Re: pgp.demonserver.de - Sync Partners

Yaron Minsky <[email protected]> Fri, 11 Jun 2004 22:43:28 -0400
Newsgroups gmane.comp.encryption.pgp.keyserver-folk
Message-ID <[email protected]>
Do you mean the wire protocol?  The current wire protocol is basically
only described in the SKS source.  The set reconciliation algorithms
themselves are described in some papers I wrote, which are linked to
from the sks web page.

http://www.nongnu.org/sks/

Unfortunately, the bar is a bit high for building a server capable of
taking part in SKS-style gossip.  You will need:

1) An implementation of arithmetic over Z_p for a large prime p
2) An implementation of polynomial arithmetic over Z_p
3) A linear equation solver over Z_p.

There are some other things that go into it.  There's a data structure
called a prefix-tree which is about 70-100 Megs, and so you probably
don't want all of that in memory.  To that end, I have it stored
partially in memory and partially on disk, and so you have to build
that too.

All told, the synchronization component of SKS is fairly non-trivial,
and really was the bulk of the implementation effort.  After that,
parsing PGP packets was a breeze....

I guess the summary is this:  if someone is really interested in
putting in the necessary work to add sks-style synchronization to
their keyserver, I'd be up for standardizing the wire protocol in a
way that would allow them to interoperate.  But the bar is a bit high,
and I've been somewhat doubtful that there's anyone who thinks it
worth the effort at present.  Personally, I think it would be easier
to learn ocaml and extend SKS.

y

On Fri, 11 Jun 2004 22:16:56 -0400, V. Alex Brennen <[email protected]> wrote:
> 
> Yaron Minsky wrote:
> 
> > That feeling can be alleviated by switching to SKS....
> 
> Where is the gossip protocol defined?  Is the SKS
> source code the only place?
> 
> 
>     - VAB
>