Transformation-based temporal aggregation using order-based buffer replacement strategy

A temporal database is essential to applications dealing with time-varying data [11], such as trend analysis and prediction in decision support systems, version management in computer aided designs, medical record management, and data warehousing. In this regard, a temporal database system is used to store and query time-varying information. There has been extensive research on temporal databases [25]. The early research focused mainly on conceptual problems like data models and query languages. An example is the temporal SQL(TSQL)[19], which extends the relational data model and the query language SQL with temporal features. Subsequently, the research shifted toward system implementation problems like query processing in which query operations have been extended to support temporal data models and queries. Among various query operations, temporal aggregation is particularly important due to the high processing cost [13]. Temporal aggregation is an operation for finding the aggregate value of an attribute over a certain period of time. Specifically, it finds the time intervals in which the aggregate value does not change, namely the constant intervals [13], and performs the aggregation in each constant interval. There are typically two kinds of aggregate functions: cumulative (e.g. COUNT, SUM, AVG) and selective (e.g. MIN, MAX). Aggregation is an expensive operation, and temporal aggregation is particularly more so because it deals with very large data accumulated over time and has to calculate aggregate values separately for multiple time intervals. There have been several temporal aggregation methods proposed to date. The early ones include the aggregation tree method [13] and its variants [24, 8, 12, 15]. Although these methods do facilitate calculating temporal aggregates, they require one aggregation tree for each aggregate function. Moreover, they require the tree structures to reside in main memory. The trees, however, are typically much larger than the available main memory because a temporal database