"For example, if the baseline traffic on a data center network is 1Gbps, but peaks at 10Gbps for just a few minutes at lunch time… do you wire the entire data center up with costly 10Gbps gear, or just admit that there’ll be lots of latency and packet loss at peak times?"
Heh. The article makes this a 1G or 10G choice. Haven't they heard of 802.3ad?
10Gbps is an aggregate for all links in the data center.[1]
Even $99 low end consumer ethernet switches have an aggregate speed of 16Gbps.
For this to make any sense at all you have to have a switch in the rack aggregating traffic but still be using a 1Gbps ethernet back to a more central switch. The cost of a 10Gbps link isn't too bad when spread over a rack of equipment. Compare that to the cost of an RF system that has to be engineered for the exact physical installation, still needs a >1Gbps uplink port on the rack switch, and silently breaks if someone sets a cardboard box on top of a rack.
[1] Their number, not mine. The report talks about using the spatial restrictions of the links to make many "multiple Gbps" links in the same space.
Even $99 low end consumer ethernet switches have an aggregate speed of 16Gbps.
Not really. Most only have a 1 or 2 Gbps backplane and you're talking about absurdly priced switches that have a backplane that will support full saturation on all ports.
I'm talking about a Netgear GS108Tv2, just because that what is sitting on my desk at the moment. It claims 16Gbps right there in the brochure. This is a notch above the lower end, it has a management interface, VLANs, QoS, port authentication, and a lot of other features from real switches.
Of course it also claims lifetime warranty but somehow that doesn't work after 90 days, so maybe the 16Gbps is not true either. In a testament to their opinion of their product, Netgear will sell you a warranty that does work for a couple years for only a bit more than the cost of the switch itself.
