Congestion Control for Scalability in Bufferless On-Chip Networks

In this paper, we present network-on-chip (NoC) design and contrast it to traditional network design, highlighting both similarities and differences between NoCs and traditional networks. As an initial case study, we examine network congestion in bufferless NoCs. We show that congestion manifests itself differently in a NoC than in a traditional network. This both reduces system throughput in congested workloads for smaller NoC sizes (16 and 64 nodes), and limits the scalability of the bufferless NoC in larger configurations (256 to 4096 nodes) even when data is mapped with locality in mind. We propose a source throttling-based congestion control mechanism with application-level awareness. This mechanism improves system performance by up to 28% (15% on average in congested workloads) in smaller NoCs, and achieves linear throughput scaling in NoCs up to 4096 cores. Thus, we show an effective application of a network-level concept, congestion control, to a class of networks – bufferless on-chip networks – that has not been studied before by the networking community.