PHYSICAL RECORD CLUSTERING IN DATABASES

Abstract

The NP‐complete problem of optimally placing tuples on a hierarchy of secondary storage devices is considered using a heuristic approach. From load specification details captured at database design time, those tuples associated with queries which merit tailored, “set‐in‐concrete”, physical access paths are placed using a two‐level graph partitioning algorithm. Experiments are reported with the pages and cylinders as the two hierarchical levels of storage for a centralised database, but the technique is applicable to an n‐Ievel storage hierarchy—as up to the “different sites” level for distributed databases. The results show up to 39% improvement over single‐level partitioning algorithms for the database considered.