Cloud computing uses a great amount of heterogeneous resources to deliver countless different services to users of distinctive quality of services (QoS) requirements. Numerous diverse tasks need to be carried out to meet the vastly different QoS and budget requirements. Workflow scheduling is therefore critical for the success of large-scale cloud computing. Particle Swarm Optimization (PSO) has been adopted for workflow scheduling in cloud computing, yet most existing works focused on a single objective. This paper proposes a tunable fitness function for the PSO algorithm, based on which a workflow schedule may be selected for minimal cost or minimal makespan (completion time), or any level in between. A heuristics is further proposed to address bottleneck problems and attains a smaller makespan. Performance evaluation and complexity analysis are both presented, which show that the proposed algorithm surpasses the existing ones in both cost and makespan while maintaining a reasonable load balance and keeping the same time complexity. We believe that the tunable fitness function-based PSO have many potential applications in other soft computing and distributed computing models.
展开▼
