Contention-Free Scheduling for Clustered Many-Core Platform
Ryotaro Koike, Takuro Fukunaga, Shingo Igarashi, Takuya Azumi · 2020
High performance embedded systems, such as autonomous driving systems require platforms that reduce power consumption and perform high-performance processing. As satisfying both requirements, multi-/many-core processors are attracting attention. However, for hard real-time applications, there exists the problem of a loss of predictability due to contention for requests to shared resources. Therefore, we propose an Integer Linear Programming (ILP) formula optimization method without contention to increase system determinism by reducing task switching overhead. To predict the occurrence of contention, a directed acyclic graph (DAG) is used to divide a task into a memory access (communication) phase and an execution phase. Existing methods use a clustered many-core processor; however, they do not consider communication between clusters or they are limited to communication between specific clusters. Our proposed method can be used regardless of the number of clusters. Using an ILP formulation, we propose a non-preemptive, partitioned, time-triggered schedule, and evaluate the makespan.