Model-driven development of multi-core embedded software

Event-driven Scheduling for Parallel Stream Processing

“…By using the newly introduced model-to-model transformations and generation support, it is now possible to automate the exploitation of this parallelism. In DSPE, unlike alternative approaches [3], [12], [17], [19], each application may now feature its specific event-driven scheduler and work-sharing/stealing infrastructure, and each software component may be handled with dedicated scheduling and coordination policies. This is a significant advantage in stream processing applications, where straightforward data divide-and-conquer [2], [12], [17] may not prove effective, for example, in the case of irregular data availability.…”

“…In DSPE, unlike alternative approaches [3], [12], [17], [19], each application may now feature its specific event-driven scheduler and work-sharing/stealing infrastructure, and each software component may be handled with dedicated scheduling and coordination policies. This is a significant advantage in stream processing applications, where straightforward data divide-and-conquer [2], [12], [17] may not prove effective, for example, in the case of irregular data availability. In this context, dedicated infrastructure, scheduling and coordination policies help minimize the impact of load imbalances and other interfering factors.…”

“…The data pressure problem is one of the most difficult in real-time streamprocessing applications, because the amount of data available contemporaneously for a component is usually small. Unlike other approaches [17], [19], our solution, which advantages from model-driven generative programming, is based on the event-driven scheduler and the task worksharing/stealing infrastructure. From the point of view of the run-time infrastructure, our solution is fully dynamic, as it does not assign tasks to cores statically (unless specified).…”

“…In the category of dynamic load balancing approaches, VERTAF [17] is a stream-processing tool that uses the Threading Building Blocks (TBB) [12] as a framework. VERTAF combines TBB with a software architecture based on "active objects".…”

“…Therefore, unlike other approaches [1], [2], [3], [5], [9], [12], [15], [16], [17], we concentrate efforts on flexibly and automatically combining different forms of both dynamic load balancing and static pre-compile optimizations. To best profit from the resulting synergies we use model-driven generative programming techniques.…”