Categorizing and predicting errors in software programs
Kenneth S. Mendis, M. GOLLIS · 1979
The lOAl Operational Program was formally This paper investigates the application of empirical software released to Software Configuration Management in Feberror dita-as applied to failure trend analysis and residual error ruary 1977. Since release, it has undergone four prediction. Ths paper: 1) identifies the types and frequency of reassemblies resulting from errors identified in the software errors experienced; 2) makes a comparison between errors experienced in two designs; 3) details a model for predict- program. During the reporting period a of ing the Mean Time to Failure (MTTF) of software; and 4) recom- Investigation Requests (IRs) reporting software problems mends corrective action as to how to reduce and/or limit errors were accumulated; 107 of these IRs resulted in software in future development efforts. changes. The data, when analyzed, shows a steady but gradual decrease in the number of software errors as the program enters the maintenance phase of its life cycle. The data points to the fact that there is a significant difference in the number of program errors experienced in the two designs studied. Of those errors identified, Raytheon Design Changes (RDCs) and Software Design Errors (SDEs) accounted for the largest percentage of the experienced errors; i.e., design changes were the single most critical factor that contributed to software errors. By means of error estimation, a total of 16 residual errors is estimated in the operational program. The data further identified that the majority of those errors are logical software design errors. Also discussed is a simplified approach to determining, within specified confidence levels, the MTTF of the program. The 10A2 Program was released to Configuration Management in May 1977. The PV portion of the program was added in J uly 1977. This activity doubled the program size and necessitated redefining its baseline. Subsequent reassemblies were needed to correct software e rrors. A total of three updates to t he original released baselined program has been made to date. During the reporting period a total of 32 IRs were accumulated, 28 result in design changes to the 10A2 Program.