This post originally appeared on The Next Platform.
Any time you can get a lot of companies with very technically adept and strongly opinionated people to work together on a problem, or a set of problems, then you know for a fact that there is a real problem. The many problems with Ethernet networking at scale caused the formation last summer of the Ultra Ethernet Consortium, which now has well over a hundred organizations participating and which is working to deliver its first specification.
The founding members of the UEC all have skin in the networking game, and the HPC and AI systems space where Ethernet needs to be bolstered and extended to make up for the limitations of current Ethernet and the InfiniBand fabrics that have been commonly used for capability-class systems in the past decade.
The founders include AMD, Arista Networks, Broadcom, Cisco Systems, the Eviden spinout of Atos, Hewlett Packard Enterprise, Intel, Meta Platforms, and Microsoft. Nvidia was not originally a member, but joined recently to throw its weight behind some of the ideas being tossed around about how to create Ethernet networks that have multiple terabits per port and that can scale to over 1 million endpoints. And in most cases, the endpoint in question is a vector and tensor accelerator that is used to do calculations for AI training.
To get a handle on what the UEC is all about and how the work is progressing, we had a chat with J Metz, who is chair of the steering committee of the consortium. Metz has done engineering stints at Apple, QLogic, Cisco Systems, and Rockport Networks before joining AMD as a technical director of systems design. He has also been on the boards of the Fibre Channel Industry Association, the NVM-Express consortium, and the Storage Networking Industry Association, and knows his way around standards bodies that include input from a diverse set of players and experts.
In part one of this series, we spoke about the UEC’s mission, breaking down the barriers in the OSI model and the people who maintain the technologies in the different layers of that model, and the process by which it comes to consensus about how to evolve Ethernet and when different milestones are set. We also talked about the things that the UEC is not trying to do, at least not initially, because creating a more extensible Ethernet without breaking compatibility with the past is a hard enough task as it is.
We hope you enjoy the first part of this conversation, and all you need to do is click on the video link above and you can hear what Metz had to say. If you want to see part two in this interview series, it is available here.
