A comparative study of the semi-elastic and fully-elastic mapreduce models

MapReduce which was initially proposed to handle big data in a cluster of computers, is becoming a popular programming model for big data processing in cloud computing. When MapReduce is used in cloud computing where everything is a service and the quality of service is important, a new issue that must be addressed is how to ensure a MapReduce computation will finish before a deadline in a dynamically changing cloud computing environment while minimizing its computation cost. The original MapReduce model cannot address the issue as it is not elastic, that is, it does not support adding resources to a MapReduce computation duration the runtime. To overcome the drawback of the original MapReduce model, a fully-elastic MapReduce is proposed in this paper. In addition, in this paper we study the performance of the fully-elastic model by comparing it with an existing model, namely, semi-elastic model, by theoretic analysis and by numerical experiments.