Joint Computation and Communication Analysis of Hard Real-Time Applications in Manycores

Manycores appeared as a scalable computing infrastructure to support massively parallel applications. Due to the advancements of on-chip technology, manycores present an attractive trade-off between power consumption, area, and fabrication cost. Due to the inherent distributed computing model of manycores, predicting the behavior of modern, mission-critical applications becomes challenging. On the one hand, the community introduced real-time (RT) networks-on-chips (NoC) to support the RT analysis of communication in manycores. On the other hand, implementing non-functional requirements (NFRs) adds complexity to the design, making RT-NoCs prohibitively in resource-constrained scenarios. This work presents a computation and communication approach to RT analysis in manycores. Our approach suits systems without RT-NoCs, enabling manycores equipped with simple low-cost NoCs to run RT applications. We show a proof-of-concept where we can reduce the frequency of the baseline manycore to fit the needs of the target application. Results demonstrate that our approach guarantees RT without modifying the baseline hardware.