Research On Elastic Memory Resource Provisions In Virtual Computing Environment

The new urgent needs presented by the application in all aspects of social life and the emergence of a large number of new network applications, have brought new challenge to the effective management of massive computing resources in the network. Especially, with the features of growth, autonomy and diversity, many new problem has come in to being of the process of collaborative sharing and integration utilization of resources.Specially, the problem of wide-area distributed and heterogeneous memory resource sharing and allocation has become one of the important area of the various problem. As one of the three major computing resources, memory resources, relative to CPU and I/O resources, is harder to be shared and reused, since there exists “semantic gap” between VMM and virtual machines. So the elastic memory resource allocation and sharing research work has important academic significance and commercial value. Especially in the new Internet applications, the memory usage scenarios will become more unpredictable. Around this problem, the following four main contents of this paper conducted in-depth researchs.Firstly, we have deeply analysised the XEN virtualization platform and architecture, as well as paravirtualized drivers, memory virtualization modeling techniques, which are implemented in a virtualized environment consolidate memory key supporting technologies; Then, this paper focuses on the memory balloon and method of memory hotplug technology, page sharing technology and live migration technology. Not only it covers their advantages, but also pointed out the disadvantages of these methods.Secondly, the use of memory balloon method of the virtual machine has the probability to bring big variations in performance degradation. Based on this observation and also in order to achieve the purpose of solving this problem, we propose a method called memory holes. Memory hole method in the virtual machine monitor allocates "memory hole" which does not have physical memory. When memory access operation happens in the virtual machine, virtual machine monitor only allocate real physical memory for it. Experiments show that memory hole method can indeed surge in memory in case, a reasonable allocation of memory to ensure the quality of service the virtual machine.Next, with the in-depth exploration of the guest virtual machine paging issues, we propose a host-based paging caching approach to optimize the guest paging operations. The method uses parts of the host memory to cache memory pages swapped out by the guests. Also we design algorithms to reasonably allocated caching pool size following each virtual machine memory resource usage. Our test show that, our method has the ability to balance the host and client memory usage, and also to improve the overall efficiency of memory resources.Furthermore, we propose a network-based memory paging methods to overcome a single physical node memory capacity limit. The method uses memory on other physical node and efficiently organized them as a huge storage to contain swapped out memory pages in virtual machine. With the support of advanced network equipment(such as Infiniband and 10 Gb Ethernet), network-based memory paging method has certain performance advantages overcoming the traditional method of paging in the testbed.This article is a useful exploration of elastic memory resource allocation in virtual computing environment. Also it is a good theoretical research to build a flexible and scalable of massive data processing methods and also shows practical significance to consolidate in cloud computing platform. This work has been applied in the project of National Basic Research Program of China(973 Program), National High-tech R&D Program(863 Program) and Natural Science Foundation of China.