The context of this article is the availability of enterprise IT services, a key concern for many enterprises. While there is a plethora of literature concerned with service availability, there is no previous systematic empirical study on IT service time to recovery following outages. The existing literature typically assumes a distribution, or builds on analogies to related areas such as software engineering. Therefore, our objective is to find the statistical distribution of IT service time to recovery. Method-wise, this investigation is based on logs of more than 1 $\thinspace$ 800 incidents in a large Nordic bank, corresponding to more than 11 $\thinspace$ 000 hours of recorded downtime. Five possible distributions of time to recovery from the literature were investigated using the Akaike Information Criterion to find the distribution offering the best fit. The results show that the log-normal distribution outperformed the others for all tested service channels (collections of IT services). It is concluded that the log-normal distribution offers the best fit of IT service time to recovery. Using this distribution in simulation and decision-support tools offers the prospect of better predictions of downtime and downtime costs to the practitioner community.

The Distribution of Time to Recovery of Enterprise IT Services

Bayes inference for a nonhomogeneous Poisson process with an S-shaped mean value function is studied. In particular, the authors consider the model of Ohba et al. (1983), and its generalization to a class of gamma distribution growth curves. Two Gibbs sampling approaches are proposed to compute the Bayes estimates of the mean number of errors remaining and the current system reliability. One algorithm is a Metropolis within Gibbs algorithm, The other is a stochastic substitution algorithm with data augmentation. Model selection based on the posterior Bayes factor is studied. A numerical example with simulated data is given.

Bayes inference for S-shaped software-reliability growth models

Existence conditions are given for maximum likelihood parameter estimates for several commonly employed two-parameter software reliability models. For these models, the maximum likelihood equations can be expressed in terms of a single equation in one unknown. Bounds are given on solutions to these single equations problems to serve as initial intervals for search algorithms like bisection. Uniqueness of the solutions is established in some cases. Results are given for the case of grouped failure data. >

Solving maximum likelihood equations for two-parameter software reliability models using grouped data

A variety of Software Reliability Growth Models (SRGM) have been presented in literature. These models suffer many problems when handling various types of project. The reason is; the nature of each project makes it difficult to build a model which can generalize. In this paper we propose the use of Genetic Programming (GP) as an eVolutionary computation approach to handle the software reliability modeling problem. GP deals with one of the key issues in computer science which is called automatic programming. The goal of automatic programming is to create, in an automated way, a computer program that enables a computer to solve problems. GP will be used to build a SRGM which can predict accumulated faults during the software testing process. We evaluate the GP developed model and compare its performance with other common growth models from the literature. Our experiments results show that the proposed GP model is superior compared to Yamada S-Shaped, Generalized Poisson, NHPP and Schneidewind reliability models.

https://file.scirp.org/pdf/JSEA20110800002_91478098.pdf

A New Software Reliability Growth Model: Genetic-Programming-Based Approach

In order to assure software quality and assess software reliability, many software reliability growth models (SRGMs) have been proposed for estimation of reliability growth of products in the past three decades. In principle, two widely used methods for the parameter estimation of SRGMs are the maximum likelihood estimation (MLE) and the least squares estimation (LSE). However, the approach of these two estimations may impose some restrictions on SRGMs, such as the existence of derivatives from formulated models or the needs for complex calculation. Thus in this paper, we propose a modified genetic algorithm (MGA) to estimate the parameters of SRGMs. Experiments based on real software failure data are performed, and the results show that the proposed genetic algorithm is more effective and faster than traditional genetic algorithms.

A Study on the Applicability of Modified Genetic Algorithms for the Parameter Estimation of Software Reliability Modeling

Model selection based on a predictive performance measure is compared to model selection based on maximum likelihood. Both procedures exhibit unstable relative performance when predictive performance is measured over time periods that may contain overly small proportions of total failures. Both procedures may be stabilized by bounding this proportion away from zero. In that case, both procedures exhibit similar predictive performance relative to the other. This provides evidence that model selection based on the commonly used maximum likelihood approach may in fact have good predictive performance. >

J. Sacks

Papers

Software reliability model selection