High Performance Ethernet
The high-performance computing landscape is rapidly evolving, with modern science platforms increasingly adopting cloud-like flexibility, scalability, and service models. Alongside this shift, there is growing momentum toward cross-industry, standards-based networking solutions that enable broad interoperability with enterprise computing technologies across diverse, multi-vendor environments.
While Ethernet has long been the dominant standard in enterprise networks, it is only recently beginning to demonstrate the low-latency performance needed for demanding high-performance computing workloads—opening new possibilities for unifying enterprise and HPC infrastructure.
Aims
The Lab will explore the use of multi-vendor, low-latency Ethernet as a viable networking solution for high-performance computing environments. By benchmarking a broad range of HPC applications at different scales, we aim to evaluate Ethernet’s performance relative to today’s commonly used low-latency fabrics. This research will clarify Ethernet’s role within the HPC ecosystem and help identify the performance characteristics needed to support diverse application workloads.
In parallel, the Lab will investigate software-defined Ethernet technologies, including the use of programmable networking languages such as P4®, which offer the potential to dynamically tailor network behaviour to meet specific application requirements.
Finally, the Lab will work to catalyse cross-industry collaboration around open standards for high-performance Ethernet, helping to drive innovation and ensure broad interoperability across the evolving HPC landscape.