Prefetching—a performance analysis

摘要

Reduction of page faults and thus, response times, is one of the keys to overall improvement of database access behavior. Assuming a transaction load of predominantly schematic or canned transactions, exploitation of preknowledge of transaction access patterns leads to the concept of prefetching, i.e. loading data pages into the system buffer in advance and before a time-critical access request is issued. Thus, the time behavior of a main memory database can be achieved. The justification of prefetching is only given by satisfying the time restriction in real-time application. This paper presents the concept and an experimental validation clearly identifying potential gains.