[Mnet-devel] erasure code optimization project

[Lenny G Arbage]

--- Bryce Wilcox-O'Hearn <zooko at zooko.com> wrote:

> I forgot to mention one major drawback of LDPCs
> (already discussed in 
> ancient mnet-devel archives): they are not as
> attack-resistant.  In the 
> example you ran, you generated 255 shares, and when
> trying random 
> shares it took 110 of them to regenerate the file. 
> If instead of 
> trying random shares you allowed an enemy to delete
> some shares of his 
> choosing then you might have had to try even more
> than that.  In fact, 
> I'm not even sure that recovery of the file would
> always be possible.

  This is true.  You could compensate by
over-encoding.  For example, by using n=50 m=204. 
This results in the same number of shares, with /very/
high probability that /any/ k shares can reconstruct
the file, where k<84 (and if you are lucky, with k ~=
n*1.05).

  Of course, the tradeoff is that the amount of data
you have to publish is even greater now: for the same
example file, each share is 2352B instead of 1572B
(total 583K vs 389K), which only more strongly
suggests that you'd only do this if network/disk
resources are less of a bottleneck/scarcity than CPU
resources.
 
  -Lenny


		
__________________________________
Do you Yahoo!?
New and Improved Yahoo! Mail - Send 10MB messages!
http://promotions.yahoo.com/new_mail