一种自适应的网格计算资源组织与发现机制

Self-Adaptive Mechanism for Computational Resource Organization and Discovery in Grids

资源发现是网格计算中一个重要的研究问题.计算资源作为支撑网格应用的基础资源,其组织与发现机制尤为重要,但现有的技术和方法在效率、可伸缩性、自适应的动态演化以及对查询方式的支持方面仍有较大的局限性.基于网格应用对计算资源需求特征的深入分析,通过引入计算资源的主属性概念,按照平衡二叉排序树对计算资源进行分类组织,提出基于资源分类树(resource categorytree,简称RCT)的资源组织与发现机制.首先,讨论了基于RCT对计算资源的组织机制,包括RCT的基本概念和原理、支持资源动态加入和退出以及资源状态动态变化的自组织机制、负载感知的自适应演化机制和基于备份节点的容错机制;然后,在基于RCT的资源组织结构下,设计了支持4种查询方式的搜索算法,并对算法的复杂度进行了分析;最后,通过多组仿真实验对RCT的性能进行了评估.

Resource discovery is one of the important research issues in grid computing. As computational resources are fundamental resources supporting various grid applications, it is of great importance to study computational resource organization and discovery. However, existing approaches are limited in terms of efficiency, scalability, adaptability and the ability to support a wide range of query mechanisms. After analyzing the requirement of application resources, this paper introduces the primary attribute （PA）, and describes a computational resource. Then it proposes resource category tree （RCT） to organize and discover the computational resources. This paper presents the mechanism of organizing resources based on RCT, including the basic concepts and rationale, the self-organizing design to support resource dynamics, the load-aware self adaptation and primary-backup based fault tolerance. It also designs four searching algorithms to support comprehensive query types in the framework of resource organization by using RCT, and a theoretical complexity analysis is presented as well. In the end, the paper evaluates the performance of RCT through simulations.