Re: [Sks-devel] Recon Details

[Thomas Spycher]

Oukay i see, this is fairly complex… But thanks for pointing to the hockeypuck 
Project! Is this project under heavy development or is it already running in 
production?

Thanks
Tom

On 28 Oct 2013, at 22:35, Phil Pennock <address@hidden> wrote:

> On 2013-10-28 at 15:59 +0100, Thomas Spycher wrote:
>> I’m the guy asking for Gossiping peers and the message of John Clizbe has 
>> bring me to an other mystery in the GPG World.
>> I think i do understand the hkp protocol very well. I do understand the idea 
>> behind gossiping as well. The theory about recon i think i do understand, 
>> but what i do not understand is how the reconciliation is actually done! I 
>> was digging trough the sks sourcecode and due to i’m not natively ocaml 
>> speaking, i’m having my problems figuring out how it works.
>> 
>> Can someone explain an recon session? Or giving examples based on a telnet 
>> session?!
> 
> There's some academic papers by Yaron Minsky on the set reconciliation
> algorithm he devised, linked to at:
> 
>  https://bitbucket.org/skskeyserver/sks-keyserver/wiki/Home
> 
> Port 11370 speaks a protocol which just identifies what needs to be
> transferred.  Each side then performs HKP GET requests against
> peering_port+1 (ie, 11371) to retrieve the keys, using normal key
> retrieval.  Thus you need both ports to be open to peers.
> 
> If recollection serves (it might not), the recon protocol is binary,
> bi-directional and custom.  Casey Marshall examined it closely so that
> he could reimplement it in Go (Golang), for use in Hockeypuck:
> 
>  https://launchpad.net/hockeypuck
> 
> and I'd start looking at his conflux library:
> 
>  https://github.com/cmars/conflux
> 
> -Phil

signature.asc
Description: Message signed with OpenPGP using GPGMail