Optimization On VCPU Scheduling Technologies Based On Application Characteristics In Virtual Machine

With the recent development of virtual desktop systems and the development oflarge-scale cloud computing environments, virtualization technology continues to matureand improve. At the same time, the types of applications running in virtualizedenvironment also appear to be diverse and unpredictable. As the virtual machine monitorfor virtual machines is lack of direct knowledge of the running application characteristics,the uncertainty of application type makes the allocation of resources between virtualmachines difficult. Particularly, virtual machine scheduling has a critical impact on I/Operformance. For virtual machine monitor cannot perceive the application characteristics,we propose an application based VCPU virtualization technology. Through this technique,we are able to perceive the type of application I/O-bound features, and to provide aneffective basis for VCPU scheduling algorithm to allocate physical computing resourcesefficiently, and reduce I/O response delay.This paper analyzes the key technologies of Xen virtualization platform with adetailed study of the Xen I/O virtualization technology and frontend/backend separatedvirtual device driver module. We also analyzes Xen virtual VCPU scheduling algorithm,discusses the accelerating effect of I/O responsiveness for I/O-bound applicationssupplied by BOOSTING mechanism. We present that as for the "semantic gap” betweenthe Xen virtual machine monitor and the upper application level, Xen is not sensitive tothe I/O-bound tasks mixed in hybrid applications, which means that the VCPUscheduling algorithm cannot supply the I/O-bound application with schedulingacceleration and ultimately makes I/O response delay greater. In order to solve theproblem, we propose a kind of new VCPU scheduling technologies based-on applicationcharacteristics.By establishing I/O application-aware agent in virtual network backend, we are ableto inspect the head of network packets and record network traffic history. On this basis,as the application types in virtual machine are unpredictable, we propose a confidencebased I/O-bound criteria considering the short-term and long-term I/O characteristicchanges of applications, which offers a flexible, effective and dynamic method for measuring I/O-bound features.On the establishment of the I/O characteristics measurement standards, we are ableto improve the event notification mechanism, and improve the event channel by addingin urgency flag of notification. We also modify the VCPU scheduling strategy by addingVCPU local boosting mechanism, so that scheduling algorithm is able to sense themixing application I/O-bound characteristic, makes VCPU scheduling more selective,and effectively reduce the I/O intensive response delay.We determine the parameters in the confidence based I/O-bound criteria, and analyzethe influence of the partial boosting rate value on VCPU fairness. We finally demonstratethe decline in I/O latency and rise of throughput in hybrid applications environment afterapplying the proposed application characteristic based VCPU scheduling technology.