Software reliability: measurement, prediction, application
Software reliability: measurement, prediction, application
| Hi-index | 0.00 |
This talk discusses our experiences in applying Software Reliability Measurement (SRM) techniques as outlined in reference [1] to one of our software products. Our products are computer-based systems which mechanize the management and operations of telephone and data communications networks in AT&T Communications and the Regional Bell Operating Companies.We conducted a pilot effort with one of our products to assess how SRM techniques could be applied and the benefits of using such methods with our product line. The product selected had Software Reliability Objectives established for it. However, we quickly noted two deficiencies in our objectives when we tried to apply SRM techniques: 1). the lack of an “Operational Profile” establishing conditions under which our reliability objectives would be met, 2). the distinction between elapsed (calendar) time and execution (CPU) time. One of the first things we did was to base-line an “Operational Profile” of command activity reflecting anticipated usage by our customers and to relate calendar time to execution time (SRM techniques provided a useful framework for understanding how to do this). Then, we calibrated a software reliability model using existing project data. The model was used to develop a-priori estimates of testing effort needed to certify the software would meet a specified level of reliability. Test scripts based on the “Operational Profile” were set up and procedures were put in place to collect failure data during system test. Following several months of system testing our product, post-analysis was done on the collected failure data and compared to the a-priori estimates of test effort generated by the model. Estimates of initial failure rate (one of the model parameters) using actual failure data validated well with our a-priori estimates. Computed failure rates during latter stages of testing deviated from our model's predictions. Reasons for this deviation are discussed.Finally, this talk summarizes our problems and successes in applying SRM techniques and our continued use of such techniques within our product line.