Re: [802.3BA] XR ad hoc Phone Conference Notice
What is a supercomputer?
Used to be a huge custom box (sometimes several) that needed
a nuclear power station nearby. Today, it is a collection of
a lot (thousands to 10s of thousands) of tiny compute nodes,
COS, using very fast low-latency interconnects.
What is a traditional commercial datacenter?
Same as above, only on a much smaller scale (hundreds to may
be a few thousand nodes).
My point is that from a datacenter architecture and deployment
model, the two are one and the same. If 100m is enough to build
a supercomputer, it is enough to build any commercial system.
It comes down to the size of the datacenter where either one
of them is deployed. At least from our customer base, what we
hear is that links above 100m are very uncommon.
Paul Kolesar wrote:
this is a revelation to me. If I understand your brief response, you
are claiming that the direction of future data center design will be
based around a super computer. I find that very interesting. But it
does little to address all of the more traditional data centers which
are not centered around super computers and have channel lengths
exceeding 100m. Your position for all of those is to deploy a 10km SM
solution for anything longer than 100m. To me that is a recipe for
stalled market penetration that will last for much longer than the delay
we observed at 10G.