Estimating and Predicting Error Detection Trends and Relative Manpower Utilization

Abstract

The work described in this paper outlines an investigative effort focusing on estimating the number of errors remaining in a software product and the effort required to detect those errors. More specifically, we provide a description of a computational model that, through a non-linear optimization process: (a) determines an error detection trend in a given development effort, (b) provides an estimate of the errors remaining in the product and the time required to detect those errors, and (c) relates the error detection effort to current and future manpower requirements. The model is simple to use, requiring only error counts and detection dates as input, and is grounded in the time-tested theory underlying Musa's Basic Reliability Model and the Rayleigh Manpower Utilization Curve. The model has been developed under the auspices of the Software Assurance Technology Center (SATC) at NASA/GSFC. Data sets from various NASA projects were used in formulating and tuning the model. For research purposes, the model was initially developed within an Excel framework; a more robust implementation is currently underway.