Distributed time-management in transputer networks

Abstract

For real-time applications in a distributed system a common notion of time is indispensable. Clocks are used for time measurement, determination of causality, process synchronization and generating unique identifications. All this is only possible if there is a time reference of specified accuracy. Since the local clocks in a distributed system tend to drift away from each other, they need to be adjusted periodically. If the application allows an accuracy that can be met by software, this may be achieved by a distributed clock synchronization algorithm, which creates and maintains a global time reference for all nodes of the network. The design and simulation of such an algorithm for a distributed system consisting of transputers is described. It is based on second order filtered adjustment of the clock rates rather than updating the clock values at once.<>