I would find that very interesting as well. MR graph processing is fine for algorithms that scale in number of MR iterations as the diameter of the graph if they are applied to small diameter very large graphs (such as the small world graphs we all see). Various tricks like keeping first and second order links and duplicating nodes can cut the constants, but not change the basic asymptotic costs.
I agree that it isn't clear whether pregel is a layer over map-reduce or an entirely new paradigm. It would be interesting to know that. It would also be interesting to hear how people are attacking the problem of graph algorithms on large scale clusters, even in advance of getting results. On Thu, Jun 25, 2009 at 10:39 AM, Patterson, Josh <[email protected]>wrote: > I'm a little lost here; Is this a replacement for M/R or is it some new > code that sits ontop of M/R that runs an iteration over some sort of > graph's vertexes? My quick scan of Google's article didn't seem to yeild > a distinction. Either way, I'd say for our data that a graph processing > lib for M/R would be interesting. >
