Newsgroups: comp.dcom.modems
Path: utzoo!henry
From: henry@utzoo.uucp (Henry Spencer)
Subject: Re: V.42 [was: Telebit in Byte's test, Ventel PEP, V.32]
Message-ID: <1988Aug26.174021.23655@utzoo.uucp>
Organization: U of Toronto Zoology
References: <1313@neoucom.UUCP> <5916@bigtex.uucp> <Aug.12.14.09.59.1988.16394@topaz.rutgers.edu> <6069@bigtex.uucp> <8808162000.AA04427@yorkville.csri.toronto.edu> <15303@shemp.CS.UCLA.EDU> <8808251940.AA00343@ellesmere.csri.toronto.edu>
Date: Fri, 26 Aug 88 17:40:21 GMT

In article <8808251940.AA00343@ellesmere.csri.toronto.edu> ray@csri.toronto.edu (Raymond Allen) writes:
>provision for the use of a 32-bit CRC (CCITT calls it
>the "Frame Check Sequence") was added to the V.42 specification.  This is nice
>from a technological standpoint since it will reduce the bit error rate to
>about 10e-239 (note: :-) ) but from an implementation standpoint it is a real
>b**ch because there is hardware in existence that will calculate the 16-bit
>CRC but the 32-bit one would have to be done in software and this has to
>be done for each character transmitted -- a rather large overhead.
>Seems that (as usual) standards committies have reduced awareness of real-
>world considerations.

Ho ho.  Not so.  The standards committees are more aware of real-world
considerations than you are, in this case.  The fact is, 16-bit CRCs *ARE
NOT ENOUGH* for some types of modern modems, and this problem will only get
worse.  Things like RF modems will often send a substantial number of bits
as a single transition, which means that one noise hit can foul up rather
a lot of bits.  16-bit CRCs fail with significant frequency in this sort
of environment; there was an uproar in the IBM mainframe world a while ago
when clear proof was produced that IBM's 16-bit CRC did not dependably
detect errors in bulk transmission over high-speed networks.

In this case, the standards committee is being farsighted, and is accepting
implementation problems today for the sake of reliable functioning tomorrow.
(Well, they may have *done* it because 32 bits sounded sexier, but the net
result is favorable regardless of real motives.)
-- 
Intel CPUs are not defective,  |     Henry Spencer at U of Toronto Zoology
they just act that way.        | uunet!attcan!utzoo!henry henry@zoo.toronto.edu
