HOPE: hotspot congestion control for Clos network on chip

  • Authors:
  • Najla Alfaraj;Junjie Zhang;Yang Xu;H. Jonathan Chao

  • Affiliations:
  • Polytechnic Institute of New York University, Brooklyn, NY;Polytechnic Institute of New York University, Brooklyn, NY;Polytechnic Institute of New York University, Brooklyn, NY;Polytechnic Institute of New York University, Brooklyn, NY

  • Venue:
  • NOCS '11 Proceedings of the Fifth ACM/IEEE International Symposium on Networks-on-Chip
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

Hotspot congestion control is one of the most challenging issues when designing a high-throughput low-latency network on the chip (NOC). When a destination node is overloaded, it starts pushing back the packets destined for it, which in turns blocks the packets destined for other nodes. How to detect the occurrence(s) of hotspot and notify all source nodes to regulate their traffic to the hotspot node(s) can be quite complex because of potentially high volume of information to be collected and the non-negligible latency between the detection point of congestion and the source nodes. In this paper, we propose an effective end-to-end flow control scheme, called HOPE (HOtspot PrEvention), to resolve the hotspot congestion problem for the Clos network on the chip (CNOC). Specifically, HOPE regulates the injected traffic rate proactively by estimating the number of packets inside the switch network destined for each destination and applying a simple stop-and-go protocol to prevent hotspot traffic from jamming the internal links of the network. We evaluate HOPE's overall performance and the required hardware. Extensive simulation results based on both static and dynamic hotspot traffic patterns confirm that HOPE can effectively regulate hotspot flows and improve system performance. Our hardware analysis shows that HOPE has very small logic overhead.