Using Time to Add Order to Distributed Testing

Many systems interact with their environment at physically distributed interfaces called ports. In testing such a system we might use a distributed approach in which there is a separate tester at each port. If the testers do not synchronise during testing then we cannot always determine the relative order of events observed at different ports and corresponding implementation relations have been developed for distributed testing. One possible method for strengthening the implementation relation is for testers to synchronise through exchanging coordination messages but this requires sufficiently fast communications channels and can increase the cost of testing. This paper explores an alternative in which each tester has a local clock and timestamps its observations. If we know nothing about how the local clocks relate then this does not help while if the local clocks agree exactly then we can reconstruct the sequence of observations made. In practice, however, we are likely to be between these extremes: the local clocks will not agree exactly but we have assumptions regarding how they can differ. This paper explores several such assumptions and derives corresponding implementation relations.