Multiprocessor Systems-on-Chips (MPSoCs) offersuperior performance while maintaining flexibility and reusabilitythanks to software oriented personalization. While mostMPSoCs are today heterogeneous for better meeting the targetedapplication requirements, homogeneous MPSoCs may becomein a near future a viable alternative bringing other benefitssuch as run-time load balancing and task migration. The workpresented in this paper relies on a homogeneous NoC-basedMPSoC framework we developed for exploring scalable andadaptive on-line continuous mapping techniques. Each processorof this system is compact and runs a tiny preemptive operatingsystem that monitors various metrics and is entitled to takeremapping decisions through code migration techniques. Thisapproach that endows the architecture with decisional capabilitiespermits refining application implementation at run-time accordingto various criteria. Experiments based on simple policies arepresented on various applications that demonstrate the benefitsof such an approach.
展开▼
