Extended Abstract: Process Networks for Reactive Streaming with Timed-automata Implementation

International audienceMost of modern academic tool flows for embedded real-time systems support either the streaming or the reactive-control class of application programming models. These two classed have historically developed two different design methodologies. The former, such as CompSoc [9], are typically dataflow-related and is based on the analysis and optimization of timing properties in system steady state. The latter, such as Prelude [4], are based on synchronous language compilation and classical real-time schedulability analysis. However, when implementing modern complex applications (such as avionics, satellite and robotics control systems) on many-core platforms we encounter disadvantages of the separation of systems into two classes. Focusing on only one of them imposes certain undesirable methodological restrictions that are not necessarily present in the other one. We present our current ideas towards unifying these two classes. To this end, in this abstract we discuss a recently developed [11] model of computation: Fixed-Priority Process Networks (FPPN). FPPNs extend streaming models by support of time-dependent (yet deterministic) behavior and real-time task properties (e.g., sporadic/periodic activations with deadlines) for the processes and channels that are not necessarily FIFO's. These extensions are possible due to decoupling between the process blocking from the inter-process channel accesses. Our public design flow [14], [10] compiles FPPN's to executable component-based model with timed automata components. Timed automata is thus used as a 'meta-model' to define the semantics of FPPN and to provide a basis for simulation and deployment. Moreover, automata are useful means for adding the system middleware components that cannot be expressed in higher-level models of computation, such as run-time management, e.g., QoS control [1], and custom scheduling policies [13]. We demonstrate combining such automata with FPPN models in [14] and [12]. An instance of FPPN is composed of four main entities: Processes (tasks), Data Channels (communication buffers), Event Generators and Functional Priorities. The process network example in Fig. 1 represents an imaginary signal processing application with input sample period 200 ms, reconfigurable filter coefficients and a feedback loop. The filter coefficients are reconfigured by sporadic process CoefB. We see several periodic processes, annotated by their periods , and a sporadic process, annotated by minimal inter-arrival time. This process also has a non-default burstyness value m e = 2. We also see inter-process channels-the blackboards This research received funding from MoSaTT-CMP-European Space Agency project, and from CERTAINTY-European FP7 projec