Selectivity estimation of temporal data manipulations

Abstract

Temporal relations possess several characteristics that distinguish them from conventional snapshot relations. First, for each instance of the surrogate (entity) there is a set of time-ordered tuples. Second, surrogate instances may arrive and depart in some time-dependent manner. Third, the surrogate instance may arrive and depart more than once, thus creating gaps (null values) within its history. Lastly, the value of the temporal attribute may also be time-dependent. Conventional methods of estimation are incapable of providing good approximations of the cost of various temporal operations, even for those involving selections on a single relation. The problem is more acute in the case of join operations, because selectivities on time interval intersections have to be estimated. We propose a practical, yet theoretically sound model to characterize the changes of temporal relations. From this model, estimates of the cardinalities of various unary and binary operations are derived. Simulation results show that the proposed estimates are both robust and superior to conventional estimates.