Data morphing: an adaptive, cache-conscious storage technique

  • Authors:

  • Richard A. Hankins;Jignesh M. Patel

  • Affiliations:

  • University of Michigan, Ann Arbor, MI;University of Michigan, Ann Arbor, MI

  • Venue:
  • VLDB '03 Proceedings of the 29th international conference on Very large data bases - Volume 29
  • Year:
  • 2003

Quantified Score

Hi-index 0.04

Visualization

Abstract

The number of processor cache misses has a critical impact on the performance of DBMSs running on servers with large main-memory configurations. In turn, the cache utilization of database systems is highly dependent on the physical organization of the records in main-memory. A recently proposed storage model, called PAX, was shown to greatly improve the performance of sequential file-scan operations when compared to the commonly implemented N-ary storage model. However, the PAX storage model can also demonstrate poor cache utilization for other common operations, such as index scans. Under a workload of heterogenous database operations, neither the PAX storage model nor the N-ary storage model is optimal. In this paper, we propose a flexible data storage technique called Data Morphing. Using Data Morphing, a cache-efficient attribute layout, called a partition, is first determined through an analysis of the query workload. This partition is then used as a template for storing data in a cache-efficient way. We present two algorithms for computing partitions, and also present a versatile storage model that accommodates the dynamic reorganization of the attributes in a file. Finally, we experimentally demonstrate that the Data Morphing technique provides a significant performance improvement over both the traditional N-ary storage model and the PAX model.