基于极大熵方法的网格资源分配策略  

Grid resource allocation strategy based on maximum entropy method

作　　者：李志洁[1] 程春田[1] 黄飞雪[2] 李慧贤[3] 

机构地区：[1]大连理工大学水电与水信息研究所 [2]大连理工大学经济系,辽宁大连116024 [3]西北工业大学计算机学院,陕西西安710072

出　　处：《系统工程与电子技术》2007年第10期1738-1741,共4页Systems Engineering and Electronics

基　　金：国家自然科学基金资助课题(50479055)

摘　　要：针对基于市场机制的网格资源分配中的出价问题,提出了一种用于并行任务资源分配的优化策略。该策略先将具有并行任务的多个用户竞争同一资源的行为形式化为一个极大极小问题,然后利用极大熵方法将其转化为一个可微的优化问题,最后通过二分搜索最优解产生一组优化的用户出价,根据这组出价按比例划分资源的计算能力。研究了用户出价函数的特征,给出了求出价组合最优解的具体算法。研究结果表明提出的分配策略具有较低的时间复杂度,从而能够提高资源分配效率,对于动态的网格环境具有更好的适应性。To study the bid problem of market-based resource allocation in grid context, an optimal strategy of resource allocation for parallel tasks is presented. Firstly, the problem of multiple users with parallel tasks competing for a common resource is formulated as a max-rain problem. Secondly, the principle of maximum entropy is introduced to convert this problem into a differentiable optimal problem. Finally, bisection searching method is used to produce a set of optimal user bids to allocate the resource capacity proportionally. Particularly, the characteristics of the user bid function are studied, and the algorithms for searching ates the optimal solution of the user bid set are given. The results show that the proposed allocation strategy generates low time complexity and improves the efficiency of resource allocation, and thus can satisfy the dynamic nature of the grid environment.

关 键 词：网格 资源分配 极大熵 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]