A Universal Construction for Wait-Free Transaction Friendly Data Structures

摘要

Given the sequential implementation of any data structure, we show how to obtain an efficient, wait-free implementation of that data structure shared by any fixed number of processes using only shared registers and CAS objects. Our universal construction is transaction friendly, allowing a process to gracefully exit from an operation that it wanted to perform, and it is cache-efficient in a multicore setting where the processes run on cores that share a single cache. We also present an optimized shared queue based on this method.