Dealing with uncertainties in the performance modelling of software systems

Abstract

Models play a central role in the assessment of software non-functional properties like performance and reliability. Models can be used both in the initial phases of development to support the designer decisions and at runtime to evaluate the impact of changes in the existing software. However, being abstraction, the models include per-se a certain degree of uncertainty. Nevertheless, often this aspect is neglected and models are used beyond their capabilities. Recognising the presence of uncertainties and managing them, would increase the level of trust in a given software model. In this paper we exploit a recently defined taxonomy that classifies the different types of uncertainties and we define a method that, starting from a given model, helps in recognising the existence of uncertainty, in classifying and managing it. We show the method at work on an example application considering the performance of the application as target non-functional property.