Adaptive Optimal Control of MapReduce Performance, Availability and Costs

MapReduce is a popular programming model for distributed data processing and Big Data applications running on clouds. Extensive research has been conducted either to improve the dependability or to increase performance of MapReduce, ranging from adaptive and on-demand fault-tolerance solutions, adaptive task scheduling techniques to optimized job execution mechanisms. This paper investigates an optimization-based solution to control MapReduce systems in order to provide guarantees in terms of both performance and availability while reducing utilization costs. We follow a control theoretical approach for MapReduce cluster scaling and admission control. Moreover, we aim to be robust to changes in MapReduce and in it’s environment by adapting the controller online to those changes. This paper highlights the major challenges of combining system adaptation and optimal control to take the best of both approaches.