Correcting remote references to a server in distributed operating systems

Protocol mechanisms are described to locate and correct stale references to a server. In a first round of correction, the protocols do not strive to correct the references atomically, i.e., clients can use the references before correction is completed. This is because locating all references may be difficult and cause unbounded delay since they may be deeply buried in the distributed computations (e.g., a client may have passed a reference to another client). A stale reference which misses the correction is detected upon use and then corrected or handled in an application dependent manner. Such an approach to correction protocols makes them simple, efficient and suitable for different environments. For example, in a large network, the server may notify a correction to the clients by datagram multicast with no guarantee of message delivery. Performance models and supporting data are provided from an implementation. Potential uses of this research are in X500 based implementation of servers