Research And Application Of Queuing Theory Modeling For Server Cluster Load Balancing

With the rapid development of Internet and computer technology,a large number of user task requests have brought huge pressure to the servers.A single server obviously can't satisfy the needs of users,so the server cluster technology emerged as the times require.Load balancing technology is one of the key technologies in cluster system,which can balance the server load to ensure the normal operation of the system,and then improve the overall performance of the system.Thisthesis studies the load balancing algorithm from the aspects of the load balancing queuing system modeling,the load comprehensive index of each server construction and prediction,task scheduling strategy,etc.The specific contents are as follows:(1)Thisthesis proposes a dynamic load balancing algorithm based on queuing theory and comprehensive index evaluation.From the point of view of building the comprehensive index of server load,the existed load balancing algorithms are based on the occupancy rate of CPU,memory,process and other parameters to evaluate the current load situation of the server,but the complexity of the server load situation often makes it difficult to accurately evaluate.To solve this problem,a dynamic load balancing algorithm based on queuing theory comprehensive index evaluation is proposed.Firstly,the queuing theory is introduced to establish load balanced queuing model,then the queuing theory is used to establish the load comprehensive index of each server.Finally,according to the load comprehensive index of each server to evaluate the real-time load of the server,the tasks in the input queue are assigned to each server one by one.(2)Thisthesis proposes a dynamic exponential smooth prediction load balancing algorithm based on queuing theory modeling.In the process of collecting load information on a regular basis,if the time interval is too small,collect the load information frequently will increase timeconsuming and energy consumption;if the time interval is too large,it will lead to a certain lag of load evaluation indicators,resulting in a large error.To solve this problem,a dynamic exponential smooth prediction load balancing algorithm based on queuing theory is proposed.Firstly,the dynamic adaptive exponential smoothing model is introduced on the basis of building the load synthesis index by using queuing theory to predict the load situation at the next moment,and then the task scheduler assigns the tasks to each server according to the load prediction value.(3)Thisthesis proposes a improved dynamic load balancing algorithm based on greedy algorithm.According to the queuing model of load balancing cluster system and the flow of load balancing algorithm,the simulation platform of load balancing algorithm is established.Firstly,the simulation platform consists of four parts: task generation module,task scheduler module,task processing server module and information sharing model.Then these load balancing algorithms proposed in thisthesis are simulated by the simulation platform,and the experimental results are analyzed.The experimental results show that the three load balancing algorithms all achieve good load balancing effect,and reduce the average waiting time of task request.(4)In thisthesis,cluster and load balancing algorithm are introduced into the parallel training of Ada Boost classifier for face detection,which can effectively balance the load of the server to reduce the training time of Ada Boost classifier.In thisthesis,a simulation platform of load balancing algorithm is established to compare the load balancing algorithm and its application.The experimental results show that the load balancing algorithm achieves good load balancing effect,effectively reduces the average waiting time of tasks,and effectively reduces the training time of the classifier in the Ada Boost classifier parallel training of face detection.