Software Reliability
Software Reliability Growth Models 
1) DfRSoft has multiple methods for assessing software reliability
growth in testing. These can be viewed in two categories, 1) the
well known GoelOkumoto (GO) Model, 2)Cumulative Reliability Growth
Models. Each is described below:
GoelOkumoto (GO) Software
Reliability Growth Model
Often it is observed that testing
(and hence defect fixing) cycle gets elongated during product
development. Manager is asked questions like, “How long will testing go
on? How many defects are still remaining in software? How do we
predict reliability of current software? At such time, it is better to
use statistical methods to take informed decisions, rather than guessing
with past experience. The method described here is based on
NonHomogeneous Poisson Process (NHPP) GO model. It uses historical
sample test data to predict how many residual defects are there in the
software system and how many days are needed.
Cumulative Software Reliability Growth Models
DfRSoft offers a number of cumulative software reliability growth
models. These model are based on Cumulative Time vs. Cumulative Error
Data Fits. They include The Duane Model, Crow/AMSAA Model, The
Logarithmic Model, and a Polynomial Fit Model. Often the Logarithmic
model is the best chose. However, one should use the model that best
fits the data points and projected results. The key to remember is that
all the models are based on cumulative time rather than test time.
Therefore one should always keep this in mind in projecting model
results. Duane and Crow/AMSAAA are well known traditional reliability
growth models. They can also be used in software reliability growth. All
these models have advantages over the GO model above. The max errors in
the GO model is the value A. This limits the use of the model. These
models do not have this limit. They are more straight forward, being
basic regression and well understood. All models are based on data fits
of Cum Errors plotted against cumulative time. For the Duane and
Crow/AMSAA, the result is a power law fit to the data. Beta and
alpha=1Beta are considered the growth exponent for these models.
When is enough, enough for software reliability testing:
DfRSoft now offers a method of slopes test criteria to help the user in
assessing this.
