RE: Linux Kernel Summit

> Well, we already have MD5 (and SHA1) hashes built right into APR.
> HTTP also specifies the Content-MD5 header for on-the-wire checksums.
>
> At the moment, MD5 seems a reasonable way to do checksums, from the
> standpoint of the various components. Being 128 bits in length, it is also
> more resistent to false negatives than a 32 bit (or smaller!) CRC value.

Yes, I was talking about a 64 bit CRC... I have the (tiny!) code
here, but I think MD5 might be better, and the calculation shouldn't
have a huge impact on performance.
Received on Sat Oct 21 14:36:27 2006