[Http-crcsync] General comments on crcsync document

Gervase Markham gerv at mozilla.org
Fri Jul 17 13:33:29 EDT 2009


On 17/07/09 08:53, Toby Collett wrote:
> Which indicates that with 40 hashes and a page about the size of
> slashdot we have a 1/1000 error rate, assuming completely random data. I
> am not sure how much impact the asymmetric nature of CRC has on this,
> are we talking 10% better or factors of 10?

Hmm. I'd say 1 in 10,000 would be fine, 1 in 1000 is borderline. What's 
the worst-case scenario - you have to issue another request, and 
download the whole page as you would have had to normally? So 1 in 999 
is sped-up, and 1 in 1000 is slowed down by 1 RTT and a bit of 
processing time?

How hard is it to change this in an existing implementation? Do a lot of 
details depend on it, or is it fairly pluggable? Given the uncertainty 
you give above, maybe we want to just do some endurance tests using 
non-random data.

Gerv


More information about the Http-crcsync mailing list