The selection of appropriate test cases is an important issue for conformance testing of protocol implementations as well as in software engineering. A number of methods are known for the selection of a test suite based on the specification of the implementation under test, assumed to be given in the form of a finite state machine. This paper presents a new method which provides a logical link between several of the known methods. Called "partial W method", it has general applicability, full fault detection power, and yields shorter test suites than the W method. The second part of the paper discusses various other issues which have an impact on the selection of a suitable test suite. This includes the consideration of interaction parameters, various test architectures for protocol testing, and the fact that many specifications do not satisfy the assumptions made by most test selection methods, such as complete definition, a correctly implemented reset function, a limited number of states in the implementation, and determinism.

In order to test the control portion of communication software, specifications are usually fll"St abstracted to state machines, then test cases are generated from the resulting machines. The state machines obtained from the specification are often both partially-specified and nondeterministic, but no former work has been reported on test generation for such state machines. We come out with a method of generating test suites for the software that is modeled by partially-specified nondeterministic fmite state machines (PNFSM's). On the basis of intuitive notions, a conformance relation, called quasi-equivalence, is introduced for such machines, which serves as a guide to test generation. Our method is also applicable to completely-specified deterministic machines, partially-specified deterministic machines, and completely-specified nondeterministic machines, which are typical classes of PNFSM's. When applied to such classes of machines, this method yields not greater test suites with full fault coverage for each class of machines than the existing methods for the same class which also provide full fault coverage, when the number of states in implementation machines is bounded by a known integer. The test suites generated by the method can be used to check the conformance relation between a specification and its implementations.

Protocol testing for the purpose of certifying the implementation's adherence to the protocol specification can be done with a test architecture consisting of remote tester and local responder processes generating specific input stimuli, called test sequences, and observing the output produced by the implementation under test. It is possible to adapt test sequence generation techniques for finite state machines, such as transition tour, characterization, and checking sequence methods, to generate test sequences for protocols specified as incomplete finite state machines. For certain test sequences, the tester or responder processes are forced to consider the timing of an interaction in which they have not taken part; these test sequences are called nonsynchronizahle. The three test sequence generation algorithms are modified to obtain synchronizable test sequences. The checking of a given protocol for intrinsic synchronization problems is also discussed. Complexities of synchronizable test sequence generation algorithms are given and complete testing of a protocol is shown to he infeasible.To extend the applicability of the characterization and checking sequences, different methods are proposed to enhance the protocol specifications: special test input interactions are defined and a methodology is developed to complete the protocol specifications. P

No abstract

No abstract