The Research Of Performance Evaluation And Approximate Caculation For Large-scale Parallel Service System

In a large scale parallel service system, how to adjust the customers in system whoare waiting for service is the method for improving resource utilization and service rateof system, and reducing customers’waiting time. In such a system, dynamic random loadbalancing strategy is an effective way to achieve system performance optimization, andhas important application prospect.Large parallel service system has many servers, this makes us have to deal with toomany states. This characteristic of the model makes precise mathematical calculation isdifficult. So, we process double server symmetric model, JSQ model and supermarketmodel scientifically to get corresponding upper bound. Our ultimate goal is to achieveapproximate calculation of three models. Specific steps are as follows.Firstly, do orderly expression of states. In general, we express the state of system byqueue length. In this thesis, to simplify the expression the state,we do orderly expressionto states.Secondly, assume existence condition of state. In practice, customers will change toshorter queue automatically, so some state which longest queue is too more longer thanshortest queue will not exist. The setting of the threshold value T can simplifymathematical calculation.At last, design the redirecting method of state transition. To get upper model,weassume redirecting method of double server symmetric model, JSQ model and SQ(d)model. The so called upper model means that the performance indicators are the upperlimit of original model.According to this research, upper bound model of large-scale parallel service systemcan be given. Theory support for the operation and optimization of the actual system canbe given.