Exact and heuristic MapReduce scheduling algorithms for cloud federation

Geo-distributed big-data processing has recently received much attention since it ensures large-scale and geographically distributed data processing, using Hadoop or Spark, in an efficient, fault-tolerant and reliable manner. The objective of this work is to propose a new geo-distributed MapReduce-based framework and algorithm for federated cloud platforms. A distributed heuristic algorithm, called FDMR (Federated Distributed MapReduce), that takes advantage of data locality, inter-cloud data transfer and high availability of capacities offered by the federation is proposed. The aim of FDMR is to reduce job cost while respecting deadline constraint. The goal of this paper is also to propose an exact MapReduce scheduling model to serve as a baseline for benchmarking and to compare and discuss the heuristic algorithm results. The performance evaluation proves that the proposed algorithm FDMR can improve resource utilization of the cloud federation and consequently reduce cost and job response time while satisfying the deadline constraint.