Software control flow error detection and correlation with system performance deviation

• We provide a control flow error detection approach that decreases the false negatives and allows complete coverage of program connections.
• We determine the correlation of the component behavior and the error state parameters with system performance.
• We evaluate the approach by using a number of fault-injected versions of the PostgreSQL database and demonstrate the detection accuracy.
• We also illustrate its significant ability to identify the components and error state patterns responsible for performance deviation.

Detecting runtime errors helps avoid the cost of failures and enables systems to perform corrective actions prior to failure occurrences. Control flow errors are major impairments of system dependability during component interactions. Existing control flow monitors are susceptible to false negatives due to possible inaccuracies of the underlying control flow representations. Moreover, avoiding performance overhead and program modifications are major challenges in these monitoring techniques. In this paper, we construct a connection-based signature approach for detecting errors among component interactions. We analyze the monitored system performance and examine the relationship of the captured error state parameters with the system performance deviation. Using the PostgreSQL 8.4.4 open-source database system with randomly injected errors, the experimental evaluation results show a decrease in false negatives using our approach relative to the existing techniques. It also demonstrates a significant ability of identifying the responsible components and error state patterns for system performance deviation.