Performance Analysis of List Scheduling Algorithms by Random Synthetic DAGs

Abstract

Efficient utilization of high performance heterogeneous parallel computing systems requires efficient application scheduling algorithms. List based scheduling algorithms have the ability to generate the most efficient schedules. This type of task scheduling algorithm can be considered as NP-Complete problem. The applications are assumed as multiple tasks interconnected to each other with precedence constraint. Directed Acyclic Graph (DAG) is used to present these tasks. Each of these tasks can be executed on any processor with its own execution timeline. In this paper, four list-based algorithms (HEFT, CPOP, PEFT, EXT_HEFT) have been considered for performance analysis by considering randomly generated synthetic DAGs as input. Further, these algorithms have been compared in terms of the Schedule Length Ratio (SLR), speedup, efficiency, and running time of algorithms. We have also considered the effects on SLR value of algorithms by changing the value of CCR (Communication to Computation Ratio) of the graphs.