Highly Efficient and Predictable Group Communication over Multi-core NoCs

Massive multi-core embedded processors with network-on-chip (NoC) are becoming common in real-time systems. These architectures benefit real-time scheduling of tasks and provide higher processing capability due to abundance of cores. The core-to-core communication can be leveraged by adopting message passing to further increase system scalability. Despite these advantages, multicores pose predictability challenges. In this work, we develop efficient and predictable group communication using message passing specifically designed for large core counts in 2D mesh NoC architectures. We have implemented the most commonly used collectives in such a way that they incur low latency and high timing predictability making them suitable for real-time systems. Experimental results on the TilePro64 hardware platform show that our collectives can significantly reduce communication times by up to 95% for single packet messages. In addition, the primitives have significantly lower variance compared to prior work, thereby providing better real-time predictability.