Programming and analysing scenario-aware dataflow on a multi-processor platform

Abstract

Abstract-The FSM-SADF model of computation is especially suitable for analysing real-time applications with inputdependent behaviour such as different modes, variable execution times and scalable parallelism. Although FSM-SADF specifies which scenario transitions are possible, it does not specify how and when they are decided at runtime. Multiple actors of a scenario, e.g. video stream header parsing, may have to fire before it is known which scenario the application is in. We solve this causality dilemma with a concept for executing a sequence of scenarios, and demonstrate an implementation on multiple processors with rolling static-order scheduling. We furthermore present a platform-aware analysis model that covers concept and implementation, and integrate the contributions in a toolflow. A proof-of-concept confirms the low overhead of the implementation and the exact timing analysis of our model.

On the implementation and execution of adaptive streaming applications modeled as MADF

“…In [3], the FSM-SADF model is presented as another analysis model for adaptive streaming applications. To implement an application modeled and analyzed with FSM-SADF, two programming models have been proposed in [9] and [10]. In [9], the programming model is constructed by merging the SDF graphs of all scenarios into a single graph which may be larger than the FSM-SADF analysis graph.…”

“…In this way, a single static-order schedule can be used for the application in all scenarios. In contrast to [9], the proposed programming model in [10] uses a similar switch/select actors, as in MCDF [2], in the constructed graph for switching between scenario graphs at run-time. Then, the graph is reconfigured at run-time using the switch/select actors according to the identified scenario by a detecting actor(s) while updating the application's static-order schedule accordingly.…”

“…Then, the graph is reconfigured at run-time using the switch/select actors according to the identified scenario by a detecting actor(s) while updating the application's static-order schedule accordingly. However, the proposed programming models in [9] and [10] need to be derived manually, thereby requiring extra effort by the designer. More importantly, these programming models assume that actors in all scenarios of an application are active all the time.…”

“…This can result in a huge overhead for applications with a high number of modes, thereby leading to inefficient resource utilization. In contrast to [9] and [10], our implementation and execution approach does not require derivation of an additional model and enables the utilization of many different schedules rather than only static-order schedule. Moreover, our approach (de)activates actors in different modes at run-time, so we do not need to keep all modes active all the time, thereby avoiding the unnecessary overhead imposed by the approaches in [9] and [10].…”