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

[David P. Reed]

At 04:48 PM 6/11/01 -0400, Craig Partridge wrote:
>I think you've missed the point.  In a prior note, you suggested a line of
>thinking of assume an adversary.  Implicitly, that's an error model.

You're right, it's a model.  But there are advantages in not forcing the 
model to be statistical.  Instead you can bound, say, the adversary's 
computational complexity - which is what you do when thinking about a 
cryptographic adversary, since inversion of one-way functions is blocked by 
computational complexity considerations.

When you are trying to think through what kinds of errors software and 
hardware design flaws might introduce, it helps to classify them by 
computational complexity - it takes resources to screw stuff up in a 
complicated way.

The computational complexity needed to screw up cryptographic hash is known 
to be high, whereas there are computationally simple functions that can 
screw up  simple checksums.