Font Size: a A A

The Design And Realization Of The Genetic Algorithm-based Server-side Dynamic Load Balancing System

Posted on:2006-06-22Degree:MasterType:Thesis
Country:ChinaCandidate:W Y ZhangFull Text:PDF
GTID:2208360155965977Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
The workload on the Web servers is increasing rapidly with the thriving of the Internet and the Web. Availability and efficient performance are strongly required for the Web servers. Achieving load balancing by request dispatching is one of the basic strategies forimproving the performance and efficiency of this system——Cluster.Although a lot of researches have been made within this field, there are still many problems leaving unsolved. In this thesis, the work proposed here investigates how Genetic Algorithms can be employed to solve the dynamic load-balancing problem, and also introduces some methods and technologies in dynamic load-balancing system.GA is an efficient stochastic search algorithm based on the mechanism of natural selection and sexual recombination. It has been generally applied in the fields of function optimizing, task scheduling and so on. After expounding the general idea, the major processes and comparative advantages of GA, the thesis puts emphasis on its applications in the load-balancing. We propose an improved GA called Genetic Algorithms based Dynamic Load-Balancing by seeing newly about request dispatching in the view of task scheduling. Depending on the abilities of GA's stochastic search and its iterative, GA-DLB defines dynamic arrays as the structure of chromosome for the initial population in need of mass-disposing web requests. GA-DLB inducts the possession ratio of server-end CPU as load index to the fitness function, which is real-time watched in the aim of dynamic balance. In the design of select operator, GA-DLB picks out "excellent breed preserving policy" to ensure correct restraint. In addition, GA-DLB in the group structure adopts a strategy of multi-point crossover, which provides more hybridizing opportunity for optimized individuals, and designs variation rate to prevent single population and partial optimal solution. Compared with other balancingalgorithms, GA-DLB has some obvious advantages: ? It really implements dynamic load-balancing by real-time watching server load index; ?By the aid of GA self-adapt and self-study, it depresses the strong conjunction between algorithm and solution, eliminates some puzzles like setting the power etc occurred in other balancing algorithms, and improves its health; and ?It is robust on the base of improved GA and foundation of theory by powerful mathematics knowledge,.In this paper we review the state of the art techniques including server cluster's architecture and varied distribution and load balancer method. On the base of analysis of these methods, it is obvious that we should study the way of data transmission and the dynamic of load balancing. Firstly, we set up a load balancing model in which data is transmitted by direct routing from IP layer. Then we adopt GA-DLB to dispatch web requests from client-end. Besides, based on the acquaintance of the principle of Linux kernel network we modify the driver of NIC and implement the function of load balancing in the way of dynamic loading modules. Due to its running in the kernel level of Linux, the load balancing system is controlled by using virtual character device driver.At the end of this paper, we show the performance tests for the load balancing scheme GA-DLB. Its high efficiency in request dispatch and load balance is proved by experimental results. It is obvious that its application in the network load-balancing is proper and feasible. The summary and successive works are also given.
Keywords/Search Tags:server Cluster, request dispatch, dynamic load balancing, Genetic Algorithms, device driver
PDF Full Text Request
Related items