When data is replicated, an access protocol must be chosen to insure the presentation of a consistent view of the data. Protocols based on quorum consensus provide good availability with the added benefit of mutual exclusion. Of the protocols based on quorum consensus, the dynamic voting protocols provide the highest known availability. We describe a dynamic voting protocol that does not need the instantaneous state information required by the same performance as the original dynamic voting in the asymptotic case, and quickly converges to it for realistic access rates. Our protocol does this at a cost in network similar to that of statistic majority consensus voting. The first realistic analysis of the availability afforded by dynamic voting protocols is presented, taking the ac- cess frequency into account. The analysis confirms our hypothesis that delaying state information does not appreciably affect availability. Discrete event simulation is used to confirm and to extend the results we obtain using analytic models.

We present the first analytic study of stream tapping protocols, a family of protocols that provide the most efficient way to distribute videos on demand at low to medium request arrival rates, say, less than ten requests per hour for a two-hour video. The main results of this study are analytical solutions for the optimal operational points of stream tapping, stream tapping with small client buffers, stream tapping with partial preloading and stream tapping with proactive streams. In addition we introduce a new stream tapping protocol with batching that caps the bandwidth requirements of stream tapping at high to very high arrival rates. © 2006 IEEE.