Overcoming controllability problems in distributed testing from an input output transition system

Abstract

This paper concerns the testing of a system with physically distributed interfaces, called ports, at which it interacts with its environment. We place a tester at each port and the tester at port p observes events at p only. This can lead to controllability problems, where the observations made by the tester at a port p are not sufficient for it to be able to know when to send an input. It is known that there are test objectives, such as executing a particular transition, that cannot be achieved if we restrict attention to test cases that have no controllability problems. This has led to interest in schemes where the testers at the individual ports send coordination messages to one another through an external communications network in order to overcome controllability problems. However, such approaches have largely been studied in the context of testing from a deterministic finite state machine. This paper investigates the use of coordination messages to overcome controllability problems when testing from an input output transition system and gives an algorithm for introducing sufficient messages. It also proves that the problem of minimising the number of coordination messages used is NP-hard.

Towards Decentralized Conformance Checking in Model-Based Testing of Distributed Systems

“…2) messages can only be received after being sent, being the transmission time non-deterministic and the lifelines input-enabled (that is, always able to receive inputs, even if they are internally stored for later processing); 3) the possibility of a lifeline remaining quiescent (i.e., not sending output without first receiving input [9]) is also a function only of the sequence of previous event occurrences in the lifeline. The first two parts of function unintendedTraces in Figure 9 determines erroneous global subtraces, with all events valid except the last one, that can be generated if each lifeline, knowing the set of traces valid locally, behaves according to the causality rules 1 and 2 above.…”

Using schedulers to test probabilistic distributed systems

“…However, as explained earlier, in distributed testing we can only bring together the observations at the separate ports when the system is quiescent since we need to know that the local traces observed are projections of the same global trace. As a result, often we will not be able to distinguish between processes such as r and s. Alternative approaches consist in synchronising the actions of the testers through the exchange of coordination messages [CR99, RC03,Hie12] or in having the individual testers exchanging messages with a central coordinator that records the global order of events and determines when an individual tester should send an input [JJKV98]. While these approaches can be extremely useful, they can complicate testing and are not always appropriate.…”

Conformance Relations for Distributed Testing Based on CSP

“…It has been shown that, when testing from an IOTS, coordination messages can be used to allow global traces to be observed, but this requires several coordination messages for each event [12]. Recent work has shown how fewer messages can be added to a test sequence [5] to overcome controllability problems. It is this approach that we adapt.…”

“…The approach presented here is inspired by that in Hierons [5]. First, we introduce coordination messages as events coord .i.j observable by users i and j .…”

“…The approach previously devised for CSP [3] can be adapted, but the resulting test cases need not be sound. We then show how the use of coordination messages suggested by Hierons in [5] can be adapted to CSP to produce sound test cases that establish traces refinement.…”