Are CISQ Reliability Measures Practical? A Research Perspective
Recent research characterizes two types of program errors. The repeatable class, called Bohrbugs, always cause the same error with the same inputs. Prior thinking and reliability growth models were based on Bohrbugs, (Bbugs). Another class of errors called Mandelbugs, (Mbugs), are difficult to reproduce even with the same inputs, , . Mbugs may be an interaction of inputs with computer storage, where computer storage represents the past history of inputs to the computer system . A small fraction of Mbugs which increase with time, aging-related Mbugs, are ignored in the model. This paper formulates a software reliability growth model for predicting operational reliability of the software as development progresses. Based on prior experience, constant and exponentially decreasing error removal models are assumed. The inputs to the model are error removal data for both Bbugs and Mbugs, recorded during development testing and mean time to failure data from simulated operational tests. Simplified values for the model parameters are developed based on moment estimates. More sophisticated estimates using least squares and maximum likelihood techniques are discussed. The potential for improving the accuracy of reliability estimates by using models that incorporate both Bbugs and Mbugs is discussed with respect to two examples.