Convergence of Time Decay for Event Weights

Events of a distributed system are partially ordered, but additional ordering information can aid in understanding the causal impact of events on other events. The assignment of weights to events measures the temporal proximity of events and quantitates the causality among events. In particular, weights have been used in the feature location problem to determine the relevance of events to the feature. Weights propagate to other processes following the Lamport causal ordering, with an applied decay representing their time-distance from the point of certain relevance. The focus of this paper is proving the decreasing property of weights as the time-distance from preceding events increases. Specifically, we prove that a weight of zero is reached despite intermediate weight increases that may incur from interprocess communication. An upper bound is placed on the elapsed time required before an event is assigned a weight of zero.