Specification and Constant RMR Algorithm for Phase-Fair Reader-Writer Lock

Abstract

Brandenburg and Anderson [1,2] recently introduced a phase-fair readers/writers lock [1,2], where read and write phases alternate: when the writer leaves the CS, any waiting reader will enter the CS before the next writer enters the CS; similarly, if a reader is in the CS and a writer is waiting, any new reader that now enters the Try section will not enter the CS before some writer enters the CS. Thus, neither class of processes–readers or writer–has priority over the others, and no process starves.

Brandenburg and Anderson [1,2] informally specify a phase fair lock and present an algorithm to implement it with O(n) remote memory reference complexity (RMR), where n is the number of processes in the system. In this work we give a rigorous specification of a phase fair lock and present an algorithm that implements it with O(1) RMR complexity.