Phase Level Scheduler for Map Reduce Using Grained Resource

Abstract

MapReduce is one of the important concepts of Hadoop that is used for data handling used by big companies today such as Google and Facebook. Here we divide each job into the map and reduce phases and try to complete the execution of the assigned task in a parallel form. In this paper, we suggest that it would be more efficient if we make the scheduler to work at the phase-level instead of the task-level. The reason is because the task demands a lot of requirements during its lifetime. For this very purpose, we introduce the concept called PRISM, which is aphase and information-aware scheduler for MapReduce and in this concept we divide the tasks into unequal parts called as phases and apply phase-level scheduling to these phases and achieve efficient resource usage.