So...
In a few years time,could we be seeing 400Gbps FCoEoF?
After a year of discussion and setting of baseline specifications, the hard work of creating 400G bps Ethernet is about to begin. This month, according to the acting chair of the IEEE's 400G Ethernet study group, Dell's John D'Ambrosia, the group will be seeking approval of the project documentation it's been developing since …
Funny, but it's not Ethernet over Fiber. Fiber and copper are both part of the Ethernet standard. No I didn't downvote you, just being pedantic.
Yes, with the spacing, the bits are less than 1mm apart. This means that in a distance of about 12 meters, you have an entire (standard) ethernet frame.
The problems with "frames in transit" starts to make things difficult. Go longer distances and limits of TCP become evident.
So, generally nice for in rack communications, longer distances start to be problematical.
As consumers, 100Base-T works quite well, but doesn't satisfy those who believe "Instant gratification isn't soon enough".
Well yes, but we already have that problem on the Internet right now. You have 100ms latency for intercontinental links and 100 MBit connections (in cheap data-centres or with end users in developed countries like Turkey). So you easily have a megabyte in transit.
Obviously there is a reason for Gigabit Ethernet for consumers. It allows much faster file transfers. Realistically you can pipe 50 Megabytes per second even with consumer equipment. It's relevant if copying your file takes 1 or 5 hours.
Short distance 400G bps Ethernet may be useful for CERN, the Square Kilometer Array or have a roll as the backbone of a supercomputer. 'Normal' hardware cannot feed a connection like this.
In broadcast facilities underground tunnels with hundreds of parallel optical fibers are used. There is nothing wrong with that.
Well, it's not that far off 'Normal'. 400Gbps is going to pan at at less than 50GByte/s, and there's Intel CPUs that have that much memory bandwidth (certainly when inter-leaved across multiple CPU sockets). The second requirement is for a CPU -> peripheral bus that's equally fat, and that's only a matter of bus width ultimately.
So we're not far away from it being "Normal" at all, especially as anything reasonable NIC in this class would offer TCP offload facilities. Give it a few years and it will seem routine.
It does raise an interesting point. If Ethernet is the fastest interconnect we have, people will start using it inside computer architectures instead of PCI or whatever.