Model-Driven Web Engineering Performance Prediction With Layered Queue Networks Alessio Gambi 1 , Giovanni Toffetti 1 , and Sara Comai 2 1 University of Lugano 2 Politecnico di Milano 6904, Lugano, Switzerland 20133, Milan, Italy Abstract. This position paper describes an approach to predict the performances of a Web application already in the early stages of ap- plication development. It leverages the wealth of information of MDWE solutions to automatically obtain accurate representations of the running application in terms of layered queue networks (LQNs), i.e., analytical models simulating the behavior of the system and computing the perfor- mances mathematically. In particular, the paper discusses how a MDWE methodology can be exploited to generate such performance models and presents a proof of concept example. 1 Introduction Being able to correctly size the resources needed by a Web application to its incoming workload is the ultimate goal of performance engineering [4, 6]. For a company, both over- and under-provisioning turn into revenue loss: in the former case because of higher than needed operational costs, in the latter case because the Quality of Service is below clients’ expectations. The current practice in performance engineering impacts on different phases of a Web application development cycle: early in the process, coarse high level specifications are used at design-time to build mathematical models (e.g., queue networks) to estimate system performances; as the application reaches a complete implementation, staged experiments provide numeric parameters for the performance models or direct measures of the complete system behaviour. Updating the mathematical models to reflect the design choices and implementation details that happen along the development cycle is a costly manual process and very seldom done, as a result performance models tend to be misaligned with the final system and their predictions are less accurate. In this paper we propose an approach to leverage the wealth of information of MDWE solutions to automatically obtain accurate representations of the running applications in terms of layered queue networks (LQNs). The advantages of this approach with respect to the common practice are manifold: 1) formal models allow for the generation of LQNs at different levels of granularity and detail, from very coarse grained all the way to the actual code produced by model transformation, 2) they do not require manual intervention to be kept in synch with the development process, 3) any model update can be reflected directly