节点文献

网格环境下Master-Worker计算的研究和实现

Research and Implement of Master-Worker Computation in Grid

【作者】 付志祥

【导师】 王晓东;

【作者基本信息】 福州大学 , 计算机应用技术, 2006, 硕士

【摘要】 Master-Worker并行计算是一类比较典型的分布式计算,存在一个Master进行少量的运算,而将大量的计算分配给各个Worker进行,最后将各个Worker的计算结果收回Master进行汇总。国外有不少对于传统的Master-Worker计算和改进的Master-Worker计算的架构的研究。但依然存在不能更好的发现资源和对资源集中管理等等的不足。网格是继万维网之后出现的一种新型网络计算平台,它的目的就是为用户提供一种全面共享包括计算资源在内的各种资源的基础设施。通过对网格和Master-Worker计算特点的理解和研究,发现在网格平台下研究和实现Master-Worker并行计算能够克服传统的Master-Worker计算的一些不足之处,如无法对资源进行集中管理等不足,对于这一类分布式计算有着很大的推进作用。本文研发的毕方拾遗网格计算平台和毕方计算网格计算平台就是利用拾遗网格和计算网格进行Master-Worker并行计算的可屏蔽异构性和分布性的计算平台。本文首先详细给出了两个计算平台的系统架构和功能模块,其次对于本平台中的资源管理,包括资源表示、资源注册、资源发现的具体实现进行了描述,接着又对系统中作业调度算法进行了研究和实现,给出了一种改进后的蚁群算法,最后给出了平台下的实验测试数据和结果分析。

【Abstract】 Master-Worker parallel computation is one kind of typical parallel computation, which has a master doing little computation itself, distributing most computational tasks to workers and finally gathering the results. In abroad, people did many researches on the conventional and improved Master-Worker computation, but until now, there are still many problems unsolved such as no effective method to find resources easily, to administer resources concentratedly and so on. Grid is a new kind of computing platform after the World Wide Wed. It is a basic establishment whose aim is to share all resources including the computing resource for its users. According to the research on the characters of grid computation and Master-Worker parallel computation, we find that on grid platform, we can overcome the shortages of traditional Master-Worker parallel computation, such as centralized management of resource. This will improve Master-Worker parallel computation. Bifang computing platform of scavenging grid and Bifang computing platform of computing grid, which I contrived, can mask resource heterogeneity and distribution. It is a system using scavenging grid and computing grid to do master-worker parallel computation. In this paper, we introduce the frameworks and modules of these two platforms and do some research on resource management including resource description、resource registration and resource discovery. Then we research and realize job assignment algorithms in these platforms, and introduce an improved ant colony optimization. Finally, we offer the testing data and result analyses.

  • 【网络出版投稿人】 福州大学
  • 【网络出版年期】2006年 06期
  • 【分类号】TP393.01
  • 【被引频次】8
  • 【下载频次】87
节点文献中: 

本文链接的文献网络图示:

本文的引文网络