Research On Schedule Performance Of VMM Based On Multi-core

With the development of science and technology, CPU of servers and micro-computers has the transition from single core to multi-core. Future computers will have more and more computing resources. System virtualization turns a single phys-ical computer into a number of virtual computer systems, which can support multiple virtual machines running simultaneously in multi-core environment. A single system can be subdivided into multiple systems to accommodate more execution environment and more applications. This improves system utilization. While system virtualization brings flexibility to high performance computing, but different high-performance com-puting applications have different needs of computing resources. Therefore, how to manage virtual computing resources efficiently based on the need of applications and how to make full performance of multi-core architecture remain to be further studied.Based on the Intel Hardware Asisted Virtualization and Xen virtual machine mon-itor technology, new scheduling algorithm of the virtual machine on the existing frame-work of CPU virtualization is designed and implemented to improve the performance and reduce the latency of scheduler. Further performance optimization is made based on CPU hot-plug.The main contents are as follows:An analysis of Intel Hardware Asisted Virtualization and Xen VMM concludes that the scheduling algorithm plays a key role in the process of allocation of computing resources. It has a significant impact in the performance and response latency of virtual machines.Based on existing scheduling algorithms in Xen and the needs of the actual en-vironment, a new scheduling algorithm is designed and implemented. The scheduling algorithm treats domains of critical applications and domains of non-critical applica-tions separately. Domains of critical applications require high performance and low latency. Each vcpu of domains of critical applications binds a physical CPU to each other. The CPU will not be interrupted by scheduling or context switch. It improves the performance and reduces the response delay of vcpu. Vcpus of domains of non-critical applications runs on physical CPUs which are not binded. Scheduling ensures the vc-pus equitable access to physical CPUs. Load balance ensures physical CPUs having average utilization rate.Operating system in papa-virtualization support CPU hot-plug. When domains of critical applications have a large number of vcpu, which makes domains of non-critical applications can only run on a relatively small number of physical CPU, hot removal vcpu of DomainO can reduce the total number of vcpu without influencing tasks and functions. This will reduce the performance degradation caused by frequent switching of vcpu.Linpack test tool and Mini-OS are used to measure computing performance and network response latency. The comparison of measurements shows that above work can reach a higher computing performance in domains of critical applications than original Scheduler in Xen on the same CPU and hardware environment. Comput-ing performance of domains of critical applications is not affected by other domains. Response latency of external demand decreases dramatically on domains of critical applications.