A High-Performance Design, Implementation, Deployment, and Evaluation of The Slim Fly Network

Explore a groundbreaking conference talk from NSDI '24 that delves into the design, implementation, deployment, and evaluation of the first real-world Slim Fly (SF) network installation. Learn about the advantages of low-diameter network topologies like SF over traditional Fat Tree, Clos, or Dragonfly networks in terms of cost and power efficiency. Discover techniques for simple cabling, cabling validation, and a novel high-performance routing architecture for InfiniBand-based low-diameter topologies. Examine real-world benchmarks demonstrating SF's strong performance in modern workloads such as deep neural network training, graph analytics, and linear algebra kernels. Gain insights into how SF outperforms non-blocking Fat Trees in scalability while offering comparable or better performance and lower cost for large network sizes. Understand the potential impact of this research on facilitating SF deployment and the applicability of the associated open-source routing architecture to accelerate any low-diameter interconnect.

NSDI '24 - A High-Performance Design, Implementation, Deployment, and Evaluation of The Slim Fly...