Distributed Web-server systems have been widely used to provide better Internet services. With the development of multimedia Web sites and E-commence Internet applications, existing load balancing approaches can no longer satisfy the requirements of either the service providers or the users....[ Read more ]

Distributed Web-server systems have been widely used to provide better Internet services. With the development of multimedia Web sites and E-commence Internet applications, existing load balancing approaches can no longer satisfy the requirements of either the service providers or the users.

In this thesis, a new two-tier routing control mechanism is proposed for Web-server networks. This mechanism is designed to achieve reward-based dynamic routing control with high system reward, low response time and high cache hit ratio. The first tier is for wide area network (WAN) control. The second tier is for local area network (LAN) control. The implementation plan of the mechanism is also presented in this thesis.

At the WAN tier, a new reward-based control mechanism is introduced. This modified dispatcher-based mechanism satisfies the dynamic content-based control requirement while overcoming the disadvantage that the dispatcher itself being a network bottleneck. Based on an MDP model, an efficient algorithm is proposed. A centralized dipatching with decentralized admission (CDDA) approach is used to distribute the control related computation to each server pool to make the problem easier to solve. A state-block heuristic approach is developed to reduce the state space so that the problem is computationally feasible. Simulation results demonstrate that the proposed state-block approach can not only reduce the state space but also provide a good approximation of power-tailed request interarrival times.

At the LAN tier, we construct an MDP model taking into account of the general cache content replacement rule. A Weighted Load Balanced (WLB) algorithm enlightened by this MDP model is proposed to minimized the response time. The WLB algorithm can be easily implemented to any distributed web server and cache systems. Large amount of simulation results show that the WLB algorithm leads to shorter response times than those obtained from the existing algorithms including the round-robin algorithm, pure load balancing algorithm or maximum cache hit ratio algorithm.