14.1. Introduction

We have seen several concurrent implementations of sets based on linked lists and on hash tables. We now turn our attention to concurrent search structures with logarithmic depth. There are many concurrent logarithmic search structures in the literature. Here, we are interested in search structures intended for in-memory data, as opposed to data residing on outside storage such as disks.

Many popular sequential search structures, such as red-black trees or AVL-trees, require periodic rebalancing to maintain the structure’s logarithmic depth. Rebalancing works well for sequential tree-based search structures, but for concurrent structures, rebalancing may cause bottlenecks and contention. Instead, we focus here on concurrent implementations of a proven data structure that provides expected logarithmic time search without the need to rebalance: the SkipList. In the following sections we present two SkipList implementations. The LazySkipList class is a lock-based implementation, while the LockFreeSkipList class is not. In both algorithms, the typically most frequent method, contains(), which searches for an item, is wait-free. These constructions follow the design patterns outlined earlier in Chapter 9.