Browse Theses

Mediator-based architectures are used successfully in applications where transparent access to heterogeneous information sources is needed. However, one major problem with the use of these architectures is the long query response times. This problem is particularly serious for mobile and wireless systems, where disconnections and expensive connections make it either unfeasible or unsatisfactory to use classical mediator systems. In this paper, we present different caching strategies for mediators cater to different information needs. The first is for mediators that offer the possibility to ask decision-support queries, which are rather complex and use aggregate functions. We provide an integrated lattice framework for all of the aggregate functions in a relational environment. Then, we introduce an improved greedy algorithm for choosing views to materialize. Second, we are looking at mediators answering queries on information sources with clearly identifiable hot spots, that is, small areas of the database that are queried frequently. We propose semantic-region caching approaches with different storage granularities for such mediators accessing relational databases. In contrast to existing approaches, we cache not only the projection result of queries but also the condition attributes, resulting in a higher cache hit rate. Additionally, we introduce a sophisticated cache replacement algorithm, PRAG (Profit-based Replacement algorithm with Aging Counter), which incorporates the semantic notion of locality into the system. We also provide solutions for wireless system that retrieve partial result from the cache when the remote information source is disconnected.