The increasing complexity of configurable software systems has created a need for more intelligent sampling mechanisms to detect and characterize failure-inducing dependencies between configurations. Prior work - in idealized environments - has shown that test schedules based on a mathematical object, called a covering array, in combination with classification techniques, can meet this need. Applying this approach in practice, however, is tricky because testing time and resource availability are unpredictable, and because failure characteristics can change from release to release. With current approaches developers must set a key covering array parameter (its strength) based on estimated release times and failure characterizations. This will influence the outcome of their results.

The increasing complexity of configurable software systems creates a need for more intelligent sampling mechanisms to detect and locate failure-inducing dependencies between configurations. Prior work shows that test schedules basedon a mathematical object, called a covering array, can beused to detect and locate failures in combination with a classification tree analysis. This paper addresses limitations ofthe earlier approach. First, the previous work requires developers to choose the covering array's strength, even thought here is no scientific or historical basis for doing so. Second, if a single covering array is insufficient to classify specificfailures, the entire process must be rerun from scratch. To address these issues, our new approach incrementally and adaptively builds covering array schedules. It begins witha low strength, and continually increases this as resources allow, or poor classification results require. At each stage, previous tests are reused. This allows failures due to only one or two configurations settings to be found and classified as early as possible, and also limits duplication of work when multiple covering arrays must be used.

There is an increasing interest in techniques that support analysis and measurement of fielded software systems. These techniques typically deploy numerous instrumented instances of a software system, collect execution data when the instances run in the field, and analyze the remotely collected data to better understand the system's in-the-field behavior. One common need for these techniques is the ability to distinguish execution outcomes (e.g., to collect only data corresponding to some behavior or to determine how often and under which condition a specific behavior occurs). Most current approaches, however, do not perform any kind of classification of remote executions and either focus on easily observable behaviors (e.g., crashes) or assume that outcomes' classifications are externally provided (e.g., by the users). To address the limitations of existing approaches, we have developed three techniques for automatically classifying execution data as belonging to one of several classes. In this paper, we introduce our techniques and apply them to the binary classification of passing and failing behaviors. Our three techniques impose different overheads on program instances and, thus, each is appropriate for different application scenarios. We performed several empirical studies to evaluate and refine our techniques and to investigate the trade-offs among them. Our results show that 1) the first technique can build very accurate models, but requires a complete set of execution data; 2) the second technique produces slightly less accurate models, but needs only a small fraction of the total execution data; and 3) the third technique allows for even further cost reductions by building the models incrementally, but requires some sequential ordering of the software instances' instrumentation.