Jacob Berlin and Amihai Motro
ISE-TR-00-03
Most virtual database systems are suitable for environments in which the set of member information sources is small and stable. Consequently, present virtual database systems do not scale up very well. The main reason is the complexity and cost of incorporating new information sources into the virtual database. In this paper we describe a system, called Autoplex, which uses machine learning techniques for automating the discovery of new content for virtual database systems. Autoplex assumes that several information sources have already been incorporated ("mapped") into the virtual database system by human experts (as done in standard virtual database systems). Autoplex learns the features of these examples. It then applies this knowledge to new candidate sources, trying to infer views that "resemble" the examples. In this paper we report initial results from the Autoplex project.
Jeff Offutt
ISE-TR-00-02
This report presents results for the Rockwell Collins Inc. sponsored project on generating test data from requirements/specifications, which started January 1, 1999. The purpose of this project is to improve our ability to test software that needs to be highly reliable by developing formal techniques for generating test cases from formal specificational descriptions of the software. Formal specifications represent a significant opportunity for testing because they precisely describe what functions the software is supposed to provide in a form that can be easily manipulated by automated means. This Phase III, 1999 report presents results from an empirical evaluation of the full predicate specification-based testing criterion developed during the first two phases of this project, and a proof-of-concept test data generation tool. The evaluation used a comparative study on a large industrial system, a research version of the Flight Guidance Mode Logic System (FGS) provided by Rockwell Collins. Full predicate tests were generated for FGS, and compared against the T-Vec generation scheme. T-Vec tests for FGS were also provided by Rockwell Collins. While creating and running the tests, one problem was found in the SCR specifications for FGS, and one problem was found in the already well tested implementation of FGS. Both T-Vec and the full predicate tests found almost the same number of faults, but T-Vec required more than five times as many tests, thus the full predicate tests were more efficient. The proof-of-concept test data generator creates full predicate and transition-pair tests from an SCR specification. Currently, the tool requires the SCR specification to be a single mode transition table, all variables must be boolean, and the transition predicates must be single-variable expressions.
Aynur Abdurazik
ISE-TR-00-01
Increasingly, very high level designs of large software systems are being described by software architectures. A software architecture expresses the overall structure of the system in an abstract, structured way. The Unified Modeling Language (UML) is widely used to express mid- and low-level designs of software, and recent proposals have been made to adapt the UML for use as an architecture design language (ADL). This research is looking into problems associated with creating system-level software tests that are based on architecture descriptions. This paper discusses issues with using the UML as an ADL, and general problems with then using the architecture descriptions as a basis for generating tests.