The performance of concurrent red-black tree algorithms

Relaxed balancing has become a commonly used concept in the design of concurrent search tree algorithms. The idea of relaxed balancing is to uncouple the rebalancing from the updating in order to speed up the update operations and to allow a high degree of concurrency. Many different relaxed balancing algorithms have been proposed, especially for red-black trees and AVL trees, but their performance in concurrent environments is not yet well understood. This paper presents an experimental comparison of three relaxed balancing algorithms for red-black trees. Using the simulation of a multi processor environment we study the performance of chromatic trees, the algorithm that is got by applying the general method of how to make strict balancing schemes relaxed to red-black trees, and the relaxed red-black tree. Furthermore, we compare the relaxed balancing algorithms with the standard red-black tree, i.e. the strictly balanced red-black tree combined with the locking scheme of Ellis. We find that in a concurrent environment the relaxed balancing algorithms have a significant advantage over the strictly balanced red-black tree. With regard to the average response time of the dictionary operations the three relaxed balancing algorithms perform almost identically. Regarding the quality of the rebalancing we find that the chromatic tree has the best performance. (orig.)SIGLEAvailable from TIB Hannover: RR 5165(115) / FIZ - Fachinformationszzentrum Karlsruhe / TIB - Technische InformationsbibliothekDEGerman