Performance Evaluation of Scheduling Algorithms in Heterogeneous Cloud Systems

Advanced heterogeneous parallel computing systems require effective scheduling algorithms for efficient execution of complex applications. List-based scheduling algorithms are recognized for their ability to generate the most efficient schedules, despite the NP-complete characteristic of the task scheduling problems. The application is represented as set of inter-dependent tasks that contain precedence constraints that can be modeled as Directed Acyclic Graphs (DAGs). Each task is executed differently based on processor assigned for execution. This paper presents a performance evaluation of five scheduling algorithms, including HEFT, CPOP, PEFT, LSTD and IPPTS on various complex applications such as Gaussian Elimination (GE), Fast Fourier Transformation (FFT), Linear Algebra (LA) in heterogeneous cloud systems. The algorithms were evaluated based on two key parameters: the average makespan w.r.t number of tasks and the number of virtual machines.