Cache consistency management in mobile distributed environment

Abstract

In a mobile computing environment, database servers disseminate database items to multiple mobile clients via wireless channels. Due to the low bandwidth and low reliability of wireless channels, it is important for a mobile client to cache its frequently accessed database items into its local storage. This improves the performance of database queries and the availability of database items for query processing during disconnection. Since the conventional cache management mechanisms are not suitable in mobile environments due to their reliance on high bandwidth networks, a set of cache management mechanisms has been proposed to address this issue. There are three issues for consideration in the proposed management mechanism, namely, cache granularity, cache coherence strategy, and replacement scheme. Along the dimension of cache granularity, three levels of granularity are suggested, namely, attribute caching, object caching, and hybrid caching, which is a hybrid approach of attribute and object caching. Regarding cache coherence strategy, a client-pull based solution is proposed for mobile environment. Also we have proposed several cache replacement schemes that can adapt to the access patterns of database items. Via a detailed simulation model, we examine the performance of various levels of cache granularity with our cache coherence strategy. We observe, in general, that hybrid caching could achieve a better performance. In order to preserve the consistency of database, an optimistic concurrency control protocol has been incorporated into our caching model. However, in practice, it is hard to enforce one-copy serializability in mobile computing environment due to the narrow bandwidth and unreliable connectivity of wireless communication links. Instead, we have employed bounded ignorance as the correctness criterion in our model. Through simulations, it is observed that the system throughput is significantly enhanced when a controlled degree of inconsistency is allowed in read-only transactions especially when the update probability is low. Finally, we have implemented the proposed caching mechanism as a system prototype, and a set of experiments has been performed accordingly. The outcome of these experiments is in line with the performance observed in simulations.