Software-Based Fault-Tolerant ClockSynchronization for Distributed UNIXEnvironments

| Fault-tolerant clock synchronization is often used in distributed systems with requirements such as close interaction between its components, measurements of elapsed time and ordering of events in the system. Diierent implementation approaches can be used to achieve fault-tolerant clock synchronization, depending on criteria such as performance, cost and availability of hardware and operating system support. This paper describes a low-cost, application level implementation that synchronizes the clocks of UNIX-based computers connected via an Ethernet local area network. The implementation is highly portable and modular, allowing the utilization of diierent fault-tolerant, interactive convergence clock synchronization algorithms. Experimental results obtained with the Fault-Tolerant Midpoint Algorithm (FTMA) 1] are presented and compared against typical results of hardware and operating system-based implementations. We show that a software implementation of the FTMA algorithm is sensitive to environmental aspects such as process priority, CPU and network load, and that it is possible to obtain a more stable algorithm by using a technique referred to as adaptive exponential averaging. Finally, we compare our fault-tolerant implementation with a synchronization protocol used in the Internet, namely the NTP (Network Time Protocol).