Rockport Fabric Implementation at Durham University

This case study details the implementation of the Rockport fabric technology at Durham University's Institute for Computational Cosmology (ICC). The objective of deploying the Rockport fabric was to enhance the performance and reliability of advanced workloads, particularly in simulating cosmic phenomena. The document outlines the challenges faced by the ICC, including network congestion that hindered workload completion times and underutilization of computational resources. The Rockport fabric's architecture eliminates traditional switching layers, thus reducing latency and improving network performance. The Autonomous Network Manager (ANM) is also highlighted for its role in providing real-time insights into network traffic, which aids application developers in optimizing their codes. Initial tests demonstrated that the Rockport fabric performed equally or better than existing InfiniBand networks under congestion. Following these results, Durham University expanded the Rockport deployment to half of its COSMA7 cluster, indicating a commitment to leveraging this technology for enhanced computational efficiency in research.