Architectures for testing distributed systems

Stabilizing network infrastructures has moved the focus of software system engineering to the development of distributed applications running on top of the network. The complexity of distributed systems and their inherent concurrency pose high requirements on their design and implementation. This is also true for the validation of the systems, in particular the test. Compared to protocol testing the test of distributed systems and applications requires different methods for deriving test cases and for running the test. In this paper, we discuss architectures for testing distributed, concurrent systems. We suggest three different models: a global tester that has total control over the distributed system under test (SUT) and, more interestingly, two types of distributed testers comprising several concurrent tester components. The test architectures rely on a grey-box testing approach that allows to observe internal interactions of the SUT by the tester. In order to assure a correct assessment of the test data collected by the distributed tester components, the tester has to maintain a correct global view on the SUT. This can be achieved either by the use of redundant points of control and observation or by test coordination procedures. We outline the features of each approach and discuss their benefits and shortages. Finally, we describe possible simplifications for the architectures.