The Research On Client Semantic Caching In Mobile Database

With the coming of Information Age, mankind's desire for all sorts of information has increased tremendously. A Mobile environment provides the possibility for people to access information anywhere anytime. Hence it is becoming an important component of the future computing environment. With the widespread use of portable devices and the rapidly expanding wireless communication infrastructure, the mobile computing environment is maturing. But considering the resource-poor mobile devices and networks, the mobility of devices and the variance of network connections, information access in mobile environments faces many problems and challenges.Traditional client-server database systems based on "query-shipping" architecture rely totally on processing at server sides, so can't perform well in a mobile wireless network which characterizes itself as low bandwidth, high expense and poor reliability. Client caching provides a solution for ensuring data accessing performance in mobile environments. But most research in caching field has assumed traditional environments, meaning fixed hosts and networks. What's more, research on client caching that supports associative access also requires more investigation. Therefore, this thesis's purpose is to improve the performance of associative database access in mobile environments through work on client caching.The thesis proposes a mobile client caching scheme based on semantic caching, which explores the semantic locality among client issued queries. The cache is dynamically loaded with former query results, and is organized by cache item descriptions corresponding to query statements. Later queries are answered by cache items via cache description matching. If only parts of the query can be obtained from the cache, then the original query is trimmed and the trimmed part is processed on the server. In this way, associative database access is supported, and system performance and data availability during disconnection improve. Based on this caching scheme, the thesis investigates thoroughly key issues on cache replacement strategy, cache coherency policy, and query processing algorithms.To support the mobile client cache, the thesis first presents a hybrid client-server database architecture. Then the main idea behind the caching scheme is given, which is followed by the definition of cache admissible query. The content and organization of cache and the combination strategy of cache items are also discussed. Then the structure and the function of each component of the cache system are shown.Client semantic caching is different from traditional object caching in granularity and query processing. To get better cache hits in a semantic cache, a cache replacement strategy, called LWI(Least Weight Item), is proposed. LWI replacement strategy replaces the lowest weight cache item, and the weight is determined by the following three factors: access frequencies of the cache item's projected attributes, access probability of the tuples satisfying cache item predicate on each projected attribute, and latest access time of the cache item. Simulation results show that LWI replacementstrategy outperforms LRU and LFU replacement strategy in semantic cache.After qualitatively and quantitatively analysing several classic cache coherency policies, the thesis designs a cache coherency policy CCPMD for mobile environments. This policy is novel in the sense that it adopts befitting policies for different network connections with different optimization aims, and provides support for cache coherence upon transition between different situations. To support cache item view, the policy uses incremental view maintenance. Because of the high cost of maintaining cache coherency and the resource-poor characteristic of mobile environments, the thesis also discusses cache weak coherency. Several ways to specify allowable divergence are given. For clients with weak enforcement of deviation, a cache coherency policy based on period_of_validity is presented.Query cache hit is decided by c...