Building Optical Networks Digitally
Every optical vendor pitch over the past three years has included some reference to a time when optical networks will become dynamic, reconfigurable, “transparent.” Though carriers have made limited moves in this direction, they remain mere dabblers when it comes to all-optical networking. Is it because the technology just isn’t mature enough, or does something more fundamental lie behind the reluctance?
It’s worth looking hard at this word transparent. It’s often applied to an optical network interface or system because it operates entirely in the “optical” domain and is indifferent to protocol, bit rate, or formatting. In essence, it is truly optical, having no need to process a signal, only to shunt a wavelength towards its ultimate destination. There has long been a sense of inevitability tied to this notion of the transparent optical network; time would yield the fruits of low-cost, scaleable, photonic infrastructure. Optical would someday break free of the electronic.
Today, my guess is that the photonic future will forever be out of reach, not because of technology but because of network economics. I’d argue that a purely photonic network (one in which wavelengths are created at the edge then networked throughout the core without ever being electronically regenerated) is in fact an analog network that gives the appearance of ultimate scaleability and protocol flexibility, while driving up overall network operations costs and capital costs – and reducing reliability.
It’s become common wisdom this year that carriers have spent too much on their core networks for too little revenue. On the data side of the house, IP revenues couldn’t pay for core router ports, while in the transport network wholesale bandwidth sales couldn’t keep up with the cost of deploying 160-channel DWDM systems.
The answer from many vendors has been to place the blame on the immaturity of the optical equipment. All those OEO (optical-to-electrical-to-optical) conversions among Sonet ADMs, metro DWDM systems, optical switching systems, and long-haul DWDM line systems just cost too much. Scaling a network this old-fashioned way would always be too costly, and yet another generation of optical equipment would be required to bring carriers back to profitability.
The answer, many have argued, is to eliminate those OEO conversions by making them optical – simple passive connections that direct wavelengths from one port to another or one box to another. While OC48 ports on transport equipment hover around $10,000, an optical port on a photonic switching system, for example, is maybe half that, and throws in the benefit of staying that price whether you put OC48, OC192 or OC768 through it, since a beam of light looks quite the same no matter how it’s modulated.
So far, so good. But consider this: What if those savings realized there at the switch or OADM (optical add/drop mux) suddenly cause some unforeseen effects elsewhere in the network? For example, the path length of a wavelength can be dramatically altered depending on which port it is switched to in the node. Where one port may send it from Chicago to Milwaukee, another may send it to Denver. To make it that far, the wavelength either needs to be optically regenerated (no small feat and damned expensive today) or it needs to have started out with enough optical power to stay detectable all the way to Denver. One minute you have cost savings at the node; the next you have Raman amps, ultra-long-reach optics, and wavelength converters through the network.
This, in a word, is expensive. But there’s more. Since the switches at the nodes in these networks are photonic, and therefore transparent, they do not process the content of any signal traversing them. They may employ some device-level technology to monitor OSNR (optical signal-to-noise ratio), wavelength drift, or even bit error rate, but they can’t tell you what’s happening inside the wave. The digital info is off limits. That’s not very good news when customers begin complaining about their service, and it certainly complicates matters when connections need to be made among different carriers or different management domains within a large carrier. Purely optical networks just don’t let carriers sleep well at night.
The enthusiasm around transparent optical networks was driven by the belief that the pace of bandwidth demand in a network core would consistently outstrip Moore’s law, driving electronics costs through the roof. The only solution seemed to be one that eliminated electronics, replacing them with optics. Eventually, some argued, DWDM networks would reach all the way to the home and users’ desktops at work. In this “wavelengths everywhere” architecture, scaleability is the key driver, as a network like this assumes massive growth in bandwidth demand, which can only be cost-effectively met via a conversion of the network core from electronic to optical.
Two things: Bandwidth isn’t growing as fast as we’ve been led to believe (see Did WorldCom Puff Up the Internet Too?) – and there are other ways to skin this cat.
Acknowledging that transponders represent the majority of costs at any given network node, it’s important to eliminate them whenever possible, while maintaining the ability to process signals digitally. This doesn’t mean replacing electronic switches and routers with optical ones; it only means consolidating functions wherever practical.
First, integrating switching (STS1 through OC192) and DWDM transport onto a common platform eliminates banks of redundant transponders at core- or edge-nodes by putting ITU grid lasers directly on the optical switching system or bandwidth manager. This system has the benefit of consolidating the functionality of Sonet ADM, super broadband digital crossconnect (STS management), and a “wavelength” switch; though, in this case, every wavelength is fully processed and regenerated at the electronic level. An extra benefit is had if these are tunable transponders – as cards are added, you simply tune them to the proper wavelength and leave them alone.
This is a bit easier said than done, as most optical switch vendors have found. It takes quite a bit more than just putting tunable transponders on a switch. Issues of control plane integration between bandwidth management and transport must be addressed. Often times a complete redesign is necessary, since the long-reach optics required to support DWDM transmission are often larger and consume more power, dissipating more heat. It will likely turn out that vendors will have to build this kind of switch from scratch. A retrofit won’t yield optimal results.
Beyond consolidation of switching and transport in the node, the next step is to optimize spans around cost and capacity. With full signal regeneration implemented at every node, span design remains quite simple: Get to the next node as cheaply as possible, without regard to the rest of the network. If one span requires significant capacity and is relatively short, then 40-Gig could be used between two nodes, without having to architect the entire network for 40-Gig. If another span is quite long but capacity is only moderate, then dense OC48 or OC192 links can be deployed with ultra-long-reach optics to eliminate or reduce the need for valueless electronic regeneration along the way. This type of network architecture is transparent between nodes, but opaque at the node. Bandwidth management is preserved at every juncture, as is performance monitoring and STSn-level provisioning and protection.
As electronics improve, wideband (1.5-Mbit/s granularity) crossconnect capability can be added to these integrated switching systems, further reducing optical connections within a POP while improving provisioning speeds and network reliability. These are not “God boxes” by any means – they stay well within the confines of transport network functionality.
This network is quite scaleable and can be cost effective over the long run, riding the decreasing cost curve and increased density and performance of electronics, while at the same time taking advantage of optical component developments that improve span design. They also can offer some limited values of transparency by “passing through” circuit management information if required or implementing rate-adaptive electronics to terminate and process a variety of signal formats on a single interface. From all appearances, this network architecture can scale indefinitely and is not inevitably headed toward extinction, to be replaced by photonics.
What does this mean for optical component vendors? They stand to be affected the most, since they are building the devices that live or die by the future shape of optical networks. If networks remain more or less “opaque” as described here, then there will be little need for photonic switch fabrics and wavelength converters. Components facing reduced demand in this scenario include optical add/drop multiplexers, dynamic gain equalizers, ultra-long-reach optics and amplifiers (since they will only be needed on a few spans in any network), optical layer monitoring devices, and active dispersion compensation subsystems.
Who benefits? Chip vendors certainly do, since it will be essential to have the lowest power, smallest footprint chips to keep electronics costs down. In the transponder, chips include framers, transceivers, mux/demux, forward error correction (FEC), and modulators, among others, which will be pushed for greater performance and improved integration. Backplane chips, SerDes, and electronic switch fabrics will also prosper. Others benefiting include tunable laser vendors (eventually, but not necessarily immediately), since they can be used to reduce total capital costs of ownership. Down the road, I’d think optical regeneration would be useful, as well as denser and denser DWDMs and, riding on top of it all, a scaleable optical control plane.
So, while we watch carriers crumble and consolidate it’s worth taking this breather to have a look at what is really coming next. It won’t be soon, but the ones left standing know that an optimal network doesn’t necessarily have to be all-optical. They are certainly examining the technology closely, but getting a sense of timing from them is nearly impossible now, because the numbers aren’t making a compelling case for transparency yet. Component vendors need to take notice, as do systems vendors. The latter, especially, ought to start thinking about deleting that ubiquitous “photonic future” slide and replacing it with something more realistic – an optical network that field engineers aren’t afraid to touch for fear of disturbing the fragile waves careening along these nearly invisible fibers, lenses, and mirrors.
— Scott Clavenna, Director of Research, Light Reading