Estimating temporary files sizes in distributed realtional database systems

Abstract

The estimated sizes of temporary files are one of the most important statistics used by an Optimizer in generating a minimum cost processing strategy. Statistical information is the primary input to the estimation technique. The amount of statistics kept concerning the key attributes and the distribution of data values within a domain in a relation will greatly affect the accuracy of the estimates. However, the cost of storing this information may outweigh its value. Ultimately such a determination is left to those responsible for designing and implementing distributed realtional DBMSs. This paper presents a new method to calculate temporary files sizes. It also describes specific data structures and algorithms to implement the proposed method. The major tools suggested are a Log File and several specialized data matrices. The latter contain information unique to each relation. The Log Files is a posting files that records the number of occurrences of different values in the database.