Scheduling in MapReduce Clusters

摘要

MapReduce is a framework proposed by Google for processing huge amounts of data in a distributed environment. The simplicity of the programming model and the fault-tolerance feature of the framework make it very popular in Big Data processing.;As MapReduce clusters get popular, their scheduling becomes increasingly important. On one hand, many MapReduce applications have high performance requirements, for example, on response time and/or throughput. On the other hand, with the increasing size of MapReduce clusters, the energy-efficient scheduling of MapReduce clusters becomes inevitable. These scheduling challenges, however, have not been systematically studied.;The objective of this dissertation is to provide MapReduce applications with low cost and energy consumption through the development of scheduling theory and algorithms, energy models, and energy-aware resource management. In particular, we will investigate energy-efficient scheduling in hybrid CPU-GPU MapReduce clusters. This research work is expected to have a breakthrough in Big Data processing, particularly in providing green computing to Big Data applications such as social network analysis, medical care data mining, and financial fraud detection. The tools we propose to develop are expected to increase utilization and reduce energy consumption for MapReduce clusters. In this PhD dissertation, we propose to address the aforementioned challenges by investigating and developing 1) a match-making scheduling algorithm for improving the data locality of Map- Reduce applications, 2) a real-time scheduling algorithm for heterogeneous Map- Reduce clusters, and 3) an energy-efficient scheduler for hybrid CPU-GPU Map- Reduce cluster.