Authors:
Thomas B. Puzak
and
Chun-Hsi Huang
Affiliation:
The University of Connecticut, United States
Keyword(s):
Cache Aware Binary Trees, Binary Tree Spatial Locality, Binary Tree Cache Performance.
Related
Ontology
Subjects/Areas/Topics:
Business Analytics
;
Data Engineering
;
Data Management and Quality
;
Data Semantics
;
Information Quality
Abstract:
The topological structure of binary search trees does not translate well into the linear nature of a computer’s memory system, resulting in high cache miss rates on data accesses. This paper analyzes the cache performance of search operations on several varieties of binary trees. Using uniform and nonuniform key distributions, the number of cache misses encountered per search is measured for Vanilla, AVL, and two types of Cache Aware Trees. Additionally, concrete measurements of the degree of spatial locality observed in the trees is provided. This allows the trees to be evaluated for situational merit, and for definitive explanations of their performance to be given. Results show that the balancing operations of AVL trees effectively negates any spatial locality gained through naive allocation schemes. Furthermore, for uniform input, this paper shows that large cache lines are only beneficial to trees that consider the cache’s line size in their allocation strategy. Results in the p
aper demonstrate that adaptive cache aware allocation schemes that approximate the key distribution of a tree have universally better performance than static systems that favor a particular key distribution.
(More)