An Empirical Model to Measure Dependability of a Software System

Abstract

Dependency on automation processes in science and engineering imparts the importance of user- confidence on software. This proposal has the objective to measure and provide a quantitative prediction of dependability for a software system with the help of a measuring model. The articulation of the model will be based on empirical findings on NASA test-bed at Mountain View, California. Dependability is a collection of several individual attributes like reliability, availability, safety, security along with confidentiality, integrity, maintainability and robustness. fu some of the related works, attempts were made to measure individually some of these attributes, e.g., reliability or security or availability of a software system. Unfortunately, there exists no such work that aims to measure and predict dependability of a software system as a whole in a quantitative manner. In the empirical approach that has been proposed here, each individual attributes that impacts on dependability of the whole system will be considered. The central theme of the work would be to utilize the NASA test beds to execute test cases on a wide range of domain-specific software systems and to use the test results to identify appropriate index ranges for each ofthe dependability factors. The idea ofindex ranges is that a software system remains dependable if the individual attributes for dependability are maintained within respective index ranges, once those are identified. The research team will design test cases to check for independence between attributes and then for finding the index ranges for each of the dependability attributes. The variation of dependability will be observed as each individual attributes will be varied one at a time and the researchers will record the comparative results. fu the next phase, the test results will be analyzed and the measurement model will be formulated that would produce an indicative index for dependability for a set of input values selected from safe index ranges for relevant dependability attributes. With this empirical model the research team will go back to use the NASA test bed once again to validate the model on new case studies. Absence of any standard and benchmarking makes it impossible at present to compare two software solutions of the same problem in terms of dependability. The proposed work would provide a much- needed tool for dependability, where we can make a priori estimation that if we decide to go for e.g., more availability, how it may impact on over all dependability ofthe software system? Apart from estimating dependability of existing software systems, the proposed work will also be useful and effective to provide a mechanism for designing a software system maintaining dependability in all through the development process. An a priori estimation of dependability after trading off the conflicting needs, would allow the designer to make a better feasibility study whether the end-product would meet all the functional needs as well as maintain the desired level of dependability. The success of the proposed work holds a lot of potential and promises to look ahead even before designing starts as far as dependability is concerned. This would ease the job of project management and reduce the cost of overall software development. The Software Engineering Automation Center of U. S. Naval Postgraduate School has an excellent research environment with research faculty members and several post-doc researchers leading research groups of varied interests comprising of more than 70 enrolled PhD students and about 200 Masters students. This would ensure the continual flow of fresh ideas that is an essential requirement to conduct this kind of experimental work.