[Dev] P2P Architecture

[John Anderson]

Pastry and IRIS look interesting, I'll check it out. Your absolute right 
about Jaber and the server requirement. Our hope was that there would be 
enough Jaber servers around that most people wouldn't need to go to the 
trouble to set up and administer one. Jaber would provide presence and 
allow us to pass small messages between peers, large transfers would 
happen out of band directly between peers.

John Anderson


Charles Reis wrote:

> Hello--
>   I just found the project page for "Chandler" from the recent 
> Slashdot story, and I'm quite intrigued.  It sounds like an 
> interesting project, and I'm happy to see new directions being taken 
> in this area.
>
>   After browsing through the Feature Summary and Technology overview, 
> I got the impression that you intend to use Jabber to provide 
> peer-to-peer-like services, such as browsing others' data and 
> subscribing to contact information.  I think these could be very 
> powerful features, but Jabber appears to be based more on a 
> client-server model than true P2P, as noted on their own Technology 
> Overview page.
>
>   Have you considered looking at any of the distributed hashtable 
> technologies for a stronger P2P architecture?  Systems such as Pastry 
> (http://www.cs.rice.edu/CS/Systems/Pastry/) at Rice University or 
> Chord at MIT seem very strong for this sort of application, and have 
> recently been getting increased attention (eg. the NSF grant for IRIS, 
> http://iris.lcs.mit.edu).  In particular, these systems provide a true 
> P2P model with very good scalability and performance properties. 
> Although their implementations are not as widely used yet as Jabber, 
> it could be a very good direction for your "Interpersonal Information 
> Manager".
>
>   I'd be interested to know your thoughts on this, or if I'm mistaken 
> about Jabber.  Take a look at the IRIS page mentioned above for more 
> high level information and links to related papers.
>
> Hope it's useful,
> Charles Reis
>