【作者】 张伟；

【导师】 岳德权；

【作者基本信息】 燕山大学 ， 运筹学与控制论， 2015， 硕士

【摘要】 计算机服务系统通常由多个服务器组成,是常见的多服务器排队系统。服务器任务分配不均衡会导致某些服务器太忙或太闲,太忙的服务器会延长任务的等待时间,太闲的服务器造成资源浪费。所以在保持服务器中负载平衡的前提下,有必要对计算机服务系统任务分配进行优化。论文的主要内容如下:首先,研究了多服务器排队系统中不等概率任务分配负载平衡策略的M/G/1排队模型。通过概率方法获得前端分发器对末端服务器的分配概率,并考虑了服务时间服从负指数分布和对数正态分布的情况,求出了性能指标等待时间的期望和方差。利用数值分析的方法,同多服务器排队系统等概率任务分配负载平衡策略进行了比较分析。其次,在两个服务器的排队系统中,研究单边队长控制负载平衡策略的M/M/1排队模型。通过马尔可夫过程理论,获得了平稳概率方程。利用矩阵分析的方法,获得平均队长和平均等待时间的计算公式。通过数值分析,对单边队长控制负载平衡策略和无队长控制负载平衡策略进行了比较分析。最后,在两个服务器的排队系统中,研究双边队长控制负载平衡策略的M/M/1排队模型。通过马尔可夫过程理论,获得了稳态概率方程。利用矩阵分析的方法,获得平均队长和平均等待时间的计算公式。通过数值分析,对双边队长控制负载平衡策略和单边队长控制负载平衡策略进行了比较分析。更多还原

【Abstract】 Computer services system is a common multi-server queuing system which usually consists of multiple servers. The unbalancing allocation task in severs will lead to the server is too busy or too idle, too busy servers will extend the waiting time of the task, while the server is too idle creating waste resources. So in keeping server load balancing premise, it is necessary to allocation the task to a computer service system optimization.The main works are as follows:Firstly, we are researching M/G/1 queuing model of different probability for allocation tasks of load balancing policies in the multi-server queuing system. We obtained allocation probability of the front diapatcher to the end server by probabilistic methods, at the same time, considering service time of the negative exponential distribution and lognormal distribution of cases, obtaining the expectation and variance of the waiting time of performance index. The load balancing policies of different allocation probability for tasks in multi-server compared with the load balancing strategy with equal probability allocation task by using the method of numerical analysis.Secondly, We are studying M/M/1 queuing model of captain unilateral control of load balancing policies in the two servers queuing system. Obtaining the probability of a stable equation in Markov process theory method. Knowing how to calculate the performance index of the average queue length and average waiting time by the method of matrix analysis. The load balancing policies of captain unilateral control compared with the load balancing strategies of captain no control by using the method of numerical analysis.Finally, We are researching M/M/1 queuing model of captain bilatera control of load balancing policies in the two servers queuing system. Obtaining the probability of a stable equation in Markov process theory method. Using the method of matrix analysis to obtain the method to calculate performance index of the average queue length and average waiting time. The load balancing policies of captain bilatera control compared with the load balancing strategies of unilateral captain control By using the method of numerical analysis.更多还原