Showing papers by "Christian S. Jensen published in 1990"

Replication Gives High Performance Query Processing in Relational Models Extended with Transaction Time

Abstract: Our research project can be described as follows. We have defined a static rollback relational data model with tuple time-stamping [5, 6]. For each relation defined by the user, the system maintains a backlog that records all requests for changes to its user-defined relation. The backlogs contain all the data necessary for answering any query without replication. We have designed an extended query language that allows access to previous states of relations (including backlogs) and allows to ask queries about the change history of relations. We also have developed vacuuming facilities to control otherwise ever-growing amounts of data in a transaction time database [7]. To support queries efficiently, we maintain a cache of views stored as either actual data (true replication) or as pointer structures (pseudo replication). A special cache index is used for fast identification of cache entries potentially useful as outsets in differential (i.e. incremental or decremental) computations of queries. The cache index is augmented with statistics about the cache entries. We use state transition networks for enumerating query plans. The number of alternative query plans is reduced by pruning rules, and plan selection can be done using the statistics of the cache index in conjunction with either dynamic programming techniques or heuristic techniques (e.g. the A∗ algorithm). Our framework is the first to integrate pointer/data storage and incremental/decremental/recomputation.