It's also worth pointing out GraphLab, an abstraction for parallel programming over graphs that is higher level than MPI and Pthreads, but lower level and more expressive than MapReduce.
Many machine learning algorithms cannot be expressed in MapReduce without sacrificing scalability, but can be expressed in GraphLab and maintain scalability.
See this thread on MetaOptimize, including commentary from the GraphLab authors:
3 comments
[ 3.2 ms ] story [ 15.4 ms ] threadMany machine learning algorithms cannot be expressed in MapReduce without sacrificing scalability, but can be expressed in GraphLab and maintain scalability.
See this thread on MetaOptimize, including commentary from the GraphLab authors:
http://metaoptimize.com/qa/questions/285/when-should-you-pre...
I would be interested in a comparison between Pregel and GraphLab.