Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
To find a configuration of a distributed system satisfying performance goals is a complex search problem that involves many design parameters, like hardware selection, job distribution and process configuration. Performance models are a powerful tools to analyse potential system configurations, however, their evaluation is expensive, such that only a limited number of possible configurations can be evaluated. In this paper we present a systematic method to find a satisfactory configuration with feasible effort, based on a two-step approach. First, using performance estimates a hardware configuration is determined and then the software configuration is incrementally optimized evaluating Layered Queueing Network models. We applied this method to the design of performant EDI converter systems in the financial domain, where increasing message volumes need to be handled due to the increasing importance of B2B interaction.