Clock Synchronization and Event Ordering

Message synchronization is key to distributed coordination. Each cohort of a distributed system runs an independent clock. The absence of a common clock and the variations of physical parameters of network links lead to non‐determinism in the arrivals of messages at destinations. Internal synchronization of local clocks suffices for event ordering within a cohort. External synchronization depends on sending and receiving messages. If a receive event predates the corresponding send event, the system state becomes inconsistent. Therefore, defining the clock and a consistent temporal ordering of events is essential for coordination. Christian's algorithm, Berkeley algorithm, and network time protocol were proposed for external synchronization clocks. Lamport described a logical clock and the “happened before” relation among events. Lamport's clock uses message timestamps to synchronize local clocks at cohorts. It captures causal dependencies among events. However, “happened before” fails to capture the causal dependence of messages. Vector clock is an extension of Lamport's clock. It captures causal dependence multicast messages. Causal order, FIFO order, and total order of multicast messages are essential for distributed applications. So, FIFO ordering and total ordering of multicast messages are possible. Further we talk about interval events and their relations in this chapter.