Please release me let me go,
No bugs are left here anymore.
To waste your time would be a sin.
Release me, don't test me once again.
The lectures deal with quantitative techniques for the determination of the reliability of a software system. They consider both commercial software packages, for which considerations about the time-to-market may dominate over the possibility of prolonging the testing phase, and safety-relevant software system, whose licensing must rely on a well-founded reliability assessment. With this distinction in mind the lectures present systematic procedures in order to derive reliability estimations on the basis representative samples. The theoretical introduction of the topics above is completed by reports on real-world industrial experiences.
- Random Test Data Generation
- Reliability Growth Models
- Testing of Reliability Hypotheses
Please release me can't you see,
You'd be a fool to cling to me.
There're no bugs left, just features, dear.
Release me, don't wait another year.
- terminology: correctness, reliability, availability, safety
- foundations of stochastic: probability, conditional probability, random variable
- foundations of reliability theory: survival probability, reliability function, fault and failure
- bathtub curve, failure rate, MTTF
2. Random Test Data Generation
- random testing vs. systematic testing, functional and structural testing
- random distribution, partition, seeding und tagging
- different contributions of individual faults to the total failure rate
3. Reliability Growth Models
- reliability prediction, estimation of failure rate
- parameter estimation methods: Maximum-Likelihood method, least squares
- models on the basis of inter-failure times, fault counting and Bayesian methods
- Jelinski & Moranda's model, Moranda's geometric model, Littlewood & Verrall's model
- models on the basis of cumulative failure numbers, finite / infinite fault models
- failure intensity, homogeneous / inhomogeneous Poisson process
- Musa Basic Model, Modell von Goel - Okumoto, Modell von Musa - Okumoto
- measuring predictive accuracy: bias, u-plot, Kolmogorov-Smirnov test
- noise, prequential likelihood function, y-plot, prediction recalibration
- combination of two models by Bayesian inference or prequential likelihood maximisation
- examples in practice: obtaining failure data, software tool CASRE 3.0, pro and cons
4. Testing of Reliability Hypotheses
- statistical testing, error probability
- sequential analysis
- determination of operational profiles: explicit and implicit profile
- generation of test cases from operational profiles