[bittorrent] New P2P protocols

Nick Johnson arachnid at notdot.net
Sun Mar 27 21:05:42 EST 2005

On 28/03/2005, at 1:01 PM, mgp at ucla.edu wrote:

> Just a warning about Online Codes: they _are_ patented by Digital
> Fountain. In fact, the Online Codes paper cited [1] actually fell under
> the scope of DF's patents, unbeknownst to the authors. As a result, 
> after
> (or perhaps just before) the authors started the company Rateless [2],
> they developed new "rateless codes, based on decoders that don't use
> chain-reaction, message passing or belief-propagation techniques." If
> you've read the Online Codes paper, you'll find this quite interesting,
> since this is how Online Codes and practically every other similar code
> based on sparse graphs works. Unfortunately, I haven't seen any papers 
> on
> their new technique -- perhaps they are keeping quiet until they 
> receive
> patents on it.

Yup, I'm aware of that unfortunate circumstance. As I stated, though, 
for myself it doesn't apply, as I live in New Zealand. I still think 
it's worth implementing, even if it means that only non-us users can 
use that portion of it.

> Before I found out that Online Codes were wrapped up in an IP 
> quagmire, I
> implemented the encoder and decoder from the algorithm in the paper,
> including niceties such as a "pretty" interface (allowing it to be used
> as a library in any application) and a disk cache for improved
> performance. Unfortunately, only after I finished it all did I realize 
> I
> would get sued if it ever saw the light of day under the GPL.

You could publish it under a more permissive license, with the proviso 
that it not be used in the US without a patent license.

> Now, I haven't read the paper on homomorphic hashing for rateless 
> codes,
> but I was wondering if after applying it to a "vanilla" rateless code,
> would the new code fall within the scope of Digital Fountain's patents?
> Unfortunately(or, depending on your point of view, it was simply good
> fortune), IANAL -- but if not, it might be something worth pursuing. If
> needed, you can build on what I've written. (Right now it is in Java, 
> but
> it should be easily translated into any OO-language.)

Having skimmed one of the patents, I believe it would still be covered: 
The only change required to implement homomorphic hashing is to change 
the xor for block combination to addition modulo a large prime, and the 
patent is very wide in scope - if I remember correctly, it applies 
regardless of how you amalgamate blocks.

More information about the BitTorrent mailing list