Study On Power-Aware Scheduling Of Virtual Resources Within Private Clouds

As a novel large-scale distributed computing paradigm based on virtualized resource pool, cloud computing derived from the ever-increasing interaction and profound development of distributed computing, parallel computing, grid computing, utility computing, web services, etc. If only with accesses to cloud data centers, cloud computing could enable users all over the planet to on-demand leverage a variety of IT-related services based on a pay-per-use model, e.g. infrastructure, platform, software, etc. In addition, cloud computing also corresponds with the basic idea of green computing. Through elastic and high-scalable management of the resource pool, as well as relaxation of the terminal (the"thin"terminal), cloud computing can conserve much energy and cut the cost of both user's and data center's down. It is one of the most promising computing paradigms in the coming low carbon economy.Recently, one of the applied models of cloud computing in industry, private cloud, is getting more and more popular among lots of companies and research centers around the world. Compared to public cloud, private cloud has some special features and requirements, especially respect to resource management and power-aware scheduling. After in-depth exploring theses features, we propose couples of power-aware approaches, which could conserve more energy with acceptable overhead on job efficiency and data center's throughput rate, including the following aspects, among which (2), (3) and (4) are contributions of this dissertation.(1) Under in-depth exploring the cause and the applied scenarios of private clouds, in association with the basic problem of resource management and power-aware scheduling in virtualized environments, we sum up series of sensitive characteristics of private clouds. In this manner, three metrics are applied, i.e. response time of requests, conserved energy and load balancing level, to evaluate the performance of power-aware scheduling solutions. Additionally, we also introduce four paradigms to manage the tradeoff problem between energy consumption and load balancing.(2) As respect to power-aware scheduling of virtual machines, a layout-based approach is proposed using active sleep mechanism, other than addressing the modes of target nodes via thresholds according to most previous researches. Through this approach, the response time can be cut down by a pre-power technique, and the workloads across nodes can be balanced by a min-load-first algorithm. And the experiments show that, this approach can not only shorten the response time of requests, conserve more energy, but also achieve higher level of load balancing.(3) As respect to power-aware scheduling of virtual disks, a power-aware optimization approach is proposed, which is derived from the layout-based approach proposed in power-aware scheduling of virtual machines. It can dynamically adjust the volume of the working pool and alleviate the prolonged problem of response time in case target disk is being stuck on sleep mode. And the experiments show that, this approach can effectively cut job's response time down, meanwhile, reduce the idle time of physical disks.(4) As respect to power-aware scheduling of virtual network, a symmetric multi-processing virtual machine based approach is introduced against the traffic jam problem due to communications among multiple virtual machines within local network. This approach can effectively reduce the number of virtual machines within a secure group, lower the communication threshold and promote the performance of parallel computing jobs. Moreover, through cutting down the number of communication zones, this approach can greatly relax the coupling connection between virtual group and local network, and therefore, it can meet the objective of power-aware network.(5) After reviewing existing computer-supported collaborative learning systems, some limitations related to underlying resources, such as insufficient scalability and high energy dissipation, are concluded. Using private clouds as the base infrastructure, an elastic cloud service based power-aware collaborative learning system is proposed using the above-mentioned power-aware approaches. It can enable users with virtualized collaborative environment on demand, support elastic scalability of resources and make good use of energy.