the 3COM cards is that the throughput is limited by the single buffer on the card. To receive a packet, you enable the receiver, and when the packet arrives it's stored in the on-card buffer. The card interrupts the host, which commands the card to DMA the data into the host memory. After that the host resets the interface card, and it's prepared to receive again. Should a packet arrive during the fairly long interval between the end of the previous packet and the card's reenabling, it's jammed and lost. I realize that the single buffer is due to real estate limitations on the board, but it's still a pain. John Levine, ima!johnl