Goal-oriented memory allocation in database management systems
Kurt P. Brown · 1996
In order to meet the individual performance goals of each class in a complex multiclass database workload, today's database management systems require the adjustment of a number of low-level performance "knobs," such as buffer pool sizes, multiprogramming levels, data placement, dispatching priorities, etc. As the complexity of database systems is increasing, while their cost is declining at the same time, manually adjusting low-level DBMS performance knobs will become increasingly impractical. Ideally, the DBMS should simply accept per-class performance goals as inputs, and it should adjust its own low-level knobs in order to achieve them; this self-tuning capability is called goal-oriented resource allocation. This thesis makes three contributions in the area of goal-oriented resource allocation for database management systems. First, it defines an overall architecture for goal-oriented resource allocation that includes techniques to insure a stable and responsive system and to accu...