Lock-Free Red-Black Trees Using CAS

The negative side-effects of using lock-based synchronization mechanisms are well known and include unexpected scheduling anomalies such as priority inversion and convoying as well as unnecessary synchronization overhead and the potential for deadlock. Despite these drawbacks, the use of lock-free techniques has not gained widespread acceptance. The reasons for this lack of acceptance include the complexity of hand-crafting lock-free algorithms for complex data structures and the relative inefficiency of the algorithms resulting from the use of generic methods for converting sequential algorithms into lock-free concurrent ones. Given these factors, we believe that the best approach is to create a library of efficient lock-free data structures that can be easily employed by others. For such a library to be successful, however, it must be based on synchronization primitives that are generally available on a wide range of machine architectures. In this paper, we describe the implementation of efficient lock-free algorithms for operating on a type of balanced binary search tree — red-black trees. Such tree structures are of potential importance in a wide range of search-based applications (including text mining and nucleotide sequence matching) that are being implemented on shared memory parallel machines. Our algorithms require only the widely available CAS (Compare And Swap) synchronization primitive.