A Dynamic Load Balancing Algorithm For Web Server Cluster Based On Stochastic Petri Nets

The Web server cluster system has excellent scalability, high performance and high availability, and can be built by using the low-end equipments, with high ratio of performance to price. By virtue of this unique technical superiority, the Web server cluster is being in rapid development in recent years, with wide applications. However, some performance of the Web server cluster system is not desired, the load balancing algorithm is the key to improve it.Based on a study on the Web server cluster related technologies and a analysis of the current main load balancing algorithms owned advantages and disadvantages, a new dynamic load balancing algorithm named LBRQC is proposed based on quantification and classification of requests. Mechanisms of request overhead quantification and classification are introduced into the algorithm. And the mechanism of queuing the ready request processes by category reduces the complexity and overhead cost of fetching the servers load information, improves the accuracy of the servers load information so as to achieve effective load balancing. An admission control mechanism is also introduced to avoid system overload and maintain a high level of performance. A stochastic Petri net modeling approach is used to build an advanced stochastic Petri net model (SHLPN) for the web server cluster based on our algorithm. The model is simulated by the stochastic Petri net package (SPNP). The simulation results show that, compared with the dynamic WRR algorithm and CAP algorithm, our algorithm achieves performance of higher throughput and lower response time.