Research On Optimization Of Xen Network I/O Performance

At present, cloud computing and cloud storage technology is developing rapidly.Because the bandwidth throughput of network service affects the user experience of cloudcomputing and cloud storage directly, so people attache great importance to the target ofthe bandwidth throughput of cloud network service, in using of cloud computing cloudand storage. In the growing internet environment of cloud computing, virtualizationtechnology as the core support of cloud computing is more and more concerned orinterested by people. Thereinto, Xen virtualization technology plays an important role.Network bandwidth throughput is also one of the core target in Xen virtualizationtechnology, and also has direct impact on the user experience of cloud computing serviceplatform supported by virtualization technology. However, in the Xen virtualizationtechnology, the usage fo10G network interface card is not so ideal, and the throughput ofits virtual domain is not very high. So this paper researches on the optimization ofbandwidth throughput of cloud computing service supported by Xen; That is theoptimization of Xen network I/O performance.The virtualization of Xen network I/O mainly have two defects: one is that, the singlepair of tasklets for Tx/Rx data transaction in Dom0backend driver become the bottleneckof its network performance, influencing the throughput of whole network; Another is that,Xen system will produce a page granting for each small packets in the processing ofDom0and DomU data exchange,and thus there will be a lot of page granting supercall inthe system. It will affect the speed of sending and receiving data packets, and then affectthe network throughput.This paper presents two optimization points to improve the10G Nic throughput rateof Xen virtual domain. The first one is we use pairs of tasklets for Tx/Rx data transactioninstead of the single pair of tasklets, increasing parallelism, thus improving processingspeed of network packet forwarding in Dom0management domain; The other one isaiming at the small packets transaction process, concretely applying for page storage poolsin advance to reduce the number of page granted,reduce the time overhead of networkpacket forwarding.Through these two optimization points improvement,we accelerate thenetwork data packets transaction processing so as to enhance the10G Nic throughput rate.