On Tue, Jun 29, 2010 at 12:04:12AM +0200, Valerio Schiavoni wrote: > > - Given the objectives of the project, are there some problems and not so > > well-known > > caveats I should be aware of when using a Chord-like protocol ? I'm aware > > of the > > general litterature on Chord (optimal routing, rotating virtual nodes, > > deBruijn graphs, > > ...). Should I look into something in particular ? > > Do you plan to use the DHT also as a storage facility, or just its > key-routing layer?
The DHT is used for routing purposes, and light storage (the so-called 'search groups', or groups of peers having issued similar queries). Note that since the queries are hashes, they never circulate in clear on the network or on the storage host. This is in sharp contrast to existing search engines. > If you need storing values persistently and with some fault-tolerance > mechanism built-in, you might want to look carefully at how the > replication system works. > The Chord paper is far from being detailed from this point of view, > and in a real world implementation it's not trivial at all to get it > right. In particular, it's not trivial to tune the protocol parameters > to ideal values according to your context. Interesting. I have devised a replication protocol and I am in the process of implementing it. You are correct that the Chord paper leaves the replication to the application higher layers and thus does not truely discuss it. Though I would be interested in any reference to works in this area (replication over a DHT), and especially wrt. the Chord protocol. Very roughly, my current design is pretty simple, with two special calls, one to replicate forward, the other backward. This is in response to a change of predecessor or successor (from the pov of a single node). The tuning of parameters is a real concern. I would love to hear feedback from anyone with experience in this domain. Most especially, Seeks lets its users install the software and tweak its parameters. For now, all parameters are opened to configuration. I do believe the DHT can function with heterogeneous parameters but this makes its performances less predictable. One solution would be to hardcode some of the parameters, thus hiding them from the eyes of the common user. Sincerely, Em. _______________________________________________ p2p-hackers mailing list [email protected] http://lists.zooko.com/mailman/listinfo/p2p-hackers
