Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Path: utzoo!mnetor!seismo!lll-lcc!ptsfa!ihnp4!ihlpl!jhh From: jhh@ihlpl.ATT.COM (Haller) Newsgroups: comp.dcom.modems Subject: Re: Error correction without notification is a road to disaster Message-ID: <1893@ihlpl.ATT.COM> Date: Mon, 9-Mar-87 10:35:34 EST Article-I.D.: ihlpl.1893 Posted: Mon Mar 9 10:35:34 1987 Date-Received: Tue, 10-Mar-87 06:26:37 EST References: <1264@steinmetz.steinmetz.UUCP> Organization: AT&T Bell Laboratories - Naperville, Illinois Lines: 31 Summary: CRC/FCS implemented by modem may cancel higher level error detection In article <1264@steinmetz.steinmetz.UUCP>, davidsen@steinmetz.steinmetz.UUCP (William E. Davidsen Jr) writes: > In article steckel@alliant.UUCP (Geoff Steckel) writes: > >A number of vendors have introduced "error correcting" modems > >recently. I have heard of none that notify the user that correction > >has taken place. Some manufacturers actively discourage users from > >using error correction protocols above their modems! None, of course, > >detail the manner in which they provide this service. From experience > >and reading in networking, this is a disaster waiting to happen. > > I agree with you premise, but I *think* you're working from a > false starting point. The error correcting modems *seem* to work > by sending a packet and doing a resend if it's munged in > transit. I get the impression that you are assuming error > correction ala Hamming or Fire codes. See the Febrary, 1987 Data Communications, page 62. It is possible that if the software uses the same polynomial that the modem does that no further detection will be available by the software. The case described in the magazine had a modem whose scrambler had the same polynomial as the BISYNC protocol above it. It was a noisy link, and large amounts of undetected errors were being passed. I would not trust CRC-CCITT for error detection of that was what the modem used for its error detection. Maybe it would work, and maybe it wouldn't. The point is, when there is a noisy line, you probably want to know about it, so as not to run into cases where the CRC algorithms don't work. An occasional bit error is fine, but an extremely noisy line should not be used, expecting the CRC to find all errors. John Haller ihnp4!ihlpl!jhh AT&T Bell Laboratories