Fast Distributed Mutual Exclusion

Abstract

A technique is proposed for quickly distributing mutexes in a system of multiple nodes. Evaluated in the context of a physical network, it is relatively agnostic with respect to the underlying topology and can be modified to work with more abstract arrangements of mutex repositories (e.g. distributed databases). To prove its usefulness, it is compared to a variety of other proposals for distributing mutual exclusion from the operating systems and database realms and found to provide significant performance and flexibility benefit. Additionally, a method is presented for applying the proposal to atomic operations in GPGPU applications to allow the GPGPU space to efficiently support a broader range of data parallel applications.