>I was actually considering (perhaps) using Paxos in combination with a >DHT of some sort. I don't see how a DHT can provide an election or >lease service, or even provide mutable state, so I don't think it >would suffice for the system I have in mind. However I still have a >lot of reading to do so I'm probably completely wrong on all points. > >Any rate here is a link to paxos. >http://klinewoods.com/papers/p2paxos.pdf > >Thanks, I started reading that paper last night but hadn't realized it >included performance results. I was somewhat discouraged after seeing >"P2's lack of concurrency control and well defined execution semantics >provided a source difficult bugs and race conditions, and we remain >uncertain as to wether or not our implementation is robust" in the >abstract and moved on.
Glad you had a peek at the Paxos-in-P2 work. The P2 Declarative Networking infrastructure has evolved a fair bit since then, and some of the issues raised in that Paxos paper are being dealt with now. You might want to have another look. Website is http://p2.cs.berkeley.edu, there's discussion via the Yahoo group "p2devel". Interestingly, P2 was designed to make it easy to write overlay networks like DHTs, so we were pleasantly surprised to see how it helped these guys deliver a working Paxos implementation in 50-odd lines at that time. And it did push us to consider some issues we had not been focusing on prior to that, which are helping the system become more versatile. We'd be happy to help interested folks work through their protocol implementations over P2. It really does make these implementations much more pocket-sized -- e.g. Chord in ~50 lines, Paxos in ~50 lines, etc. It's research-quality code and under constant revision (we're officially still pre-beta), but a bunch of folks are using it successfully outside our group. Best way to get in touch is [EMAIL PROTECTED] Joe Hellerstein _______________________________________________ p2p-hackers mailing list [email protected] http://lists.zooko.com/mailman/listinfo/p2p-hackers
