The Analysis Of Fairness-efficiency Allocation Strategy On Heterogeneous Cloud System Resources

Since the rapid development of cloud technology and applications,the cloud services require more types of resources to support the high performance.But there are also some new difficulties and challenges in the allocation and scheduling of heterogeneous cloud resources.The employment of heterogeneous cloud resources(e.g.CPUs and GPUs)often requires the optimization according to two important metrics,that is,fairness and efficiency.To solve these problems,this paper has made the study about the fairness-efficiency allocation strategy on heterogeneous cloud resources.Our contributions can be summarized as follows:1.We proposed a fairness-efficiency heterogeneous cloud resource allocation algorithm according to the subgradient method of optimization problems.It is designed as a dynamic-adaptive,iterative heuristic allocation algorithm.2.We completed the design of heterogeneous cloud resource allocation platform with CPU-GPU heterogeneous cloud resource.The function of the platform includes collecting workload runtime data,invoking the heterogeneous resource allocation algorithm,managing workload execution and monitoring runtime data.We developed the system testing platform with different techniques about data-collecting and process-controlling such as Easy Hook and CUPTI for the graphic and general computing workload execution requirements in the implementation respectively.3.We evaluated the resource allocation strategy about the CPU-GPU heterogeneous resources.The result indicates that there is an apparent fairness metric improvement with our strategy,especially in scenarios that there are many computing-intensive workloads.Compared with the VGASA strategy,the fairness metric value improved 45% with our strategy.4.We performed the large-scale simulation for the strategy in the large-scale heterogeneous cloud resource scenario using the simulation platform based on the Cloud Sim.The simulation results proved the availability and the scalability of our strategy in large-scale environments.