Electrical or photonic: ying and yang of switching

Will the industry opt for one over the other-or will they coexist in tomorrow's network infrastructures?

IAN WRIGHT, MARC SCHWAGER, and ROB NEWMAN, Altamar Networks

There has been considerable interest and discussion recently in the telecommunications industry re-garding photonic switching versus electrical switching. With the explosion in demand for optical transmission, the question is: which is the better switching technology to scale and support the rapidly evolving requirements? As with the introduction of any new disruptive technology, predictions abound that photonic switching will soon completely replace electrical switching in the core of the carrier's network.

A close look at electrical- and photonic-switching technologies reveals that they have very different characteristics and therefore one is not a direct substitution for the other. It is more likely that these two capabilities will co-exist in the carrier's network.

To trace the source of the excitement for switching in the telecommunications transmission network, it is interesting to review with carriers what they are doing today. Discussions have been held over the last few months with a few of the largest long-haul carriers. These discussions revealed some very interesting challenges and contradictions. On one hand, the engineers and planners feel overwhelmed by the need to add new bandwidth. On the other hand, operations people are tightening budgets.

In times when new investment is required but budgets are tightening, the usual business response is to look at productivity gains and focus investment on only the core business. That's what the long-haul carriers are doing today. They are tuning their investment dollars toward the business they know best-providing long-haul transport-and refocusing their spending to lower both operational and capital costs.

SONET has come under the microscope within carrier networks, because it is a critical component in the delivery of transport services. However, SONET was designed in a time when voice traffic dominated. Today, the situation is very different. While voice still represents a significant part of the carrier's revenue, the value of that revenue is declining and the voice growth is very slow. Data transport, in contrast, is growing very rapidly, already exceeding the voice traffic on the network-and the value of data is increasing.

Figure 1. Due to data traffic growth, if the total traffic requirement increases 10-fold in three years, then the costs of SONET equipment would equal the revenue. However, revenue is not expected to increase 10-fold, rather only by a few percentage points over those same three years.

The challenge for the carrier, however, is that the revenue per bit for data is much less than that for voice. Here lies the crux of the carrier's issue: the economics of SONET don't work for data transport. Figure 1 shows this problem from another perspective. If capital equipment costs associated with SONET today represent only 10% of the carriers' revenue, the carrier has a profitable business model. Due to data traffic growth, if the total traffic requirement increases 10-fold in three years, then the costs of SONET equipment would equal the revenue. It is unlikely the carriers can convince their enterprise customers to pay a lot more for their services. Therefore, revenue will not increase 10-fold, but probably only by a few percentage points over those same three years.

This realization has already hit home with the carriers and is impacting their buying decisions today. It is noticeable in the revenue projections for the telecom-equipment companies that are the large SONET pro-viders-they have all lowered forecasts. Meanwhile, providers of DWDM equipment, which increases the bandwidth in the network, are raising estimates.

So if more bandwidth is being de-ployed over the fiber using DWDM, and SONET gear is not being used as the terminals on those lines, what is being used instead? Projections from industry analyst RHK Inc. (San Francisco) give us the answer: optical switching. That's where the technology debate begins. Everybody agrees that optical switching is the right direction, and leading carriers have already deployed first-generation optical switches. But what is the best way to build one of these switches?

Two technologies exist today that can meet the broad set of requirements for optical switching. Optical switching in this context is defined as the requirement to switch signals transported optically. For example, suppose an OC-48 (2.5-Gbit/sec) or OC-192 (10-Gbit/sec) stream on the fiber needs to be routed from San Francisco to New York. DWDM will provide the long-haul transport of this signal from central office to central office. But in each central office, the signal needs to be directed on to the next path, which includes identifying the link, fiber, and specific wavelength of the next hop toward New York.

Optical switching can be achieved with either photonic technology or electrical technology. Photonic switches are generally defined as switches that use either mirror or refractive effects to redirect the stream of light. Electrical switches will act on the electrical bit stream derived from the optical signal. Electrical switching can draw on a variety of architectures and technologies that have been in use for other types of switches.

At first glance, there is instant appeal for photonic switching. The signals carried on the fiber are optical and would stay optical through the switch, with no need to translate to electrical and back to optical. There is another benefit to a photonic switch: It operates independent of the bit rate of the optical signal. Whether the signal on the fiber is modulated at OC-48, OC-192, or OC-768 (40 Gbits/sec) would not matter. Light is light, and the switch will reflect it regardless of the speed. It is this feature that leads to the claim by photonic-switch vendors that these switches will scale with transmission speed.

Figure 2. In a typical model for an electrical switch in the core of the carrier's network, a number of optical functions are performed on the signal.

Based on these two benefits, there has been considerable research and commercial investment into photonic-switch fabrics. This class of products is not yet shipping in volume, but some vendors are currently undergoing trials or promising them within the next few months.

However, the photonic switch fabric is not a panacea to optical-switching needs. The initial appeal is also the source of some limitations of the approach. Photonic switches are transparent to the data going through them, which leads to challenges in management and grooming of traffic.

The typical model for an electrical switch in the core of the carrier's network is shown in Figure 2. As the fiber enters the central office from the street, there are a number of optical functions performed on the signal. These functions typically include amplification, optical performance monitoring, and demultiplexing the wavelengths onto separate physical fibers. All of these functions are common to optical-switching solutions.

The next step is to take each of the long-reach DWDM signals into trans-ponders. A transponder will receive the optical signal from the long-haul side, generate an electrical stream and, in turn, convert that to a short-reach (1,310-nm) optical signal. Overall, the transponder takes in the long-haul signal and converts it to a short-haul optical signal. This short-haul signal is then passed within the central office across the optical switch. The optical switch will then convert the short-reach signal to electrical, where it is switched to another port and converted back to an outgoing short-reach signal. These conversions result in these switches being called optical-electrical-optical (OEO) switches. The last step in the process is to take this short-reach signal through another transponder and generate the outgoing long-haul signal for multiplexing onto the fiber.

In this model, the primary function of the optical switch is to connect each incoming wavelength on a fiber to a different wavelength on another fiber. In general, any incoming stream can be switched to any available wavelength on any fiber. That's a very flexible architecture, which also has the benefit of performing regeneration of the signal and allows SONET-level performance monitoring of the stream. The price of this architecture results from all the optical-to-electrical (OE) conversions.

There are two models for using photonic switches in the long-haul network. Figure 3 shows the initial appeal of the idea. In this case, like the electrical model, the first stage is the optical signal processing. After the separate wavelengths are identified, they pass directly to photonic fabric, where they are switched to an output. The output would then pass to the optical signal-processing layer for transmission on the next link. This model clearly has fewer OE conversions but loses some key functionality.

Figure 3. This model clearly has fewer optical-to-electrical conversions, but also loses some key functionality due to wavelength blocking, the inability to replicate performance monitoring, and the lack of high-speed wavelength granularity.

With the all-optical model, there are three primary limitations, the most significant of which is that the outgoing and incoming wavelength must be the same. In a large network, that very quickly becomes a problem. For example, channel 35 on two different incoming fibers needs to go out the same fiber as the next hop in their path. That is not possible, since there can only be one channel 35 on the outgoing fiber. It's just like saying you can't send two TV stations on the same channel. So one of the signals will be allowed to pass, but the other signal cannot use the same route. That is known as wavelength blocking.

The next problem with this simple model is that it cannot accomplish performance monitoring at the SONET layer. This function is not easily replicated at the optical layer. The third significant problem with all-optical is that switching only occurs for the whole wavelength. Most typically, carriers want to switch at OC-48 in the core of their networks. If the transmission speed on the fiber is OC-48, then that works. But it is more common for the transmission speed to be OC-192 in the core, and that is not as granular as the carriers would like. The situation only gets more difficult when transmission speeds increase beyond OC-192. (Note: We are ignoring the case where multiple wavelengths are switched by a single mirror. That is a different application.)

One further consideration is transmission-link design. In the simple photonic model, the path length from transmitter to receiver will vary depending on the switching actions. For example, a single hop path of 900 km may become a five-hop path of 3,000 km after a switching event-for protection or for traffic routing reasons. The long path may be over completely different fiber. The implication is that all transmission links must be designed for worst case, often requiring ultra-long-haul techniques. That adds cost, and there is a tradeoff to be made between distance and the bit rate that can be transmitted down a fiber. As a result, the network may be constrained in throughput or size.

Given these challenges, it is most common for the carrier to consider using photonic switches in the more comprehensive method. In this case, the transponder function is re-introduced before the signal reaches the photonic switch. That not only prevents the wavelength blocking issue, but once the signal is converted to electrical, with the addition of further electronics, the performance management and switching granularity problems can be solved.

Of course, introducing the transponder eliminates the two benefits of the photonic switch that gave it its initial appeal. The OEO conversion is not eliminated and the solution is no longer transparent to the bit rate of the optical signal on the line. The transponder will be bit-rate-dependent.

With the transponder introduced in the photonic-switch solution, which will be the usual practice, the comparison between the two switch fabrics largely comes down to comparing the switch features. Specifically, deciding between the two types of fabric would involve tradeoffs of cost, scaling, and reliability. Regarding cost, it is typically expected that electrical-switch solutions are more economical where the stream being switched is below OC-768, which involves all cases today.

Scaling to larger switch sizes was assumed to be a benefit for optical switches. Up until recently, there has been a popular expectation that electrical-switch fabrics would only scale to 512 OC-48 ports-or possibly 1,000. However, in the first half of this year, at least three vendors have announced much larger switch fabrics based on electrical cores-one scaling as large as two million OC-48 ports. Reliability is still an open question for photonic fabrics. Some of the photonic-switch fabrics are based on micro-electromechanical systems (MEMS), which have not been proved for long-life operation, since the technology is still fairly new.

Given all of these tradeoffs, it would appear that electrical-switch fabrics would remain over the next few years as the preferred fabric for high-function switching of OC-48 and OC-192 streams. Specifically, that means carriers are still seeking to migrate current SONET architectures to a more scalable and cost-effective solutions, while maintaining the kind of management and control of SONET. For these carriers, electrical-switch fabrics will be the clear choice.

With the electrical-switch fabric, the SONET overhead processing is natural to support. The electrical-switch fabric can groom and support SONET tributaries at speeds slower than the speed of the optical signal on the fiber. The typical need for that in the long-haul network evolves when carriers prefer to switch at OC-48 granularity, but OC-192 is the preferred modulation speed of the wavelength on the fiber.

The photonic fabrics will be introduced in the carrier network, but in applications different from the electrical-switch fabric. One simple application for the photonic fabric is in fiber protection switching. With fibers carrying more and more wavelengths, switching from an active to a bypass fiber is best done using a photonic fabric. Doing that with an electrical-switch fabric would require demultiplexing all of the wavelengths on the fiber, performing protection switching, and multiplexing them all again. The photonic switch could take all the wavelengths on the fiber and switch them as one. That's clearly much easier, and in the case of many wavelengths, much more cost-effective than using the electrical-switch fabric.

The other application for photonic switching will involve switching OC-768 wavelengths. When carriers introduce OC-768 into their networks and want to switch the whole wavelength rather than the channels within it, then photonic switch fabrics may reach economic parity with electronic fabrics. It would require 16 ports on an OC-48 electrical switch fabric to achieve the same function as one port on the photonic switch fabric, assuming the whole 40-Gbit/sec signal is switched together. While electrical switch fabrics are considerably cheaper on a "per port" basis, that advantage erodes when the speed of the signal being switched increases to 40 Gbits/sec and above. All of the above applies to fabrics of 1,000 ports or less. For much larger fabrics, electronic fabrics have a significant scale advantage.

Both electrical- and photonic-switch fabrics will continue to evolve, which will change the landscape over time for the roles for these switches. In the past six months, there were two interesting developments in electrical-switch fabrics that will broaden its range of applications.

One development is the integration of the transmission system (DWDM system) as the direct input and output of the switch. For the central office with an electrical-switch fabric, the transponder equipment and the switch equipment were considered different products. The result was an additional OE conversion via the short-reach interface between the transponder and the switch shelf. But second-generation switches have re-thought this assumption. Why not use the DWDM system itself as the input to the switch?

The other development is scalability. While first-generation OEO switches were limited to 512 ports, second-generation switches have much better scalability. The leading solution in that regard can scale to 2 million ports, but more important, uses a pay-as-you-grow approach so the carrier can buy a switch of any size from eight ports up and add to the switch as required.

What is clear is that both electrical and photonic technologies will exist in the carrier network, and over the next few years, electrical fabrics will be used where management, control, and flexibility are important. Photonic fabrics will be used for switching fibers and very-high-speed streams.

Ian Wright is senior vice president of engineering for optical-networking products, Marc Schwager is vice president of marketing, and Rob Newman is executive vice president of market development at Altamar Networks (Mountain View, CA). They can be reached via the company's Website, www.altamar.com.