Towards efficient resource allocation for embedded systems

Více o knize

The main focus is on dynamic resource allocation in embedded systems, particularly for computing time and network traffic in multi-processor systems on chip (MPSoC). The goal is to dynamically schedule a mobile communication signal processing pipeline to enhance hardware resource efficiency without significantly increasing resource consumption due to scheduling overhead. Both software and hardware modules are analyzed for resource consumption hotspots and optimized accordingly. Utilizing static data flow (SDF) graphs for signal processing, the dynamic management of these graphs is refined to surpass conventional static scheduling methods. A hybrid dynamic scheduler is introduced, merging advantages from processing networks and task graph scheduling, enabling optimal balancing of computation parallelization and dynamic scheduling overhead. This results in a 50% reduction in resource consumption with only a 20% increase in runtime compared to static schedules. Furthermore, a distributed dynamic SDF scheduler is proposed, segmenting scheduling into distinct parts connected through a pipeline for multiple parallel processors. This design alleviates the bottleneck of centralized scheduling, allowing for the management of seven times more processors in typical signal processing applications. The system assumes three communication modes among processing cores, addressing performance issues when emulated over the remote direct memory