Dave, that's a rather intriguing set of thoughts. I'd like to sketch it, remove some abstraction and apply some known constructs to see how it looks.
For example, what you haven't explained yet is where the rubber hits the road.. the convergence point where the optical sublayer receives information from above.
Consider IP over SONET, where most lambdas today are still employing SONET framing and PPP as a shim between pure optical and the IP network layer.
Given that those features exist, everything that rides over the lambda is synchronous and timed at the same data rate as the SONET framing rate (e.g., OC-192 at 10 Gb/s). Every bit sent must fit the periodicity and timing of a 10Gb/s signal, in other words.
If your approach avoids SONET framing (and it sounds as though it does), then how are you organizing (framing & synchronizing) individual payloads before depositing them onto the glass?
In SONET, there is a Sonet Payload Envelope, or SPE. Something similar must exist between the virtual glass (lambda) and the IP network layer, unless we are talking about aloha, i.e., free form bits in an ether which is comprised of an almost limitless supply of bandwidth on a wavelength.
BTW, the more wavelengths you have per strand, the less available bandwidth you can expect to find on each. I know you knew that, but it pays to mention it here, because it increasingly dispels, as a function of the number of lambdae, the notion that bandwidth is limitless. Guard spacings in order to avoid crosstalk and non-linear disturbances (aliasing caused by sidebands) between lambda centers will likewise eat up spectrum, further reducing the spectral real estate we once called limitless. |
