Skip to content

IterativeReduce

jpatanooga edited this page Oct 8, 2012 · 2 revisions

Overview

  • A simple abstraction in the same concept as MapReduce but for parallel iterative algorithms
  • similar in nature to the AllReduce primitive or some of Spark's concepts
  • Involves writing a ComputableMaster and ComputableWorker
  • Work is done in the worker and then aggregated every so often in the master

ComputableWorker

  • does the work in parallel on batches of data

ComputableMaster

  • Aggregates partial results from workers, computes a bit, and then returns an aggregation to the workers
Clone this wiki locally