[e2e] Re: [e2e], [Tsvwg] Really End-to-end or CRC vs everything else?

[Jonathan Stone]

In message <200106081723.f58HN3t24578 at codex.cis.upenn.edu>,
Michael B Greenwald writes:

>   Thu, 7 Jun 2001 17:41:36 -0700
>   "Douglas Otis" <dotis at sanlight.net>
>
>   Michael,
>   
>   The stuck bit tests for the more elaborate two Fletcher-16 sums indicated
>   that errors were undetected 1.3% of the time using disk files as test cases
>   if a stuck memory bus bit affected only half of the packet.  
>
>Could you send me a pointer to the paper that reports these measurements
>and explains your experiments?  1.3% sounds high --- especially if you are
>using a 32bit Fletcher checksum (a 16bit sum and a 16bit sum of products).
>I'd like to understand this result.

Doug posted a URL to his simulation code to tsvwg.  (Be sure to get
the moe recent version with bug-fixes for the checksum computaiton).


Doug's error model is to simulate stuck-bit errors in each of the 32
positions in a 32-bit word: the `stuck' bits forces all bit positions
in that packet to the `stuck' value.  (the simulation does
stuck-at-one and stuck-at-zero).

Doug was using mod-2^n checksums ( two 16-bit counters computed
mod-65536).  IEN-45 has a discussion explaining why a mod 2^n sum
gives weaker coverage to the high-order bit than does a mod ((2^n)-1)
sum.  It would be interesting to look at how often it's the high-order
bit for which stuck-bit errors go undetected.