On the Use of Synchronizers for Asynchronous Communication Networks

A synchronizer is a mechanism that helps simulate a synchronous communication network on an asynchronous one. In this paper, we draw attention to certain difficulties that one may face in the implementation of a synchronizer mechanism, if the model of computation assumed does not permit delaying the processing of messages once received. To illustrate the issues, we present the design of a simple, time-efficient synchronizer. Using a synchronous breadth-first-search tree protocol, it will be shown that this synchronizer will not work correctly in all cases, because the simulation may not be exact. Similar difficulties arise in the implementation of other synchronizers proposed earlier in the literature, if the messages of the synchronous algorithm do not carry the pulse number. It is easy to suggest remedies for these difficulties which are at variance with the model of computation assumed. But, if we must stay within this model, we show that only the time-inefficient β-synchronizer known earlier can be corrected. We also present simple modifications to ensure this.