Study On A Three-Tier-Based Semantic Cache Technology

Nowadays, with large quantities of data stored in database server and data accessed and analyzed by thousands of users, an effective way to release the heavy load of the database and network is to take use of cache technology through which the data are cached in the local system and queries are answered by using the cache data.Among all the cache technologies, the semantic cache technology manifests distinctive advantages, through which the data and together with their semantic descriptions are cached. The semantic cache consists of disjoint semantic regions, and each semantic region includes a description and a corresponding data pointer. In semantic cache technology environment, at first, a query is splitted into two disjoint queries, probe query and remainder query. And then the probe query is answered by local cache system and the remainder query is answered by the background database server. Finally the two parts of answers are merged into one final result.As a great improvement over semantic cache technology based on client/server system, TSCT (Three-Tier-Based Semantic Cache Technology), combines the semantic cache technology with the popular three-tier architecture, greatly improves the effect of cache system, through which the query result is cached into both the client and the application server and the cache data are fully used to answer queries.TSCT is fit for the applications of both OLAP and OLTP, and is also highly scalable. It can be easily adapted to the application of huge data and the application environment of high parallelism with the number of application servers increased. TSCT takes use of the PAX (Partition Attribute Across) page schema to organize the cache data, in which all the data belonging to a field are physically stored together. In this way, the predicate evaluation and the performance of queries involving few fields of a relation are bettered, and the performance of the cache between CPU (Central Processing Unit) and the main memory is also improved. TSCT pre-selects, computes and stores some aggregation views, and uses these views to answer queries in order to avoid the on-line computation and speed up the answering of aggregation queries. The pre-selection of aggregation views takes into account both the size and the querying frequency of the aggregation views. The computation of aggregation views is implemented through the efficient BUC algorithm, which partitions the fact table in a bottom-up way and computes the whole selected aggregation views at one time. TSCT takes use of different strategies to process usual queries and aggregation queries. As for the former, TSCT splits the query exactly into the probe query and the remainder query according to different match types. As for the latter, TSCT selects an appropriate aggregation view to answer the query according to the computation dependency among aggregation views.