An Improved Xen Credit Scheduler for I/O Latency-Sensitive Applications on Multicores

It has long been recognized that the Credit scheduler favors CPU-bound applications while for the latency-sensitive workloads such as those related to stream-based audio/video services, its performance is far from satisfactory. In this paper we present an improved Credit scheduler in Xen to facilitate such tasks on multicore platforms. To this end, we improve the Credit scheduler from three perspectives. First, given the identified Simultaneous Multi-Boost problem, we minimize the system response time by load balancing the virtual CPUs with the BOOST priority between the cores. Second, we address the Premature Preemption problem by monitoring the received network packets in the driver domain and deliberately preventing it from being prematurely preempted during the packet delivery to further reduce and stabilize the I/O latency. Finally, we optimize the frequency of CPU switch by utilizing time-variant slice instead of the existing long time-invariant one to adapt to the dynamic fluctuation of the number of virtual CPUs in the run queue associated with each physical CPU. Our empirical studies show that the proposed improvement can significantly improve the performance of the Credit scheduler for scheduling the I/O latency-sensitive applications.