[tahoe-dev] Recombinant Hashes

[Zooko Wilcox-O'Hearn]

On Thursday, 2010-02-11, at 22:31 , David-Sarah Hopwood wrote:

> According to this ticket: <http://allmydata.org/trac/tahoe/ticket/ 
> 678>, the FEC code that Tahoe uses also has the property that the  
> content of some initial subset of shares does not depend on the  
> total number of shares. If that's correct, it's not clear to me why  
> it it's not also classified as a fountain code, although I've never  
> heard of Reed-Solomon codes referred to as fountain codes.

Perhaps because there is a fixed upper limit on how many unique  
shares we can generate. The erasure code library that Tahoe-LAFS uses  
-- zfec -- maxes out at 256 shares. The library from which zfec was  
derived also included a variant that maxed out at 65536 but encoding  
and decoding would be slower.

Regards,

Zooko