Performance analysis at the software architectural design level

In this paper we present an approach to evaluating the expected performance of a software architecture (SA). The method allows the derivation of a performance evaluation model, based on a Queuing Network Model, from a SA formal specification modeled as a Labeled Transition System (LTS). The goal of this approach is to provide a set of measures to compare the performance of two or more competing SAs, even at their high level of abstraction. This means that, differently to most approaches in software performance, we do not assume to have, or to expect to obtain, further information on the system under development from the subsequent development steps, e.g. design, implementation and deployment. Rather we stick at the level of abstraction of the SA and even in the absence of relevant pieces of information we want to be able to carry on an evaluation of concurrent execution of SA which should allow the validation of the possibly critical design choices made at the SA level. We aim at deriving mathematical relations among the parameters characterizing the competing models. These relations are obtained by using implementation scenarios to relate the performance indices which can then be suitably evaluated. Besides assisting in the choice of an SA, these relations give information on how to carry on the development process in order to maintain the given performance. To a certain extent, the information obtained by means of the evaluation provides further requirements to be fulfilled by an implementation in order to satisfy the chosen performance criteria. This last step suggests the possibility of developing performance models at the SA design level that can then be used to derive the performance models related to further development steps. (C) 2001 Elsevier Science B.V. All rights reserved.