Path: utzoo!mnetor!uunet!husc6!think!ames!pasteur!ucbvax!VENERA.ISI.EDU!braden
From: braden@VENERA.ISI.EDU
Newsgroups: comp.protocols.tcp-ip
Subject: Re:  Checksums (was Re: Ping, checksum algorithm?)
Message-ID: <8803282346.AA01900@braden.isi.edu>
Date: 28 Mar 88 23:46:31 GMT
Sender: usenet@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 17

Several have pointed out that IEN45 on checksums is not online. It is true.
Jon Postel and I are getting that fixed, and as soon as it is fixed, I
will send a short note to this list.

IEN45, while interesting, does not contain the nifty fact that Phil Karn
pointed out: the checksum is invariant to arbitrary byte permutations.
I don't know where that is documented; it was pointed out to me in 1983
by Peter Higginson of University College London, and enabled me to speed
up the IBM MVS ACP checksum routine by about a factor of 2.  This is
probably one of the great underground tricks in TCP implementations
(or do they teach it to undergraduates these days??).

We should turn some of the recent messages on this topic into an RFC, or
include them as an Appendix to the Host Requirements RFC which is now
being drafted. Any volunteers to write about checksum implementations? 

Bob Braden