OS4M: Achieving Global Load Balance of MapReduce Workload by Scheduling at the Operation Level

Abstract

The efficiency of MapReduce is closely related to its load balance. Existing works on MapReduce load balance focus on coarse-grained scheduling. This study concerns fine-grained scheduling on MapReduce operations, with each operation representing one invocation of the Map or Reduce function. By default, MapReduce adopts the hash-based method to schedule Reduce operations, which often leads to poor load balance. In addition, the copy phase of Reduce tasks overlaps with Map tasks, which significantly hinders the progress of Map tasks due to I/O contention. Moreover, the three phases of Reduce tasks run in sequence, while consuming different resources, thereby under-utilizing resources. To overcome these problems, we introduce a set of mechanisms named OS4M (Operation Scheduling for MapReduce) to improve MapReduce's performance. OS4M achieves load balance by collecting statistics of all Map operations, and calculates a globally optimal schedule to distribute Reduce operations. With OS4M, the copy phase of Reduce tasks no longer overlaps with Map tasks, and the three phases of Reduce tasks are pipelined based on their operation loads. OS4M has been transparently incorporated into MapReduce. Evaluations on standard benchmarks show that OS4M's job duration can be shortened by up to 42%, compared with a baseline of Hadoop.