Generalized Spring Tensor Model: A New Improved Load Balancing Method in Cloud Computing

Abstract

Abstract. Significant characteristics of cloud computing such as elasticity, scalability and payment model attract businesses to replace their legacy infrastructure with the newly offered cloud technologies. As the number of the cloud users is growing rapidly, extensive load volume will affect performance and operation of the cloud. Therefore, it is essential to develop smarter load management methods to ensure effective task scheduling and efficient management of resources. In order to reach these goals, varieties of algorithms have been explored and tested by many researchers. But so far, not many operational load balancing algorithms have been proposed that are capable of forecasting the future load patterns in cloud-based systems. The aim of this research is to design an effective load management tool, characterized by collective behavior of the workflow tasks and jobs that is able to predict various dynamic load patterns occurring in cloud networks. The results show that the proposed new load balancing algorithm can visualize the network load by projecting the existing relationships among submitted tasks and jobs. The visualization can be particularly useful in terms of monitoring the robustness and stability of the cloud systems.Keywords: cloud computing, load balancing, collective behavior, dynamic pattern recognition IntroductionOver the last decade, business and academic requirements from technology perspective have changed substantially with a greater emphasis on more powerful computing techniques. In IT, much of these changes have been driven by prompt success in Internet improvement and economical IT infrastructure development, which resulted in novel structured computational models [1]. Cloud computing is one of these newly emerged paradigms for hosting and delivering services over the Internet.In cloud computing mapping a proper load-balancing algorithm was always an important challenge. The load on the network can be forced by CPU load, memory load, bandwidth load and tasks load [2]. Therefore due to the extensive load volume, the load balancer should prioritize the information using the distributed and heuristic algorithms. Moreover, the load should be managed in a real time manner to prevent