Data vendors ride the next wave in fiber network switching

Data vendors ride the next wave in fiber network switching

How will the role of switching change as fiber-optic networks become more ubiquitous? A number of "traditional" equipment suppliers, or so-called legacy vendors, have stressed the coming of all-optical switching in the not-too-distant future (see Lightwave, July 1998, page 63). But in the brave new world of data dominance in telecommunications, a new group of suppliers and carriers has emerged with an emphasis on the tremendous rise in Internet traffic. Accordingly, their view of switching is somewhat different. Below, a few of these firms offer their insights as to how profoundly the burgeoning data-communications revolution will affect switching`s place in this new environment. As a complement to these views, some well-known academic researchers provide a peek at trends in switching research.

Lee Branst

A change is gonna come

The implementation of switching may change drastically, according to Jeff Kiel, director of product marketing, Core Switching Division, Ascend Communications (Westford, MA). "Switches used to be looked at as black boxes. Now they`re looked at as [part of] networks," he says. Kiel believes "we`re seeing almost the total collapse of the multiple, discrete networks into more streamlined networks with fewer boxes and layers inside the network." He says that bringing transmission functionality into switches will obviate the need for an entire class of products from legacy vendors (see figure). These switches will connect directly into the optical networks that are now being made available through the deployment of wavelength-division multiplexing (wdm) equipment.

"Of course, legacy vendors aren`t sitting still either," he acknowledges. "They`re trying to adapt their products to this new world, this new public network." But he claims that it will be more straightforward for vendors with experience in building data networks to add transmission functions to their data switches than it would be for a transmission vendor to make its products more successful in the data-communications space.

Kiel sees future networks as comprising a cell-based network layer running over the optical layer. "The edges of the network are where you bring in the packets as well as the legacy tdm [time-division multiplexed] and voice-type traffic. From a carrier perspective, we will see more networks being built out based on atm [Asynchronous Transfer Mode] cores, regardless of all the hype that`s going on with other technologies."

He opines that advances in the optical layer will come at the same time. "Today, the wdm products are nothing more than a fiber-multiplier technology," he explains. "As a result, capabilities such as protection, restoration, and physical layer management fall upon the switching layer." Vendors will reduce the cost and complexity of these networks further as more capabilities start to fall on the optical layer. He feels that as more capabilities, such as protection, move into the optical layer, switches can be made even more cost-effective and provide more tightly coupled network-management integration.

A two-layer network structure is the way it will be, agrees Graeme Fraser, vice president of engineering and general manager of the optical internetworking business unit of Cisco Systems (San Jose, CA). He believes one layer will be an internetworking layer focused on Internet protocol (IP) for more intelligent backbone networks. This layer will be connected to the optical layer at OC-48 (2.5-Gbit/sec) rates and higher.

Fraser foresees huge changes in switching in the optical networking layer, which today "is just comprised of point-to-point, long-haul wdm systems. But this will evolve into add/drop components as well as more-sophisticated wavelength crossconnect systems, which will act essentially like a static or pseudo-static connection map on which the IP network is built." While today`s state-of-the-art switches and routers have from eight to 16 ports capable of OC-48, he projects "they will migrate eventually to hundreds, if not thousands, of ports, as these IP switches become the next-generation crossconnects in an IP network. So the capability will exist to support hundreds of connections or wavelengths in a software-controlled, optical patch panel to build up a mesh of optical circuits upon which the internetworking layer can sit."

He predicts that this year will see the first forays into IP directly over wdm, as some leading-edge service providers connect a 2.5-Gbit/sec IP stream across the country "using a wavelength from end to end without using any, say, sonet tdm multiplexing gear in between." On the other hand, he thinks this year will be fairly sparse in terms of connections: "There might be one or two large rings around the continental United States. Next year it will be tens of circuits. The year after, perhaps hundreds of OC-48s going across the country, for example, between city peers. So the equipment needs to scale from a few interfaces to hundreds over the next couple of years."

Next year, Fraser predicts, "is probably when we`re going to see a big push to use optical in metro areas." There will be several approaches, including the provision of metro area add/drop rings. In addition, where fiber in the ground is plentiful, systems without wdm will be connected directly to the fiber. "In terms of pushing it out to the edge of the network," he says, "I don`t think the economics play out this year, but maybe next year or the year after, it will start to look more attractive."

New demands need new responses

Says Dave Newman, director of engineering, Switching Products Div., Bay Networks (Santa Clara, CA), "It`s clear that the demand for bandwidth is being driven by the 100% annual growth in data transmission." He says three basic technologies have been deployed to meet this demand. First, the data-transmission rate on each fiber is being scaled up, from OC-12 (622 Mbits/sec) to OC-48 and in some cases to OC-192 (10 Gbits/sec). Second, wdm has been applied to enable multiple independent data "channels" to share a single fiber-optic link. Third, the use of optical amplifiers ("I could fill a book on erbium-doped fiber amps and pump lasers!") has enabled these first two technologies to be used over longer links. In addition, the ability to provision and manage these new fiber networks is leading to the development of new classes of crossconnects, switches, add/drop multiplexers, and control software.

Newman delineates three general classes of optical switching, each with its own strengths and weaknesses:

Pure optical switches, in which the data paths and control signal remain purely optical. "Techniques for this have been proposed at several research labs, and some small demonstrations have been made. But in general this technology is still years away," he says.

Hybrid optical/electrical switching (where the data path remains optical but the control is electronic), which is possible today but only for relatively small port counts (16 or 32). "Also, today`s devices are relatively slow [on the order of milliseconds], so they are usable for circuit restoration and protection but are not applicable to packet-by-packet switching," Newman says. "There is, however, a lot of work going on in this field."

Electronic optical switching, "which sounds like an oxymoron but is defined as a system where the external interfaces are optical [OC-48 or OC-192], but internally both the data and control are handled as electrical signals. This is the area where we are seeing rapid development because systems can be built from existing components," Newman explains. He points out that several startup companies have formed to investigate this area in addition to the development efforts underway at established equipment manufacturers.

As to the types of networks involved in these changes, Newman says these technologies currently are being focused in wide-area, long-haul networks. "But clearly, as equipment costs come down, they will migrate into the metro area," he offers. "Some of the techniques also have potential applicability in enterprise and campus networks, but costs need to fall."

Wayne Price, manager of network development, Williams Networks (Tulsa, OK) provides a carrier`s perspective. Williams is building a new network scheduled to be in operation by the end of this year. Because the company was starting anew, "we didn`t have all the legacy baggage of sonet and crossconnects, etc.," Price says. As a result, Williams can build its optical network using dwdm technology, allowing it to "get services in the most efficient way and interface directly to the optical network. So we have more than just an atm switch: a multi-service aggregation point where we can bring in not only the traditional legacy-type traffic like private-line data and tdm-based traffic but also frame and IP." He thinks that has reduced startup costs and will reduce operational costs in the long run.

Price says Williams does not see the all-optical network as an option today because the technology is not mature enough to provide the necessary network reliability and accommodate the different types of services that Williams wants to offer. However, he says, "we would like to see [the development of] the ability to manage the dwdm infrastructure and to handle restoration at the optical level." He believes this development will allow carriers to become even more efficient.

"When the optical technology becomes reality," he predicts, "what we probably will do is employ optical crossconnects between our switching layer and dwdm infrastructures. Then we would use those optical crossconnects to manipulate large chunks of bandwidth." In addition, Price is looking for tunable lasers that will allow interfacing at higher speeds, plus intelligent optical nodes that will perform routing in the network.

Academic research

In addition to the equipment suppliers and carriers, researchers in university environments also have a keen interest in the future of switching for fiber networks. While they have an interest in commercial research and development, their work focuses on topics with less certain or less immediate payoffs.

Katie Hall of the Advanced Networks Group, mit Lincoln Laboratory (Cambridge, MA) is one such researcher. "We think that in the future, networks will be dominated by computer communications rather than voice. And those are much more bursty than voice," Hall says. She feels that such networks are served more efficiently by tdm or packet-switched approaches, "so in the next couple of years we`ll see a growth in the technologies and architectures to support very high-speed, optical time-division multiplexing."

She further elaborates, "Today, if you tell most wdm people that they`re not making the most efficient use of the channel, they would agree but say they`ll just add another channel. But in probably five to 10 years, they`ll want to make more efficient use of the channels they have, so that`s when we`ll see time-division multiplexing come in. Right now, wdm is incredibly powerful, commercially mature, and can solve a lot of problems. But in 10 years or so, we`ll see other network architectures coming to the fore."

Hall says there are two reasons why all-optical switching is going to be very important. One is because the switches themselves can be made to run very fast. "The other reason," she continues, "is that you`ll be able to propagate the signals in the optical domain without any kind of opto-electronic conversion." That`s very important, she asserts, because it`s very difficult to do electronic processing above rates of 40 or 50 Gbits/sec: "It becomes a problem of just getting the electrical signals on and off the chip. So at those very high rates it`s a transmission problem to do processing."

Hall feels there will be test-bed demonstrations in the next three or four years with very high-speed logic in the nodes demonstrating some type of switching, "People have already done some sorts of packet switching," she explains, "but typically that is reading a low bit-rate header on a packet, throwing a 2 ¥ 2 switch, and calling that a high-speed packet-switch demonstration. But in terms of a network where single stream rates are going 40 or 100 Gbits/sec and within the nodes there is either highly specialized electronics or optical switching to do functions like reading the address and recovering the clock, we`ll see test-bed demos of that within the next three to four years." However, commercial packet switching is much farther out, because the technological development of the optical sources and the clock recovery units is still much more at a research stage.

Even though the pace of development has been rapid in recent years, there are reasons why it has not been even faster, says Hall. "Part of the problem is that it`s enormously expensive to do any kind of high-speed tdm switching experiment," she explains. "It`s mostly because you`re trying to fit together commercial parts that have not been developed necessarily for that application. So everything is custom-made. In terms of the switching, the fiber has essentially an instantaneous nonlinearity, which means you should be able to go at any rate. But there are other problems due mostly to the distortion of the optical signal."

Is there a magic bullet? Hall is still waiting. "People believe that perhaps a terabit per second is possible with the materials we have now, but everybody`s always struggling against various problems with them," she says. "We`re hoping that someday this miracle material will show up, but we haven`t seen it yet."

According to Scott Hinton, switching researcher at the University of Colorado in Boulder, the quest for all-optical switching has been spawned by wdm. "With the start of the wdm revolution," he explains, "there has been much more interest in trying to keep the signal optical and avoid the conversion to electronics. I think the main reason for this interest is not speed limitations, but cost. Since the signals have been generated and received electronically, the electronics are not the speed bottleneck. The key is if optical switches can be made cheaper than electronic switches."

Optical crossconnects (oxcs) and optical add/drop multiplexers (oadms) are the start of optical switching, Hinton says: "These structures allow you to pick off streams of information and redirect that information." Unfortunately, these structures do not allow for individual control of packets and individual quanta of data. This type of switch requires the switching fabric to be able to look at headers and other information embedded in the data stream and then to make decisions on what to do with the data.

"Unfortunately," he laments, "optics is not very good for this in the time-based protocols that currently exist. With this in mind, it is my opinion that optical logic will not play a role in this phase of optical switching. It is my guess that the role for optics will come in the form of optical interconnects between high-performance electronic subsystems that will provide this individual bit-by-bit control." Therefore, he predicts that just as optical fiber is rapidly becoming the interconnect of choice between boxes, it will soon become the interconnect of choice within the box. He points to early manifestations of this trend that include work in parallel interconnects, optical backplanes, and smart pixel technology.

Hinton believes that optical switching will come first in oxcs and oadms, followed by parallel interconnects for transmission equipment, oxcs, and oadms. Next will come parallel interconnects in high-end servers and packet-based fabrics and, finally, optical backplanes in high-end servers and packet-based fabrics. He also believes that economic viability will arrive for 50-Gbit/sec parallel interconnects by 2000, for 100-Gbit/sec optical backplanes by 2003, and for 1-Tbit/sec optical backplanes by 2006.

"There are three levels of evolution related to time and difficulty," notes Ira Jacobs, professor of electrical engineering, Virginia Polytechnic University (Blacksburg, VA). "One, the capability of switching wavelengths is close to realization now, especially if switching times are not critical. Two, a few more years into the future is the capability to switch out time slots optically within a given wavelength while within a very high-speed optical stream. Three, even further out is high-speed packet-oriented optical switching. Here we want the capability to read headers on individual packets and switch those out on a packet-by-packet basis. There is research going on regarding optical time-division multiplexing up to 100 Gbits/sec. The problem is that of demultiplexing the signal. This is far off in the future. In this case the question can be raised of whether we ever will move into networks of this nature."

In the final analysis, the future, of course, is in the eye of the beholder. Yet one thing is certain: The future seems very bright to many people. As Lincoln Labs` Hall relates, "The number of people starting their own companies is huge. It`s the biggest topic of conversation at conferences nowadays. It`s almost not the new technologies, but who`s started their own company." q

The implementation of optical switches in a more centralized role would streamline networks, reducing equipment and complexity while allowing for more efficient use of bandwidth.