Wadler's Blog

The MapReduce paper begins its discussion of related work as follows:

Many systems have provided restricted programming
models and used the restrictions to parallelize the computation
automatically. For example, an associative function
can be computed over all prefixes of an N element
array in log N time on N processors using parallel prefix
computations.

Has anyone implemented parallel prefix scan as an extension to MapReduce or a similar framework such as Hadoop, and did it prove useful?

Labels: Computing, Distributed Computing, Programming Languages