Performance Analysis of Voting Algorithms with Non-zero Network Delay and Site Processing Time

Voting is the most popular replica control algorithm due to its simplicity in implementation and low communication and computing overhead. The basic idea of voting is to assign each site a vote. Majority consensus is required for each update request so that at any time only one update can be committed in the system. In order to compare and evaluate different voting algorithms, analytic models have been developed to obtain the desired measures such as the site availability and the mean response time. Early studies related to availability/performance analysis of voting algorithms used Markov chain models [3]. The use of stochastic Petri nets (SPN) in this area has significantly increased the modeling capability and enabled comprehensive evaluations of voting algorithms [2]. All the previous analytic models ignored network delay and site processing time to remain tractable and easily solvable. Although the no-delay assumption is justifiable in the situation where all the sites are contained in a local area network and the processing of each data entity is relatively simple and quick, it is not appropriate in an environment with rapidly developing heterogeneous networks, especially when geographically dispersed sites are connected by a wide area network (WAN) and used to provide webbased multimedia database services. In this paper, we have developed a stochastic reward net (SRN) model which incorporates network delay and site processing time in a distributed database system based on a static majority voting algorithm. The model allows us to find the determining performance factors under different system conditions. The numerical results show that even a small network delay or site processing time leads to a significant degradation in performance. Our model uses the fixed point iteration technique to avoid the state space largeness problem that plagues most previous ana-