A modular approach to real-time synchronization

Abstract

In this paper, we discuss real-time barrier synchronization as an example of timing constraints on synchronization. Real-time barrier synchronization is constrained by two timing values: earliest release time and release time skew. Earliest release time is the delay from the time when the last object participating in barrier synchronization issues a barrier operation, till when one of the participating objects resumes its method. Release time skew is the delay between the first and the last resumption times of objects. The implementation of real-time barrier synchronization consists of a negotiation phase and a scheduling phase. In the negotiation phase, all the participating objects communicate with one another and agree on the time when they are going to resume their method execution. Local scheduling at individual nodes must then ensure that the agreement is implemented. Thus, earliest release time and release time skew are defined by the procedures of the negotiation and the scheduling phase, respectively. We describe an architecture which enables the implementation of real-time barrier synchronization in a distributed system with unreliable channels.