Federated Scheduling in Clustered Many-core Processors
Ryotaro Koike, Takuya Azumi · 2021
High-performance embedded systems, such as self-driving systems require platforms that reduce power consumption and perform high-performance processing. As satisfying both requirements, multi-tmany-core processors are attracting attention. This paper focuses on a clustered many-core processors represented by Kalray MPPA. Clustered many-core processors regard multiple cores as a single cluster, with each cluster having private memory. A bus within each cluster performs communication. A network-on-chip (NoC) is used between clusters. These two communication speeds are different. However, it is inappropriate to always assume that the worst calculation is the slow NoC communication speed. This paper discusses a method to improve worst-case communication time estimation when applying federated scheduling on clustered many-core processors with routes at a different speed. Furthermore, this paper investigates an approach to assign dedicated cores in multiple clusters to reduce the number of required cores.