Improved and Optimized Partitioning Techniques in Database Query Processing

Abstract. In this paper we present two improvements to the partitioning process: 1) A new dynamic bu er management strategy is employed to increase the average block size of I/O-transfers to temporary les, and 2) An optimal switching between three di erent variants of the partitioning methods that ensures minimal partitioning cost. The expected performance gain resulting from the new management strategy is about 30% for a reasonable resource con guration. The performance gain decreases with increasing available bu er space. The di erent partitioning strategies (partial partitioning or hybrid hashing, one pass partitioning, and multipass partitioning) are analyzed, and we present the optimal working range for these, as a function of operand volume and available memory.

• Type: Conference
• Year: 1997
• Where: BNCOD
• Authors: Kjell Bratbergsengen, Kjetil Nørvåg
• Keywords: BNCOD 1997, BNCOD 2007, Di Erent Partitioning, Ensures Minimal Partitioning, Partitioning Process