Optimizing Maintenance of Constraint-Based Database Caches

Caching data reduces user-perceived latency and often enhances availability in case of server crashes or network failures. Database caching aims at local processing of specific declarative queries in a DBMSmanaged cache close to the application. Query evaluation must produce the same results as if done at the remote database backend, which implies that all data records needed to process such a query must be present and controlled by the cache, i. e., to achieve “predicate-specific” loading and unloading of such record sets. Using cache constraints, the cache manager applies maintenance operations to appropriate record sets, called caching unit. In this way, the cache manager can, at any point in time, guarantee “predicate completeness” of the caching units currently present. We explore how cache groups can be maintained to provide the data currently needed. For various cache configurations, we design and optimize loading and unloading algorithms for sets of records keeping the caching units complete. Furthermore, we empirically identify the costs involved for this kind of cache maintenance.