Modeling and schedulability analysis of AFDX networks in MAST 2

This short paper reports an overview of MAST (Modeling and Analysis Suite for real-Time applications) [1], and how AFDX networks will be modeled and analyzed within this framework. MAST defines a model for describing the timing behavior of distributed real-time systems and also includes a set of tools for schedulability analysis, assigning scheduling parameters and performing sensitivity analysis. It is developed by the Computers and Real-Time Group at the University of Cantabria, and it has been conceived for research purposes as a long term project led by Michael González Harbour. The objective of this project is to propose an open model as a basis to deal with new needs for timing behavior modeling and as a work bench for future timing analysis and optimization techniques. The MAST model [2] is now aligned with the OMG MARTE standard [3], especially with the SAM profile (Schedulability Analysis Modeling). MAST defines a high-level model mainly consisting of the following basic elements: Execution Platform (CPUs and communication networks), Schedulable Resources or Scheduling Servers (tasks or message streams), Operations (code blocks or messages), Mutual Exclusion Resources or Shared Resources (resources that must be used in a mutually exclusive way), and End-to-End Flows (key elements that will be described later). It also has a rich overhead model for other elements such as Timers or Network Drivers, and also a high expressiveness for timing requirements. This high-level model is transformed into analysis or simulation models over which schedulability analysis techniques or simulation tools can be applied. The analysis model considers a system composed of distributed end-to-end flows, each released by a periodic or sporadic sequence of external events, and containing a set of steps that model tasks and messages. Each release of an end-to-end flow causes the execution of the set of steps, each step being released when the preceding one in its end-to-end flow finishes its execution. We assume that all event sequences that arrive at the system and their worst-case rates are known in advance, and we also assume that tasks and messages are statically allocated in processors and networks. The relative phasing of the activations of different end-to-end flows is assumed to be arbitrary. Messages and communication networks can be treated in a similar way as tasks in processing resources. Each step of an end-to-end flow has a worst and best-case execution times, and can have a global deadline referred to the activation of …