Modeling and Performance Evaluation

This activity concerns the assessment of performance, reliability and availability of computer and communication systems. Performance evaluation is a key step in design, analysis and tuning of such systems. The Performance Evaluation Group is particularly interested in Petri Net as modelling tool for performance Evaluation. Petri nets (PN) have been well-known for a long time as a useful tool for the evaluation of the systems performance, thanks to their intrinsic simplicity of description (which results from the graphical representation of the model), and to the ability of easily representing the occurrence time of events. While generalized stochastic Petri nets (GSPN) can describe events whose duration is represented by an exponential distribution of probability, a wide area of research deals with the study of PN in which the duration of an event can be described by any kind of distributions (non-Markovian Stochastic Petri nets - NMSPN). The GSPN introduces the concept of time, associating the events described in the model with a delay whose duration is described through a distribution of exponential probability. Since the exponential distributions enjoy the property of being memory-less (i.e. at any instant the future evolution of the process is not dependent from his past history), the analysis of the GSPN could be conducted through the solution of the continuous time Markov chains. A great number of real systems could be described through this class of models, but there are many models that have meaningfully different properties (e.g. not exponentially distributed delays or particular mechanisms of dependence from other events of the system). These systems are studied through models said non-Markovian. One of these classes of systems is constituted from the real-time systems that require the use of distributions with very narrow variance (like the deterministic or the uniform). To study NMSPN models, the WebSPN tool has been developed. WebSPN is a WEB accessible tool to solve stochastic Petri nets with no exponentially distributed firing time transitions. Actual activities of the group are:

  • Evaluation of the behaviour of wireless networks with particular interest to the internet access and energy consumption of mobile devices;
  • Distributed algorithms for the solutions of NMSPN: recently the developed algorithm has been implemented in WebSPN, so that it is available for running over a cluster of workstation; it is based on a message passing paradigm and it uses MPI library (the MPICH has been used);
  • Automatic generation and evaluation of performance models starting from a UML model; this research is focused on 2 directions: Software Development Process (SDP) and Software Performances Engineer (SPE). SDP studies, analyses and optimises the techniques and algorithms for software design. SPE is the systematic process for planning and evaluating system's performance throughout the life cycle of its development. In particular Petri Nets have been used to obtain an early performances model, derived from design phase, which has been analysed by our GSPN Tool Webspn.