Research On Pso Algorithm Based Taskscheduling In A Hybrid Cloud

Due to economy, elasticity and scalability of services provided by cloud computing, more and more corporations and individuals rent Infrastructure as a Service (IaaS) to support their business. As the foundation of other services, IaaS plays a supporting role in cloud computing. However, when an IaaS provider faces peak demands and especially no enough local resources, how to schedule tasks to meet all users’ requests becomes a big challenge.Previous researches proposed purchasing machines in advance or building cloud federation to resolve this problem. However, the former is not economic and the latter is hard to be put into practice at present. In this paper, we propose a hybrid cloud architecture, in which an IaaS provider can outsource its tasks to External Clouds (ECs) without establishing any agreement or standard when its local resources are not sufficient. The key issue is how to allocate users’tasks to maximize its profit while guarantee QoS. In our work this problem is formulated as a Deadline Constrained Task Scheduling (DCTS) problem, and a Self-adaptive Learning Particle Swarm Optimization (SLPSO) is proposed to maximize utilization and profit from the perspective of an IaaS provider. In SLPSO, four velocity updating strategies are suggested to adaptively update the velocity of each particle to make it have a good diversity and robustness. For static task scheduling, the SLPSO is compared with Standard-PSO and CPLEX. Experimental results show that SLPSO is very effective for the DCTS. Compared with Standard-PSO, it improves profit by0.25%for problem1,11.56%for problem2and2.26%for problem3, and compared with CPLEX, it improves profit by-0.04%for problem1,16.71%for problem2and2.37%for problem3.For dynamic task scheduling, a dynamic SLPSO is proposed by replacing the invalid personal best. Besides, the results got by the dynamic SLPSO arecompared with that got by static SLPSO.