Position Paper: Real-Time Task Migration on Many-Core Processors

Many-core processors with Networks-on-Chip (NoCs) offer the scalable computational power required for complex embedded real-time (RT) applications such as autonomous driving. Migrating task between the cores forms the foundation for global and semi-partitioned RT scheduling algorithms which balance the CPU load, improve temperature distribution, and enhance fault tolerance. In this position paper we present a concept that enables tasklevel migration in hard RT many-core systems. The concept checks the feasibility of a migration request at runtime and migrates tasks only via resources whose availability was previously tested. Our concept is not restricted to options precomputed at design time and hence increases the adaptivity and dependability of RT applications on many-core processors while ensuring predictability.