[p2p-hackers] A Question about user awareness in p2p

[Anoop Mavilaveettil]

Thanks Mike and Greg,
 Yes I had known that I would have to(and was planning to) use DHT for
peer discovery. But the issue I had was about a peer getting a
notification when another specific peer comes online.
As Greg said, this would have involved polling DHT for this info, and
if the buddy system is very well connected(I  mean a majority of the
peers have buddies), then this polling could become a bottleneck.
See the issue of, say, figuring out if "my" buddies are connected to
the p2p net when "I" come online is what that could be solved by DHT.
But I was and is looking a solution for is the issue of "I'm already
connected and now how would "I" know when one of my buddies come
online?".

 Greg's idea of keeping the buddy info, also in the DHT would work
out(I think) but at a larger storage cost or so would a distributed
buddy database.

Is there any other options?
-Sreedhar


On Tue, 22 Mar 2005 09:59:32 -0500 (EST), Michael J. Freedman
<mfreed at cs.nyu.edu> wrote:
> On Tue, 22 Mar 2005, Anoop Mavilaveettil wrote:
> 
> > Date: Tue, 22 Mar 2005 02:27:10 -0800
> > From: Anoop Mavilaveettil <anoopsmv at gmail.com>
> > To: p2p-hackers at zgp.org
> > Subject: [p2p-hackers] A Question about user awareness in p2p
> >
> > Hello,
> > First of all, let me thank you for all the sharing of knowledge that
> > is going on in here.I must say that, I have learned a lot, ever since
> > I started frequenting this list.
> 
> ...
> 
> > b)Using an overlay of super peers/ meta trackers for user tracking and
> > then them firing all required messages to interested peers, whenever a
> > user connects to the p2p net.
> > As I said earlier, both methods aren't that great in terms of
> > efficiency and/or robustness. Is there some other way or any other
> > techniques that could be used to let a peer instantly know about the
> > connection status of his buddy?
> 
> Hi Sreedhar,
> 
> This type of "discovering needles in the haystack" problem is exactly
> where structured routing overlays (or, perhaps slightly inaccurately,
> distributed hash tables) perform superiorly to unstructured routing
> meshes (like Gnutella et al.).
> 
> Using such a distributed indexing layer, when a peer joins the system, it
> announces itself under its system handle; when another peer joins, it
> searches for its buddies under their handles in this shared architecture.
> 
> To ensure freshness, these announcements should have a TTL; still, they
> need not be all that short, as information learned from the "DHT" can of
> course be verified by peers (i.e., ping them to check if still alive).
> 
> For more information on structured overlays, I suggest looking at the
> 
>   MIT Chord -- http://pdos.lcs.mit.edu/chord/
>   Rice/MSR Pastry -- http://freepastry.rice.edu/
> 
> For an example of using these type of interfaces for node discovery under
> such weak consistency requirements, check out our use in:
> 
>   CoralCDN -- http://www.coralcdn.org/
> 
> If you just want an "working" DHT to play with, you can use the deployed
> OpenDHT service as such a rendezvous point.
> 
>   OpenDHT -- http://www.opendht.org/
> 
> Good luck,
> --mike
> 
> -----
> "Not all those who wander are lost"             www.michaelfreedman.org
> "You make it fun; we'll make it run"                   www.coralcdn.org
>