Speed scaling in fork-join system: a comparative study

Fork-Join paradigm is model of parallel computing. This family of al- gorithms divide the input into smaller pieces that are served in parallel and then joined into a unique output. MapReduce is an example of Fork- Join algorithm, it maps and sorts the input in smaller parts and then re- duces them into the final result. With the classical implementations of MapReduce we don’t have any control on the speeds used by the nodes during the Map phase but studies [1] point out how controlling such speeds thanks to processor frequency scaling can reduce the power-consumption and resources of the system while maintaining a similar throughput. In this document we study the MapReduce paradigm and analyze how vari- ous methods of Rate control can optimize a physical implementation.