Simultaneous Resolution of Buffer Allocation and Query Scheduling Problems

Abstract

Database management systems (DBMS) use external magnetic devices (disks) for the storage of mass data. In the context of very large databases, the query response time may be strongly influenced by two main factors: (i) the amount data needed to be accessed from the disk to the main memory cache and (ii) the order of queries, especially if queries share some common intermediate results. These two factors are related to two well known problems in databases: Buffer Management Problem (BMP) and Query Scheduling Problem (QSP). Usually, BMP and QSP are treated in the isolated way and their strong interaction is ignored by the existing solutions. This situation motivates us to consider the integrated problem including BMP and QSP. In this paper, we first propose a formalization of the integrated problem and show complexity. Secondly, three heuristics are proposed: naive, hill climbing and genetic algorithms dealing with our integrated problem. Finally, two intensive experiments are conducted: (1) theoretically using a cost model and (2) a real validation on Oracle11G.