Research On Memory Resource Allocation On Demand And Co-scheduling On Virtualized Cloud Platforms

Recent years, new generation Internet applications develop rapidly, such as social networking, e-commerce, online video, high-capacity network disk and so on. These applications, with the characteristics of large scale, rapid business growth rate, require an increasing number of high-performance servers, and high maintenance costs of hardware and software. To solve these problems, cloud computing emerged as a new computing pattern, and have received sustained attention from academic and business fields. Cloud computing consolidates large-scale computing, storage, and network resources in the form of data centers, and provides Internet users a variety types of services at low cost. Users can easily apply for and release resources based on their business requirements, and pay according to their demands. Cloud computing guarantees the quality with a lower cost.Virtualization technology is the underlying support of cloud computing platform which abstracts and unifies the underlying resources. Compared with traditional technologies, the greatest advantage of virtualized cloud computing platform is flexible allocation of resources. However, because of the semantic gap between the platform and upper layer services, the flexible allocation of memory becomes the bottleneck. This thesis does some research on memory resource allocate on demand and co-scheduling on virtualized cloud computing platform. The main contributions are:Based on analysis of the features of memory virtualization technology, this thesis proposes and implements a transparent method to sense the memory usage of virtual machines under x86 architecture. This method takes use of the visibility of processes and page tables of virtual machines. Compared to other existing methods, this method has the merit of low overhead, high accuracy and real-time.Based on depth analysis of the co-action between memory and storage, this paper proposes a collaborative memory and storage resource allocation method, which takes the need of cache and the usage of swap space into consideration. This method greatly improves the I/O intensive applications.Based on the transparent sensing and collaborative resource allocation method, this paper proposes a memory resource scheduling policy which considers the fairness and priority. The policy can ensure the performance of virtual machines and support difference of priority configurations. As a result, the memory resource utilization, performance and global performance are improved.To sum up, this paper is a positive exploration of the memory resource allocation on demand and co-scheduling. It has high practical value in constructing virtualized cloud computing platform. Supported by the 973 and 863 projects, this paper has brought benefits to build an efficient cloud computing platform.