MR-Tree: a cache-conscious main memory spatial index structure for mobile GIS

  • Authors:

  • Kyung-Chang Kim;Suk-Woo Yun

  • Affiliations:

  • Dept. of Computer Engineering, Hongik University, Seoul, Korea;Dept. of Computer Engineering, Hongik University, Seoul, Korea

  • Venue:
  • W2GIS'04 Proceedings of the 4th international conference on Web and Wireless Geographical Information Systems
  • Year:
  • 2004

Quantified Score

Hi-index 0.00

Visualization

Abstract

As random access memory chips get cheaper, it becomes affordable to realize main memory-based database systems. The most important issues in main memory database indexing techniques for mobile GIS applications are to improve update performances – since numerous update operations can arise for tracking continuously moving objects – and to reduce cache misses for improving search performances. In this paper, we propose MR-tree, a cache-conscious version of the R-tree for main memory databases. To increase fan-out, the MR-tree applies a novel compression scheme to entry MBRs (Minimum Bounding Rectangles). This scheme represents entry MBRs by relative coordinates in a node. To improve update performance, the MR-tree can become an unbalanced tree as follows: it propagates node splits upward only if one of the internal nodes on the insertion path has empty space, or all height differences between the consecutive nodes on the insertion path are 1 and one of the height differences among its subtrees is equal to some unbalance parameter. Because of this feature, the MR-tree can reduce the number of internal nodes splits and reinsertions significantly. In the case where all internal nodes on the insertion path do not meet the above conditions, a newly created leaf node simply becomes the child of the split leaf, when a leaf node split occurs. This split leaf is called the half-leaf node. Our experimental result shows that the search speed of the proposed two-dimensional MR-tree increases by almost a factor of two compared to the CR-tree and the R-tree variant, which are also main memory based R-trees, while maintaining better update performance.