| Sorry, my question was a bit of tongue in cheek one, as the founding member and attendant (till the end of last year) - i know exactly what UEC is about.
To your points - this is none of IETF business why some Eth work is done outside of IEEE; if tomorrow you, me and Bob’s uncle decide to work on another protocol (as long as it is not IPv10 ;-)) we are free to do so in our spare time, and before there’s any interoperability with the existing protocols issue, no-ones business really.
We (chairs) have no intention to make these (mostly educational) meetings an arena for marketing speculations. If anything - we feel, we should go back to basics and IETF relevant technologies, what is RDMA, why it is used in AI clusters, IP transport (RoCE), routing implications, topological implications, etc
We want to see: -folks who use these technologies -folks who build these technologies -folks who innovate in the space (not slides but have running code/products) Jeff, It is a new consortium, therefore, they have not yet produced spec that we can discuss. The objective of bringing them in is two-fold:
1) discuss in details what do they plan to do and why they do not propose a project in IEEE 802.3 and IEEE 802.1 (*) to do what they want in the physical layer and link layer respectively? I am also not clear on what they plan to do in the AI/HPC transport and how it interacts with the physical and link layers.
2) how IETF can work with them and perhaps influence their direction and plans specially in the transport layer where they plan to develop a new transport protocol on top of IP which advances beyond the status quo to support the requirements of AI/HPC applications (I am actually wondering why they do not let IETF do that work? Is it because IETF is slow?)
Hesham
(*) I recall that many years ago IEEE 802.1 created a datacenter bridging (DCB) group to standardize protocols for converged Ethernet. The DCB group was terminated and the last project which was congestion isolation (IEEE 802.Qcz) was done in IEEE 802.1 TSN TG. Is it the termination of the DCB group that motivated the consortium to do the link layer standardization outside IEEE 802.1?
Hesham, What have they produced that is worth discussing? Few companies such as Cisco, HP, Microsoft, Broadcom, AMD created Ultra Ethernet consortium to create standards which optimize Ethernet networks for AI.
For more details you can refer to the consortium website:
|
