[jdf]

Lamport's description of his failures in introducing the Paxos algorithm is a highly amusing story:

http://research.microsoft.com/en-us/um/people/lamport/pubs/p...

For those that don't know, Paxos is one of the most important algorithms in distributed systems, so it's amazing to see that it wasn't even published for 8 years due to the author's ... odd structuring of the problem.

[chubot]

Whenever you have a good idea, you have to bash people over the head with it. I guess he tried that with the Indiana Jones thing and it failed. He says Butler Lampson was one of the few people who noticed its significance.

But the problem is that when you have a BAD idea you also may find yourself bashing people over the head with it :)

I wonder if he will win the Turing Award for Paxos. Awhile ago I thought it would be deserved, but I also feel like the full state machine is a bit awkward and heavy-handed for a lot of distributed systems problems (especially distributed systems over WAN, which I think is more interesting these days). I like the Bloom/CRDT work. And Raft is a simpler algorithm when you need strong consensus.

[mjb]

The core of Raft appears isomorphic to Paxos, with the addition of clearly explained and well-specified details for extending it into a working consensus system. It's cool and useful work, and is likely to be influential in the long run, but doesn't reduce the significance of Paxos at all. Similarly, CRDTs are a useful and interesting area of research, very applicable to many real-world problems, but they aren't the same problems Paxos solves.

You also can't ignore his other work in logical clocks, the bakery algorithm, the Chandy-Lamport algorithm, and TLA+.

[pkj]

A really good intro to Raft --> http://www.youtube.com/watch?v=YbZ3zDzDnrw

Very useful if you are into distributed systems and would like to see a practical implementation with clear examples.

[chrismonsanto]

Yeah, it's an interesting story. My take on it: everyone in academia has their own pet problem. I think a lot of people read papers in their fields with an eye to how the paper will help them achieve their own goals. When the problem is so abstracted from what you care about, it's difficult to stay motivated & absorb the material. I imagine a lot of people put up with that paper solely because of the reputation of the author.

I also think Lamport overstates his success with formulating the Byzantine generals problem in terms of Byzantine generals. I know the problem as the Byzantine generals problem, but think of it in terms of computers, not generals. It's just easier to understand the problem that way, the generals stuff just gets in the way. iirc my distributed systems textbook (Lynch) did not actually formulate the problem in terms of generals, but I could be misremembering.