A gradient-based optimization approach for task scheduling problem in cloud computing

Task scheduling in cloud computing is a key component that affects the resource usage and operating costs of the system. In order to promote the efficiency of task executions in the cloud system, many heuristic algorithms and their variants have been used to optimize scheduling. Since makespan is the vital metric of cloud computing system, most of the relevant research focuses on improving this performance. The gradient-based optimization (GBO) has a faster convergence rate, and can avoid prematurely falling into the local optimum. In this work, we propose a task scheduling based on the GBO in the cloud to improve the makespan performance. Since the GBO is proposed for continuous optimization, rounding-off method is used to convert the real "vector" value of the GBO to the nearest integer value, thereby representing the solution of the task scheduling problem. To evaluate the performance of the proposed GBO-based scheduling method, two experimental cases are performed. The results of the two experimental cases show that compared with current heuristic algorithms, the GBO has better convergence speed and accuracy in searching for the optimal task scheduling solution, especially in the presence of large-scale tasks.