A memo on some checksums

[julian_satran@il.ibm.com]

As several people from this list have asked us to look at various
alternatives to the TCP checksum and or cheaper alternatives for the CRC
digests - Dr. Dafna Sheinwald from the IBM Haifa Research Lab. took a  look
at 2 that gained some popularity lately and here is a preliminary report.

(See attached file: Dafna-Sheinwald-codes.wri)

As for alternative CRCs two classes of checks have to be performed on them
before we can really consider them:

   test of goodnes (there several well known rules (look at Andrew
   Tannenbaum - Computer Network, or Richard Wells - Applied Coding and
   Information Theory for Engineers or one of the books that appear in
   Dafna's list - e.g., the classical Peterson & Weldon)
   a computer scan to show how "distant" good generated codes are


As up to 32 bit we have several good polynomials and the "nesting" does not
reduce their effectiveness I am not going to look for something else but we
should be ready to accept any proposal that comes with the right
credentials.

On 64 bit we are going to spend some time as there is no agreed and tested
polynomial available  or at least I was not able to get to one. Quantum
advertises using one for the DLT tapes but I have no clue what it is.

If somebody on this list is aware of anything relevant in this area (even
patent protected - at least we should know what to avoid if we can -:)
please speak up.

Testing a new polynomial is not a trivial effort.

Julo

Word 6.0 Windows/Mac