Research On Energy Consumption Optimal Management For Cloud Computing Platform

In recent years, the keeping growing size of cloud computing leads to the constantly increasement of energy consumption, high energy consumption has become a conspicuous problem, which restricts the development of cloud computing seriously. Reasonable resource allocation and task scheduling can reduce the idle energy consumption and executive energy consumption effectively, so as to realize the optimal management of energy consumption. The server allocation and task scheduling on the purpose of energy saving are researched in this thesis, which combined with the heterogeneity of compute nodes and randomness of task arrival, the four main contributions in this thesis are described as follows:(1) The sources and reasons of high energy consumption in existing cloud data centers are expounded, the study actuality at home and abroad and characteristics of energy consumption optimization management technique are summarized, the energy consumption management framework of cloud computing system is described, and the resource allocation strategies as well as the task scheduling strategies of energy consumption optimization management are emphatically analyzed.(2) For reducing the idle energy consumption of cloud computing system, a strategy of server allocation based on adaptive threshold is proposed. The strategy adopts the queuing theory to model the task scheduling of cloud computing system, which considers the lost numbers of the tasks and energy consumption of the servers at the same time, adjusts the threshold of task queue length adaptively, and selects the awakening server according to server’s cold point area priority and time priority. The simulation results show that the strategy can ensure the task performance, and reduce the energy consumption of cloud computing system effectively.(3) For the real-time and energy-saving task scheduling, this thesis draws the advantages of Colored Petri Nets and Stochastic Petri Nets, while introducing the concept of queuing theory, then chooses the Queuing Colored Petri Nets(QCPN) as a modeling tool, and extends it with memory token for modeling the task scheduling process of the whole cloud computing system.(4) For reducing the executive energy consumption of cloud computing system, soft real-time and hard real-time task scheduling strategies are proposed, the soft real-time task scheduling algorithm analyzes the reliability and credibility of compute nodes, then sets up a stability assessment model to ensure a better service performance and energy-saving effect by scheduling task to the compute node with higher stability; the hard real-time task scheduling algorithm calculates the guarantee speed to ensure task completion before deadline and calculates the key speed to ensure minimum energy consumpution, and then adjusts the executive speed of processor according to the above two speeds dynamically. The simulation results show that the two kinds of task scheduling strategies have a higher task completion rates and a better energy-saving effects.