Memory-processor co-scheduling for real-time tasks on network-on-chip manycore architectures.

The network-on-chip (NoC) provides a viable solution to bus-contention problems, that occur in classical manycore architectures. However, NoC complex design requires a particular attention to support the execution of real-time workloads. In this paper, we consider task-to-core allocation and inter-task communications, to guarantee the respect of timing constraints. In addition, we address task-to-main-memory communications, as it generates additional traffic. Rather than separating the problems of task-to-core allocation, inter-task communications, and memory-to-task data transfers separately, we tackle these problems at the same time for a set of real-time tasks modeled using directed acyclic graphs (DAGs). We propose an allocation algorithm and sequence of transformations to simplify the problem resolving and handle its combinatorial explosion. We evalaute the effectiveness of the proposed approaches using a large set of synthetic experiments.