A comparison of tree-balancing algorithms

Abstract

Several algorithms — height-balance (i.e. AVL and extensions), weight-balance (i.e. BB and WB), and total restructuring — for building balanced binary search trees are compared. The criteria for comparison encompass theoretical aspects (e.g. path lengths) and implementation independent and machine/algorithm-dependent measures (e.g. run time). A detailed analysis of code is also presented at a level believed to be language- and compiler-independent. The quality of the resulting trees and the overhead spent on building them are analyzed, and some guidelines are given for an efficient use of the methods. If insertion and subsequent queries are the only operations of interest, then “pure” AVL trees present the overall best qualities.