云环境下分级资源分配模型的研究

发布时间：2018-03-09 13:12

本文选题：云环境　切入点：资源分配　出处：《大连理工大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着信息产业的急速发展,迎来了大数据的时代。网络中数据量的剧增给已经成熟的网络结构提出了巨大的挑战。云计算是依托成熟的虚拟化技术,从网格计算、分布式计算和协同计算的基础上发展出来的。而云环境具有异构性和动态性,如何根据用户及任务的特点和需求进行资源的合理分配是需要解决的重要问题之一。针对大型云计算环境下的多节点协作问题建立了动态分级的网络计算模型,并进一步提出了动态分级的资源分配算法(Dynamically Hierarchical Resource-Allocation Algorithm,DHRA).动态分级网络计算模型采用模糊模式识别理论,根据任务和资源节点的信息将其动态地分为不同的等级。从而形成动态分级的网络计算模型。因此对于每个任务只需要在相应等级的节点中寻找合适的节点执行即可,有效地减小了问题的规模。在此基础上,在资源分配算法中引入多Agent机制,增加了系统的可靠性和自主性。综合考虑了任务的完成时间、节点的负载、系统通信量等因素,使得算法在各方面都有较好的性能和效率。对于由大型应用分解的相互关联子任务的并行计算问题,由于所有任务的计算量和所需资源等信息都是已知的,采用随机搜索类算法中的遗传算法。并为了实现多方面的性能优化,提出多目标遗传算法(Multi-Object Genetic Algorithm, MOGA)。采用任务完成时间和任务节点相关性两个适应度函数共同控制种群的进化方向。实现了在保证完成时间的前提下减少通信量的目的。对于DHRA算法和传统的协商算法产生的通信量,进行了定量的理论分析,证明DHRA算法可以有效地减少系统通信量。并且对DHRA算法和MOGA算法在不同的任务和节点数时进行多组仿真实验。将DHRA算法与MinMin算法进行对比,DHRA算法有效地减少的系统通信量的产生,同时保证任务完成时间也有一定的减少。有效地提高了系统的稳定性和执行效率。同样地,对MOGA算法与传统GA算法进行比较,在相同的条件下MOGA算法获得了比传统遗传算法更少的任务完成时间和通信量。都有效地提高了系统的稳定性和执行效率。
[Abstract]:With the rapid development of information industry, the era of big data is ushered in. The huge increase in the amount of data in the network poses a great challenge to the mature network structure. Cloud computing is based on mature virtualization technology, from grid computing, Developed on the basis of distributed computing and collaborative computing. The cloud environment is heterogeneous and dynamic. How to allocate resources reasonably according to the characteristics and requirements of users and tasks is one of the important problems to be solved. A dynamic hierarchical network computing model is established to solve the multi-node collaboration problem in large-scale cloud computing environment. Furthermore, a dynamic Hierarchical Resource-Allocation algorithm is proposed for dynamic resource allocation. Fuzzy pattern recognition theory is used in the computing model of dynamic hierarchical network. According to the information of the task and resource nodes, they are dynamically divided into different levels. Thus, a dynamic hierarchical network computing model is formed. Therefore, for each task, it is only necessary to find the appropriate node in the corresponding level node to execute the task. On the basis of this, the multiple Agent mechanism is introduced into the resource allocation algorithm, which increases the reliability and autonomy of the system. The factors such as the completion time of the task, the load of the node, the traffic of the system, and so on, are considered synthetically. The algorithm has better performance and efficiency in all aspects. For parallel computing problems of interrelated subtasks decomposed by large applications, the information of all tasks is known, such as the amount of computation and the resources required. The genetic algorithm is used in the random search algorithm, and in order to optimize the performance of many aspects, A multi-objective genetic algorithm named Multi-Object Genetic algorithm (Moga) is proposed to control the evolution direction of the population by using two fitness functions: task completion time and task node correlation. The goal of reducing traffic while ensuring completion time is achieved. For the traffic generated by the DHRA algorithm and the traditional negotiation algorithm, A quantitative theoretical analysis was carried out. It is proved that the DHRA algorithm can effectively reduce the system traffic, and the simulation experiments of DHRA algorithm and MOGA algorithm are carried out in different tasks and nodes. Compared with MinMin algorithm, the DHRA algorithm can effectively reduce the system. The generation of communication traffic, At the same time, the task completion time is also reduced, which effectively improves the stability and efficiency of the system. Similarly, the MOGA algorithm is compared with the traditional GA algorithm. Under the same conditions, the MOGA algorithm achieves less task completion time and traffic than the traditional genetic algorithm, and improves the stability and efficiency of the system effectively.
【学位授予单位】：大连理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP3

【参考文献】