Examining the 'all' in all-optical networking
ANDREW McCORMICK and ELIZABETH BRUCE, Optical Strategies
All-optical networking today is no more than an industry buzzword. The term is often used to refer to optical networks that actually include a great deal of electronic components. The question is, will telecommunications networks ever be completely optical? While some herald the dawn of the all-optical network (AON), in the strictest sense, it remains many years away.
The ability to read and store packets of light (photons) are only in the research lab today. The explosion in the number of optical-networking companies over the last few years and the overall growth of the industry have created an illusion that AONs are just around the corner, ready to bring the benefits of unlimited bandwidth at low cost. In reality, many advancements in optical networking are simply incremental improvements to fundamental breakthroughs that occurred over a decade ago, particularly the developments of the erbium-doped fiber amplifier (EDFA), DWDM, and singlemode optical fiber. Though new technologies such as tunable lasers and optical-switching modules promise to advance the cause of all-optical networking, these are difficult technologies to commercialize.
Telecom networks differ from pure data networks in that they must be able to accurately monitor, manage, and guarantee delivery of the information they carry. Today, that can only be done with electronics in the network. A number of advances must be made before we can even begin to consider what is truly "all-optical networking."
One of the major debates in the market revolves around whether there is a bandwidth glut or a bandwidth shortage. While both sides can present statistics supporting their cases, the real problem is getting adequate capacity where it is needed at the time it is needed.
Most major communications carriers continue to double the amount of data traffic on their networks every year. While that may be a drop from the 300% growth rates of the recent past, there is still significant demand and a high rate of growth. Yet, it is impossible to predict where that growth is coming from year to year or even month to month.
To support increasingly dynamic enterprise operations, new applications for processes such as customer-relationship management, supply-chain management, and enterprise resource planning are being used. The client/server architecture of these applications requires high-speed connectivity from remote offices to a centralized server. In many cases, separate connections are required to support individual applications because of their mission-critical nature. Additional bandwidth is needed for offsite data storage and data mirroring. At higher layers, new ways of providing connectivity for these applications are also being developed and deployed. Transparent LAN services, virtual LANs, and IP-based virtual private networks are changing the way information is distributed.
On top of all that, new networking relationships and bandwidth applications are evolving in the carrier industry. Managed wavelength services and other wholesale bandwidth products are changing from fixed, long-term contracts for static links to yearly, and even monthly, contracts for varying levels of capacity. The commoditization of capacity with the development of bandwidth trading markets will bring the need for network interconnectivity and fluidity to a whole new level. For these new applications and markets to thrive, this connectivity must occur at the optical level, since it provides the high capacity, rapid provisioning capability, and service transparency to serve the multitude of needs in the marketplace.
The immediate impact of these market changes on carriers has been eroding profit margins. The cost of the electrical components required in existing systems to regenerate and switch this traffic is outstripping the incremental revenue from the new services these systems support.
Optical in the network
In the truest sense, an all-optical communications network would generate, pro cess, store, route, and transport information entirely in the optical domain as photons, in the same manner data packets are handled today as electrons. AONs today are, in reality, point-to-point optical links sewn together by electronic means. Add/drop multiplexers can drop and insert individual or bands of wavelengths and perform the multiplexing/demultiplexing function optically-but getting to the subcarrier (i.e., below 52-Mbit/ sec) level for distribution at the edge of the network requires electrical conversion. Because of these processing limitations, optical transport is really the only application for all-optical-networking devices today. Because of network economics, the backbone has been its only domain.
While commercially available all-optical-switching/routing functionality is available in optical crossconnects (OXCs), such as the CorWave OS from Corvis Corp. (Columbia, MD) and WaveStar LambdaRouter from Lucent Technologies (Murray Hill, NJ), that functionality is limited to redirecting individual wavelengths from an input port to an output port. Photonic packets cannot be processed on an individual basis like packets in an IP router. That means OXCs have no ability to combine or regroup information within an individual wavelength or across different wavelengths to optimize network performance.
The advantage of OXCs is that they redirect large amounts of information much faster than data routers and switches. Because of this ability, OXCs do have application in the core of large networks where multiple high-speed wavelengths need to be switched, though the market for that is currently limited (see Figure 1). For example, Broadwing (Cincinnati) has announced that it is deploying the CorWave OS in its terrestrial network to switch traffic across the three major rings that constitute its national backbone network, but accomplishing that requires only six OXCs. Global Crossing Ltd. (Bermuda) has likewise deployed the LambdaRouter, but only at a few cable-landing sites.
For intracity traffic, the cost of optical components has at last decreased to the point where it is now economically feasible to deploy DWDM equipment for use in MANs that connect data centers or central offices (COs) in a given city or localized region. Technologies like selective DWDM as well as the optical bidirectional line-switched ring developed by ONI Systems (San Jose, CA) also play a key role. They adapt the point-to-point nature of optical links-well suited for backbone topologies-to ring-based networks typically found in the metro environment.
The last mile
Though it is common for large enterprises today to be served directly by fiber links, fiber-to-the-business and -home are far from being readily available. There are two access technologies that can provide optical services to the end user. The first is passive optical networking (PON). As its name indicates, PON technology is characterized by relatively inexpensive passive optical elements such as splitters and filters in the outside plant.
The advantages of PON include higher capacity than standard telephone cabling or coaxial cable and greater reach than DSL. The gating factor, of course, is that for a service provider to use PON, it must have fiber to each subscriber. As a result, while most of the RBOCs have implemented PON on a trial basis, none have fully committed to it. Somewhat contrary to the typical technology deployment model, several rural independent telephone companies are using PON to overbuild their networks in competing service areas with fiber to expand their service territories.
The second optical access technology is free-space optics (FSO). FSO uses roof- or window-mounted transmitters to direct an optical signal to a receiver mounted in another building. Standard FSO technology uses inexpensive 850-nm lasers for data rates up to 1.25 Gbits/sec and distances up to 2 km. Lightpointe, an FSO supplier in San Diego, has recently developed a 1550-nm system using more expensive lasers for data rates up to 2.5 Gbits/sec.
Historically, the biggest drawback to FSO has been the deterioration of signal quality due to atmospheric conditions, resulting in less than carrier-class reliability. Improved systems are now on the market using techniques such as spatial diversity and beam divergence to compensate for both atmospheric variations and building sway. Future development will include adding DWDM capability to increase network capacity.
The advantages of FSO are that it operates in an unlicensed portion of the spectrum where no expensive governmental licensing is required and it is faster and cheaper to install than fiber. With low-end systems starting under $10,000, FSO technology is being looked at as an increasingly viable solution to providing high bandwidth directly to the end user.
The main obstacles to the realization of the AON are the technical challenges presented by optical technology itself. Though this field is well researched, there are many technologies still in the research lab and far from commercialization. Chief among these technologies is the ability to process and store information in the form of photons. Networks make routing and switching decisions by reading the header bytes in the IP packet, ATM cell, or SONET/SDH frame. When information is in the form of photons, networks lose that ability. Until this problem can be solved, true photonic routing will not exist.
Storage is also necessary for buffering information as it comes into a switch or router. Electronic buffers temporarily hold information until a decision can be made on how it should be processed. The closest anyone has come to emulating that optically is through the use of circulators within a switch-in effect, "slowing down" certain wavelengths by forcing them to take a longer path.
Though DWDM transport has been used in the network for almost a decade now, systems developers still face challenges to increase the distance optical links travel, the number of channels per fiber, and the bit rate per channel. Unfortunately, these metrics can have an inverse relationship with one another. Performance improvement in one area may have negative consequences in the others.
The objective for many long-distance optical links is to keep everything in the optical layer for as long as possible to avoid expensive electrical regeneration. The challenge is maintaining the system performance and acceptable bit-error-rate requirements over long distances. Network engineers have developed many techniques for accomplishing that, including optical amplification, dispersion compensation, different modulation schemes (i.e., solitons), and forward error correction (FEC).
Optical amplification using EDFAs has become one of the most fundamental tools of the AON. One of the difficulties in using EDFAs is that the gain profile is not flat, resulting in uneven amplification of different channels across the band. This uneven gain causes severe signal degradation at the receiver if not corrected.
In addition, each inline amplifier introduces a certain level of noise and gain ripple, which accumulate over the total distance of a link. As such, gain-flattening filters are designed to manage a portion of the signal degradation caused by EDFAs. The problem is that these filters are fixed, meaning the current generation of EDFA-based systems is relatively static and ill-equipped to handle a changing or dynamic environment. New remotely configurable modules such as dynamic gain equalization filters and variable optical attenuators are currently being developed to address some of these limitations.
While EDFAs have been designed to operate in the C- and L-bands, they cannot be used to amplify shorter wavelengths (in the S-band). That is one of the advantages of Raman amplification: It can be used to amplify any portion of the spectrum. Raman amplification is currently used in long-distance optical networks offering high levels of amplification across multiple bands (see Figure 2).
While Raman amplification is less efficient than EDFA technology (for typical single-band systems with approximately 80 channels), distributed Raman amplification does not require a separate gain medium for operation. One of the difficulties with Raman technology is that the 14xx-nm pump lasers required for Raman amplification tend to be relatively costly and have only recently become available at high powers. In addition, there remain many challenges in managing gain tilt in Raman systems. Although the principle behind Raman amplification has long been understood, the commercialization of the technology is a relatively new phenomenon.
Dispersion is a concern in optical networking as engineers design systems for higher capacity by increasing the number of channels and data rate. Chromatic dispersion (CD) occurs because wavelengths travel at different speeds in a fiber, effectively broadening each pulse. Polarization-mode dispersion (PMD) is also an issue in higher-data-rate systems (40 Gbits/sec and above) and can cause the significant distortion of optical signals. The effects of chromatic and polarization dispersion accumulate as a function of the distance a signal travels, and the impact of dispersion increases significantly as a function of the data rate.
For example, a 2.5-Gbit/sec channel can travel over 6,000 km. This distance decreases to 400 km for a 10-Gbit/sec channel and only 25 km for a 40-Gbit/sec channel before encountering the effects of PMD. CD has a similar effect on performance as a function of distance, though at lower distance thresholds. Because CD and PMD have different causes, different dispersion compensation modules must be specifically designed to counteract the negative effects. That's a problem for long-haul network planners, since each network-and even each optical span-is unique.
The challenges of optical switching remain numerous. Optical switch fabrics range from micro-electromechanical switching (MEMS) to liquid crystal to planar waveguides and even holographic. Despite the amount of research and development that has gone into this field over the last few years, an ideal switching fabric has not yet emerged. Because each is more suited for different applications, it is likely that a number of technologies will be used. While MEMS is the technology of choice in core OXCs because of its superior scalability, planar waveguides may be used for configurable OADMs because of the inherent advantages as an integration platform.
What is not known is how network architects actually want to deploy optical switching. Some switch-fabric manufacturers feel there may not be as great a demand for large OXCs as originally anticipated, and instead there will be a greater number of flexible OADM sites.
Another reason we have not seen a dominant optical-switch technology is that component suppliers have yet to present a cost-effective, reliable switch fabric that can be manufactured in volume. In particular, component packaging remains a serious issue for scaling production. For systems developers, reliability of these emerging technologies continues to be a concern.
Many of the issues ultimately come down to cost. Today, optical switching is just not proving cost-effective; as data rates increase, it will eventually prove out, but that crossover point has been stretched for some time as the cost of semiconductor technologies continues to decrease as performance increases. In fact, 128x128 switch chips with 2.5-Gbit/sec line rates exist today that can be used to create switches up to 8,000 ports bidirectionally, or 2,000 ports at 10 Gbits/sec. Given that no carrier is using a switch even close to this size yet, it seems that the market for all-optical switches is well into the future.
To balance the competing requirements of the network, some optical system vendors are developing hybrid systems that have both electrical and optical capabilities. The current generation of OXCs, including the CoreDirector from Ciena (Linthicum, MD), SN16000 from Sycamore Networks (Chelmsford, MA), and Aurora Optical Switch from Tellium (Oceanport, NJ), require an optical-electrical conversion to perform their switching functions. The next generation of OXCs is being developed to include purely photonic-switching modules-most likely based on MEMS-so traffic requiring processing can be directed to the electrical fabric, while pass-through traffic can be handled optically.
"All" is just a marketing term
Despite the amount of effort in the development of optical networking, the concept of the all-optical network has yet to be achieved. The industry is still at the very beginning of development in this process, and it will take a great deal of time before the ultimate realization of these efforts. There are certainly some very promising technologies, however, and most people in the industry are confident the technology issues can be solved.
It needs to be understood though that the problems associated with commercializing these technologies are far from trivial. These problems can only be truly appreciated by the relative handful of optical and design engineers actively engaged in solving them. Unfortunately, the industry marketing machine and venture capital eagerness obscured these issues to the point where the realization of the AON seems to be a foregone conclusion, or worse yet, already exists.
While we may never see the all-optical network in the purest sense, there are portions of the network that will make increasing use of optical technologies. The rate of adoption of these technologies is not entirely clear, since so many barriers remain. As long as these barriers exist, electronic and optical technologies will continue to coexist in the network.