تجزیه و تحلیل مدل سازی و عملکرد توازن بار QoS آگاه از خوشه های وب سرور

This paper introduces mechanisms to correlate contents and priorities of incoming HTTP requests used for server process scheduling with the load balancing policies for Web-server clusters. This approach enables both load balancing and Web quality of service (QoS). Another contribution is a modeling and analysis technique based on stochastic high-level Petri net methods for QoS-aware load balancing. We propose an approximate analysis technique to reduce the complexity of the model.

The Internet is evolving from a communication and browsing infrastructure to a medium for conducting business and selling services. Enterprises and service providers are increasingly motivated to migrate mission-critical services to the Web [1]; on-line banking, stock trading, reservations, product merchandising, to name a few, are services being offered via the Web. Since not all transactions are equally important to the clients or to the servers, e-commerce applications generally desire preferential treatment foreign to the egalitarian philosophy of TCP/IP and HTTP. To that end, Web servers must be capable of providing differentiated quality of service (QoS). This paper introduces mechanisms to correlate contents and priorities of incoming HTTP requests used for server process scheduling with the load balancing policies for Web-server clusters. This approach enables both load balancing and Web QoS. Another contribution is a modeling and analysis technique based on the stochastic high-level Petri net (SHLPN) [24] to investigate QoS-aware load balancing. We propose an approximate analysis technique based on the decomposition and iteration methods to reduce the complexity of the model. The paper is organized as follows: In Section 2 we review related work and in Section 3 we discuss scalable server architectures and load sharing models. In Section 4 we provide an informal introduction to SHLPNs and then present the SHLPN model of the Web-server cluster. In Section 5, we specify the server process scheduling policy and load balancing policy concerned, and introduce a QoS-aware load balancing policy based on them. The metrics used in performance analysis is also described. In Section 6 we present the numerical results by two examples to show the performance benefits of QoS-aware load balancing. In Section 7 we propose an approximate analysis technique to cope with the state-space explosion problem, and give the corresponding numerical results and validation. Finally, Section 8 concludes the paper with discussions on the future work.

We propose to integrate the priority levels used in HTTP process scheduling at server nodes with the load balancing policy used by the request dispatcher in the Web-server cluster to simultaneously achieve load balancing and high Web QoS. We present the architecture for QoS-aware load balancing in the Web-server cluster, and explore the performance benefits of QoS-aware load balancing using SHLPN modeling and analysis technique. To cope with the state-space explosion problem, we propose an approximate analysis technique based on the decomposition and refinement methods as well as iteration technique, which significantly reduces the complexity of the model solution and are also validated efficient for both complexity and accuracy for numerical results. Our analysis shows that the QoS-aware load balancing policy E-FSPF outperforms the QoS-unaware load balancing policy FSPF in balancing the high-priority requests and can thus provide better QoS to the high-priority requests. This demonstrates that the E-FSPF policy is a better choice than the FSPF policy for load balancing in the domain of e-commerce. This study is only a preliminary step in exploring the performance of QoS-aware load balancing for Web-server clusters. We believe that there are other mechanisms and policies deserving scrutiny, for example the performance of the cache within the Web-server cluster [26]. The content-aware request dispatchers enable the use of a partitioned server database and specialized server node, and high degree of locality in the back-ends' main memory caches becomes an important issue [27]. Future work will take into account caching performance within the cluster. Besides, a complete QoS solution would require a combination of networking QoS and Web QoS supports. Future work will investigate the geographically distributed Web-server systems, and the networking impact will also be considered for QoS-aware load balancing.