Scaling Internet Routers Using Optics

This paper explains how a 100 TB router may be designed using future optical technology. The basic idea is to take the Load-Balanced Switch design and fix some of its problems, including creating a fabric that can switch rapidly, packet mis-sequencing (which is fixed by the FOFF policy for spreading work evenly across second-stage queues and a buffer), and working when some linecards are down (which is solved by a complex modular design). The switch design is notable because it requires no centralized scheduler, achieves full throughput, and is agrued to be feasible using technology a few years from now. However, the main shortcoming of the paper is that there is no implementation even of a prototype, and so no way to actually evaluate the switch design. This means that for practical purposes, this is still only design work for improving the Load-Balanced Switch, and may not necessarily lead to a good implementation. Since I don't know much about the hardware used in switches and trends in its manufacturing, it's very hard to tell how realistic the design is.

One interesting aspect of the paper is the point that using optical fabric can reduce power usage. Certainly, power is a very significant concern in network equipment, and the availability of nearly zero-power switching fabrics using optical technology is exciting. However, electronics will apparently still be needed for routing decisions. How much power will be wasted in converting from optical signals to electronic ones? The paper seems to gloss over these points, maybe because the answer is very well-known, but it's hard to tell how feasible the design is without this knowledge.