Design And Implementation Of Virtual Machine Scheduling In Cloud Platform

As the ordinary data centres and minicomputers fade in the recent years, Cloud computing has become a mainstream solution. With the continuous expansion of enterprise data, BI business, which is closely related to big data processing, has shifted to cloud platforms. However, the virtualization provides very little support for big data processing. Current VM scheduling is one of the critical bottlenecks. BI-PaaS is a Business Intelligence provider based on cloud computing technology. And most of its business is Big Data-Oriented, which makes it the ideal experimental platform of this article. In order to improve the overall performance, designing a scheduler which can provide adequate support for big data processing is the key issue of urgent need at present. Driven by such demands, this paper puts forwards a new virtual machine scheduler optimized for big data processing in cloud computing.This work mainly covers these aspects as following:First, by analyzing the monitoring data of BI-PaaS and the scheduling mechanism in Xen, conclude the characteristics of BI-PaaS workload and the limitations of current scheduler.Second, considering the I/O bounded characteristic of BI workload and the privileges of Domain0in Xen, design an optimized scheduling solution.Third, design a new mechanism to ensure the fairness of VM scheduling.Fourth, design a coarse-grained load balance mechanism for the symmetric multi-processing hardwares.Finally, design an IaaS platform-level VM scheduling solution.