答辩 面向实时应用的虚拟CPU调度优化研究.ppt

为了测试vGEDF调度器的效果，本文在四种策略下进行了测试。Baseline就是默认的情况， We conduct experiments under four scheduling strategies to demonstrate the advantages of the vGEDF scheduler. Four PCPUs are dedicated to Domain0 to handle communication and interrupts for other VMs. Fourteen VMs (VM1~VM14) are running on the other PCPUs. VM1 hosts testing real-time applications and the others are interfering VMs which run lookbusy. baseline is the default configuration in cloud environment that only the Credit scheduler is adopted to schedule VMs. Risa is our framework. In this strategy, seven VMs (VM1～VM7) are set as RT-VMs. Risa allocates two PCPUs to rt-pool and applies the vGEDF scheduler to rt-pool automatically. sp+SEDF uses the multi-core static partitioning mechanism to simulate an environment like Risa. The SEDF scheduler is adopted to schedule RT-VMs in rt-pool. Because it does not support load balance among multiple PCPUs, a PCPU hosts four RT-VMs (VM1～VM4) and the other hosts three RT-VMs (VM5～VM7) after VM creation. sp+SEDF(overload) is similar with sp+SEDF, except that a PCPU is overloaded. Because the SEDF scheduler does not support load balancing among multiple PCPUs, it is possible that a PCPU is overloaded while the other has slight load. In this strategy, a PCPU hosts six RT-VMs (VM1～VM6) and only one RT-VM (VM7) runs on the other PCPU. * Voice over Internet Protocol (VoIP) server is a typical soft real-time application. In this slide, we describe experimental results using VoIP server. We start up several concurrent calls that range from 5 to 50 to simulate the real world environment, and measure call quality with PESQ, which ranges from 0 to 4.5. Typically, if the value is greater than 4, it means that the VoIP service has good quality. From the test results, we can observe that Risa is the best among these scheduling strategies, and the call quality is guaranteed under Risa. The Credit scheduler is a proportional fair share scheduler and does not consider real-time const