Using a waiting protocol to separate concerns in the mutual exclusion problem

Abstract

How to implement process synchronization in a general-purpose software library while incurring a minimum of policy decisions on the system as a whole? We propose that in dealing with the problem of mutual exclusion in concurrent systems, a separation of concerns between the mechanism of detecting contention and the policy decision of what to do when such contention is detected is appropriate. We show with code how a waiting protocol can guide and aid the design and implementation of both lower-level primitives and the higherlevel, policy-making parts of a system. We also show how the technique can be extended so as to help preventing some deadlock situations.