Design And Implementation Of Scheduling System Based On OpenStack

With the development of IT technology, cloud computing technology has become more and more mature in recent years. Due to the rapid increase of the number and scale of the data centers, how to reasonably use the resources and effectively save the energy consumption has become a hot topic.IaaS layer as the bottom layer of cloud computing is the basis of the entire cloud computing architecture, and the most important technology is virtualization in this layer. The typical technology of virtualization is server virtualization, which is to run multiple virtual machines on a physical server. The resulting virtual machine migration technology not only solves the problem of system failure and upgrade, but also plays a role in improving the quality of service and saving the energy consumption in a data center. OpenStack which is a popular open-source cloud computing platform also has introduced the strategy of virtual machine migration. Virtual machine migration in OpenStack requires administrator privileges and we has to use the command line or operate on the OpenStack management panel to trigger the migration manually. Non dynamic virtual machine scheduling is obviously unsatisfactory, and the host selection algorithm of virtual machine migration of OpenStack is based on the residual amount of memory of hosts, which is too simple. Therefore, a complete set of dynamic virtual machine scheduling algorithm is needed.After analyzing the existing scheduling strategy of OpenStack platform in detail, combining with the existing virtual machine migration technology, this dissertation designs a scheduling system based on OpenStack. The system uses AR(n) model to predict the future performance of a host of m moments, and uses several thresholds to determine when the machine migration happens; the system proposes the concept of transfer volume to select the virtual machines, choosing virtual machines in accordance with the order of volume from small to large; when select destination host, the system selects only one destination host for one virtual machine. The system filters out hosts that don’t meet the demand at first, and then gives different weight coefficients to each physical resources of a host, and then sort the candidate host with the total weight, and select one host randomly.In this thesis, CloudSim is used to simulate the proposed algorithm, which proves that the algorithm can effectively reduce the energy consumption in the data center, improve the utilization of resources and maintain the stability of the system.Finally, the system is realized. After several experiments and observation of system operation log, it proved the algorithm is feasible on the platform of Open Stack. Based on the simulation results, the effectiveness of the system is proved.