分布式系统中任务分配的一种结点自适应算法

基于一个非常简单的假设：各独立结点对系统反馈的认识具有一致性，提出了一种结点自适应算法，文中运用概率论的相关知识，分析了算法的各项重要性能指标，给出了一般情况下的表达式，并运用分段讨论和求极值等方法得出了这些表达式在特殊情况下的简化形式，算法所需条件较弱，具有广泛的适应性。

A Self-Learning Algorithm for IPUs in Distributed Systems

In a distributed system with many independent processing units (IPUs), such as a computer network, one of the most important things is to establish an assignment method for distributing tasks. If there were a central administrator who knows well about both the abilities of all IPUs and the situations of all tasks, this problem would be fairly easy to solve. But in most cases, such an administrator does not exist. Worse yet, IPUs cannot know one another's condition, the value of the best result, and so forth. All what they know may only be: 1, their own current situations; and 2, the feedback of the system after tasks are chosen. Basing on a simple hypothesis, that is, all IPUs have the same knowledge of the system's feedback, this paper proposes a task assignment algorithm. Using some knowledge of probability theory, authors analyze some performance aspects of this algorithm and get their expressions under all situations. Some simplified forms of these expressions are also given under some situations using divide and conquer and extremum finding methods. In the range of this paper, this integrated analysis method is more simple and practical than Markov chain.