Research On The Web Server Cluster Technologies

Web application is the most important and popular application mode in Internet, and now the B/S computing model based on web server has been used in more and more applications. As the complexity of applications and increasing users, the capcatiy of web server system is facing great chanllege. Substituting legacy system with more powerful servers will lead to great cost, and the previous investment can not be protected. Also, this is not a real solution. The web server cluster technologies that organize a group of web servers together in some mode and construct them into a single server image provide good approachs for the persistent extending of capacity in web server system. High price/performance rate and good system realiabilty can be reached in these systems, and the capacity of system can be expanded through adding servers into the cluster. So, the web server cluster technologies become crucial in constructing large web site system.Presently, there are a few problems in web server cluster area need to be studied and sloved. Firstly, with existence of some processing bottleneck, the current architecture of web cluster system can not meet the requirements in large web cluster applications, because of the limitation of system scalability and reliability. Then, the load distributing stretegies and algorithms play important roles in web cluster system. With the wide usage of dynamic page techniques, the load characteristics in web site become more complicated than ever, the current algorithms can not adapt to different load conditions. So some new kinds of load distributing alogrithms with high performance that can be used in any load conditions are expected to appear. Additionally, with the solving of QoS problems in network tranmitting, the research of QoS control is now moved to web application level. Therefore, studying on QoS in web applications, especially in web server cluster system, has great meaning to the web applications in next generation.As at the problems above, a few crucial technologies in web server cluster system have been studied in this disseration, especially on the architecture of web server cluster system, strategies and algorithms of load balancing, and the application QoS control in web cluster system. Several innovative results are achieved as the follows:a) A high-scalable distributed architecture of web server cluster, named as DCA, is proposed. In this architecture, the functions of dispatcher are totally distributed on eachbackend server, so the processing bottleneck that results from centralized processing of all system loads through single diapatcher can be avoided, and the system scalability is improved. In the meantime, the single dispatcher's negative effect on reliability can be eliminated. The control mechanisms for load balancing and locality under DCA are discussed, and appropriate strategies in implementation are proposed. A two-level DCA based on the revised DNS mechanism is proposed, which provides a method to improve the scalability of DCA.b) A content-based dynamic load distributing strategy and the Combined Locality Aware Request Distribution algorithm, as CLARD, are proposed. Due to the different characters in page providing mode for static and dynamic pages, this strategy analyses the content of HTTP request by layer-7 switching technology, identifing the type of request loads and distributing them in different ways. As the distribution of static loads, locality control is considered as the primary factor to elevate the cache hit rate, so the page accessing time can be decreased and the throughput of whole system can be improved; As the distribution of dynamic loads, load balancing is considered as the primary factor, the requests which have the same URL are distributed averagely among servers, so the balanced load distribution can be reached.c) A web application level QoS metric and a load modeling method are proposed. Based on the alalysis of characteristics in typical web applications, and combined with ideas in QoS metric for network transmitting, a Q...