RE: iSCSI CRC considerations

[dafna@il.ibm.com]

Doug,

We explicitly said that either modulo (as originally in RFC1950) or some
test (as you showed) is necessary.
(In fact, there is even a better tradeoff between (1) calculating a maximal
number of additions that can be
made for sure before the modulo needs to be addressed at all, and (2)
taking the test for each input
char).

RCF1951 and RFC1952 describe the popular DEFLATE format used in Gzip and
PKZIP, to which Adler
contributed valuably. They do not mention the Adler-32 checksum.  The same
goes to RFC 1979 and RFC 2394.

We agree that CRC is slower in software implementation than Adler-32.
But, Adler can detect much less errors, as we pointed out,  than CRC.
There are too simple error patterns that are missed by Adler.

Dafna Sheinwald.