I'm going to bail out of this little exercise which focuses on the last mile, for now. Except for its metaphorical value, in retrospect, it was ill advised to use the residential cable TV model as a realistic example, due to the momentum in that space, as ill advised as I think they are, and despite it being a highly plausible application for the future. I'd rather continue examining the relative merits of DWDM, and possible reasons why some alternatives (in a number of different roles) might be needed.
----
Like the SONET/SDH-, T3/E3- , and copper-based- ceilings before it, the WDM ceiling will now prove to be extremely unyielding, because it appears to be working satisfactorily for the time being. The user community for DWDM, namely service providers, has all it can do at this time to catch their breath and assimilate this new paradigm which deals with colors, primarily, instead of lower denominations of quasi-optical transport bundles in the opto-electronic space. Of course, I'm referring to T3s and E3s, OC-n's and STM-n's, which represent the SONET and SDH hierarchies, respectively.
It's only now, close to three years later, that most operations level personnel are getting over the shock waves that were caused by the initial announcements that were made at the September '96 Supercomm Show, when the first commercial-grade variants of today's WDMs/DWDMs were unveiled.
At some point in time, carrier engineering and operations level staffs will become totally overwhelmed by all of the colors, causing some of them to become blinded by the light, so to speak, before they are forced to consider some more economic and administratively friendly alternatives. That's how I see it, in the main, right now.
But some carriers and SPs are able to see beyond the current horizon earlier than others, and they have already seen the need to begin preparations by entertaining new approaches in anticipation of that eventual onslaught of flows.
Wavelength management is easy enough to contend with when the number of lambdas are restricted to several dozen or under a hundred or so wavelengths, on a relatively few high density routes. But this is before meshing occurs. Once optical meshes are created, the carriers must either find new ways of porting optical sub-payloads between optical links, or resort to downshifting (back to back muxing) them to traditional SONET and subrate formats, and then putting them through digital cross connects as they have done going back to the late Seventies.
Okay, they'll do it with ports that are optically defined, but for the most part many of those ports will still contain in their designations the numbers OC- 3, 12, 48, 192 and so on, or somewhere else in their definitions.
[[This is not exactly what optical is supposed to be about, at least not at the edge and into the core. But at the end office level, closest to the end user, this model will persist because of customer access constraints. I.e., most customers will only be able to afford aggregated access through traditionally defined means (T1s, T3s, OC12s, etc.) In the near term, only the more affluent will have native fiber pulled to the service providers' nodes. ]]
Granted, DWDM is relatively inexpensive to maintain when viewed against the alternative of pulling additional glass, when route distances can justify them. But again, when those routes are few and far between and don't require optical-level interchanges between them in a pluralistic multivendor/multicarrier environment.
But when you begin raising the number of colors exponentially (as I believe will happen soon, indeed we've already begun to see this in the past several months), DWDM doesn't scale very well due to wavelength element specifics (the optical elements being what they are today) which are required, and the number of physical level paths (and their subordinate payloads) that must be managed. Here, we deal with the physical (strand), the virtual (the wavelength), and the hypo-virtual (the individual user frames amd packets which reside within wavelengths). Frames and packets, accordingly, will more often than not yield further subdivisions down to individual end point originated data.
From the standpoint of providing "private lines," allocating colors to individual end users is fine. The more you can deliver (again, if the distance tradeoffs justify), the more revenues a carrier will garner per strand. But from a "networking" perspective, i.e., allowing for the interchange of data between lambdas, it's got its fair share of problems.
At some point in this respect, DWDM stops scaling to breakeven capabilities, altogether, when viewed in the light of total architectural considerations, when using today's state of the art. This, even taking some accelerated improvements into account over the next several years. Paradoxically, this becomes so "especially" when taking the greater number of optical channel creating improvements into account. One reaches a point, in other words, when any further improvements in color fabrication requires an increased distance to break even. When this happens at the shorter distances, it's cheaper to simply use fewer wavelengths and add more strands.
Perhaps these conditions will change sooner than I can see, at the moment, but there's still a race right now to see who can best optimize optical resources over 'any-distance' conditions the best, from what I can see.
The past half dozen posts or so were only meant as a metaphor to exercise the concept of SR's approach in terms which were generally familiar. Instead of using "homes passed," as in the residential cable TV model, consider "data centers passed" or "mega-POPs/NAPs passed" instead. Then, a single beam approach, if it were truly capable of scaling terabits, begins to make more sense, purely from the perspectives of costs and administrative ease.
Having dabbled in this model with DWDM parameters in mind, I can attest that box management, and element level tracking vs. lambda end-point assignments, becomes very expensive and administratively challenging once you go beyond several dozen colors of the sort currently available, and using the legacy constructs which are characteristic and most familiar to carriers.
The greater the number of colors used, the greater the distances required to justify the overall model. Maybe someday an inflection point will occur, when organically synthesized color generation methods are improved and managed entirely under software control, but not anytime soon.
And yet, some means beyond basic physical level mapping onto glass is required, due to the need to route optical payloads from one fiber 'container' to another, lest we are left with stranded payloads on individual strands.
Here lies the crux of the matter, then: how to parcel individual streams carried by optical, at some layer, no matter how thin that layer may be, above Layer One, or residing as close as possible to the upper Parts of Layer One convergence, and allow those parcels to migrate from one fiber to another, in established POPs and along established routes.
Monterey thinks they have the answer with DWDM and WaRP. Corvis and others are making even bolder claims, still, but their methods wont be unveiled until some time into the future. Yet, as far as I can see, they are all dependent on, and left with the same scaling problems I previously mentioned, in conjunction with DWDM's basic attributes. While I'm actually encouraged by such measures, they are proprietary at the current time. Just like IOS was originally, from CISCO. Which, in itself, leads to some interesting conjecturing, but I've gone long enough in this post.
How extensible will each of these proprietary schemes be, beyond the borders of individual carriers' reaches? The name of the game is supposed to be making optical less expensive to deploy (and, hopefully less expensive for end users to purchase at the same time), in ways which are ubiquitously communicable between different providers meshes. Are we on that road today? Or have proprietary architectures, for the sake of differentiation, already begun to take us in the opposite direction, again? |
