When we pick up the phone or launch a browser, we expect to be able to connect to someone or something. Usually, it happens without incident. And, while we may occasionally bemoan the lack of bandwidth, we rarely stop to consider the largely hidden world of wires and fibers, routers and switches, chips and subsystems, let alone communications protocols, that allow us to connect at all.
As more and more people discover the benefits of email, Web collaboration and especially e-business, the need to ensure the continued evolution of that largely invisible communications infrastructure will only increase. In particular, the growth of pervasive computing and mobile networking have raised the expectation that soon it will be truly possible to communicate with anyone, anywhere, anytime on any device.
That, at least, is the goal. Of course, anyone who has ever wanted to send or check email while cruising, say, at 35,000 feet over the Atlantic knows there are still some pretty important exceptions. But now, a joint research project between IBM's Zurich Research Laboratory and Boeing is developing the means to extend mobile networking to new heights, not to mention other mobile platforms such as ships and trains.
The solution is based on asynchronous transfer mode (ATM), a protocol -- or set of rules -- for sending voice, images, video or other kinds of data over high-bandwidth networks. In a parallel effort, the Zurich researchers are participating in a project to support mobility in a worldwide ATM network for the U.S. Department of Defense. The goal in both projects is to create a nearly seamless connection between the mobile and ground-based ATM networks. If done properly, the fact that some parts are mobile will be hidden; from the perspective of the end user, there will simply be one network.
Getting there is far from easy. For one thing, a mobile solution has to cope with the low-bandwidth satellite links through which connections between mobile and land-based ATM switches would typically be made. ATM technology is ideally suited to such a situation, because it uses bandwidth so efficiently: if an ATM connection does not use all the bandwidth it reserved during the initial set-up, the extra capacity is automatically made available to other applications.
Another challenge for mobile networking is to remain connected as a plane or ship moves, especially from one satellite link to another. Mechanisms to ensure smooth handovers are essential. It is also necessary to give priority to the most important connections if bandwidth suddenly shrinks. The Zurich approach would do just that. "When a commander picks up the phone, we want the call to go through and be handed over with high priority, even if it means the connection of a soldier playing remote Tetris is dropped," says Zurich researcher Laurent Frelechoux. "Our solution is aimed at ensuring that these high-priority connections will never be lost unless the bandwidth shrinks to zero."
The basis for meeting such requirements already existed in the ATM routing and signaling protocol called private network-to-network interface (PNNI), which grew out of earlier work at Zurich. PNNI enables the ATM switches -- the systems that compute a path through a network from the source to the destination every time a connection is requested -- to maintain the addresses of all the devices attached to the network. PNNI periodically floods the network with information about what resources have been added or dropped, enabling the address tables in the switches to be updated.
To avoid an explosion of update information as a network grows, PNNI is based on a hierarchical principle. Groups of ATM switches, called peer groups, are represented by what is known as a logical group node, which is hosted by one of the machines in the group. For users outside the peer group, the logical group node provides a summary of critical information, such as reachable addresses within the peer group and the links and bandwidth available to other peer groups.
As part of its mobile ATM solution, the Zurich team introduced the concept of a mobile logical group node. "Such a node," says Frelechoux, "can move within the PNNI hierarchy to reflect the current location of its corresponding mobile network." So if a plane or ship that has been using an access switch in California has now reached Japan, its ATM cells will be routed to a local access point switch. These changes, however, are apparent only to a small number of mobile-aware switches in the PNNI hierarchy. The switches in the static network do not have to know the actual location of the mobile networks.
The Zurich team also extended the ATM rerouting protocol -- which in a static network would deal with failed or overburdened switches -- to manage the handover of connections as a mobile network moves from one access point to another. "Our solution begins the handover when it becomes apparent to the satellite modem, which monitors the quality of the transmission, that an existing link is fading," says Frelechoux. "During that time, the connections on the fading link are rerouted to a new link, and the flows of data are switched from the old link to the new one in such a way as to minimize the loss of data. Also, higher-level protocols ensure that any high-priority data that is lost gets re-sent."
While the solution has not yet been implemented aboard aircraft, a prototype running at Zurich simulates the mobility features. In September, IBM will demonstrate, as part of a Department of Defense experiment, how a group of mobile ATM networks can be joined to a static ATM network.
(2) A LIGHT TOUCH
The protocols that allow one to seamlessly communicate with someone in the next office, across the country or on the other side of the earth are useless without a physical medium to carry the data. For high-bandwidth backbone networks, and increasingly for local area networks, the medium of choice is optical fiber.
Although laboratory demonstrations have shown that a single fiber can transmit terabits (trillions of bits) of data per second, the electro-optical circuitry needed to convert electrical signals into optical ones and back at that rate is too expensive for most practical applications, points out ZRL physicist Gian-Luca Bona. Currently, conventional CMOS circuits can support 2.5 gigabits (billions of bits) per second (Gb/s), and more elaborate circuits made of silicon germanium, 10 Gb/s. At the high end of cost and performance, devices based on gallium arsenide can push that number to 40 Gb/s.
There are two other fundamental ways to attain higher bandwidths in optical fiber, says Bona. The simplest is to combine two or more fibers. But that is only practical for short distances, because of the cost of the additional fiber. A more efficient approach, known as wavelength division multiplexing (WDM), uses a single fiber but multiple colors, or wavelengths, of light -- one for each separate data channel. "The technique works," says Bona, "because, unlike electrons in a copper wire, photons of different colors in a fiber don't interfere with one another."
Currently, up to 40 channels, or wavelengths, can be handled in a single fiber, although as many as 120 have been demonstrated in the laboratory. The more wavelengths one uses, the more difficult and expensive it is to discriminate between adjacent ones. "The real limit is how much you want to spend to send data from point A to point B," says Bona.
In 1995, for example, IBM introduced the 9729 optical fiber multiplexer, which can send 10 different colors in each direction over a single fiber. The different wavelengths are separated by multiplexers, devices consisting of mirrors, prisms, gratings and other optical components aligned to within 50-nanometer mechanical tolerances. Each of these marvels of "watchmaker technology" costs thousands of dollars. While that is a reasonable expense for companies that need to back up mission-critical data at a remote location, it is not competitive in the everyday world of network technology, where the cost of the components for connecting to the network is pegged at about 10 percent of the price of the server itself.
That realization led Bona and his colleagues to search for a low-cost solution based on silicon-chip technology. Specifically, they undertook to build devices based on planar optical waveguides formed on a silicon wafer using standard optical lithography and chip processing. The optical waveguide devices provide control over different wavelengths, so that data can be sent and received by machines assigned to particular wavelengths.
The devices were developed as part of a project known as the Corporate Optical Backbone Network (COBNET), funded by the Advanced Communications Technologies and Services (ACTS) agency, a European agency similar to DARPA in the United States. In a field test of COBNET technology in late 1998, the fiber-optic network consisted of two ringlike structures interconnected over a wide area network. Each ring linked several communication network nodes (up to 12, each representing, say, a different building). One node served as the switching and systems management center. All light carried in the ring passed through this central node, where it was "demultiplexed," or split into the separate colors corresponding to the different data channels (up to 12 in a ring). It is in this same node that data-channel paths are assigned and the various wavelengths are multiplexed together before being sent to the fiber-optic network.
When the different-colored signals arrive at a given building, the wavelength carrying data for that building must be dropped off, while the data destined for other buildings on the ring or elsewhere must be added to the light stream. That add/drop filtering function is provided by Zurich's planar waveguide device, through which light entering and leaving the building is sent. The device has two inputs and two outputs. One input contains all the different-colored data channels; the other is used to add, or send, data using the color assigned to the building. Similarly, one output consists of messages to be dropped off at the given building, while the other contains the other colors, which pass along unchanged.
The adding and dropping of specific colors in a device exploits the wave nature of light. When waves of the same color interfere so that the troughs of one match up with the peaks of the other, they cancel each other out. Similarly, when peaks of one wave match up with peaks of another, they reinforce one another. That phenomenon is the basis of special components known as Mach-Zehnder (MZ) interferometers, which are arranged in series on a chip. By electrically heating part of the MZ elements, they can be tuned so that different colors can be added or dropped as required.
To fit the series of MZ interferometers on a chip, they had to be designed with sharp bends, which forced the Zurich researchers to find a way to make the light follow this curved path. They developed a novel technology in which a layer of silicon oxynitride is deposited between two layers of silicon dioxide on top of a silicon wafer. Because silicon oxynitride has a higher refractive index than silicon dioxide, the light is trapped in that layer. That enables the light to travel around "corners" with a radius as small as 1.5 millimeters, one-tenth that of competing technologies.
The combination of functionality and low cost makes the planar waveguide technology a unique optical solution -- and not just for networking. "Optical buses between high-end computers could support data streams of a gigabit per second or more per channel," says Bona. "And the next-generation Prizma switch chip will have a similar or even higher bandwidth. Packaging constraints limit the number of high-data-rate input-output pins, however, and an optical solution based on our technology is a possibility."
More information
(3) SMARTER CHIP
Optical fiber, the backbone of most wide area networks, plays the same role for data as railroad tracks do for freight. And just as the freight must be suitably loaded onto railway cars before it can proceed to its destinations, so cells or packets of data must be properly "framed," or prepared, before they can be put onto the fiber. Not all the data traffic is of the same type, nor does it all travel at the same rate, and up to now the chips that frame data have had limited flexibility in the way they handle different types of traffic.
A novel chip to perform that function was recently developed by a team at IBM's Zurich Research
Laboratory and the IBM Microelectronics Division. Called SMART, the chip fully conforms with two
global standards: North America's Synchronous Optical Network (SONET) and its European equivalent,
Synchronous Digital Hierarchy (SDH). The chip's flexibility sets it apart from other framers. "SMART
was designed to be scalable," says Andreas Herkersdorf, leader of the Zurich team. Its modular design
allows a single chip to be configured in software to support a single data stream at 622 megabits per
second or partitioned to handle four data streams at 155 megabits per second, each
carrying a different kind of traffic --
asynchronous transport mode (ATM), Internet Protocol (IP), or traditional T1 or T3 voice or data channels. Four chips can be operated in parallel to support OC-48, a 2.5-Gb/s signal.
Unlike other SONET/SDH framers, the SMART chip includes the line interface functions, which normally are performed by a second chip. The line interface has to analyze the serial stream of bits from the fiber-optic channel that arrive at a rate of 622 megabits per second and to determine where the 1s and 0s are. That is not a trivial task, because temperature variations in the fiber can cause minute changes in the minimum duration of a single bit. "One has to reconstruct the duration of a 1 or a 0, a process called clock recovery," says Herkersdorf. "Because the data rate is so fast, this has normally been done with bipolar technology, rather than with the standard CMOS used in the rest of the chip." However, a team led by Dale Pearson at IBM's Thomas J. Watson Research Center devised a scheme that allows the clock rate to be recovered by splitting the original bit stream into two, each half as fast as the original. "Thus, by integrating the functions of two chips into one, SMART provides not only greater func
tion but lower system-level cost as well," says Herkersdorf.
(3) A FRESH LOOK AT INFRARED
For some technologies, the second time really is better. Infrared (IR) communication technology is a case in point. Although widely used for remote controls, it has not succeeded in becoming the basis of wireless networking.
But now, with IBM's recent announcement of a next-generation wireless protocol called Advanced Infrared (AIr), developed by IBM and other members of the Infrared Data Association (IrDA), IR has gained the functionality needed to beam it into mainstream communications. AIr allows any number of devices within 8 meters (26 feet) of each other to communicate without interference, according to Fritz Gfeller, of IBM's Zurich Research Laboratory, who was instrumental in creating AIr. Able to reliably connect one device to many, or many to many, at speeds of up to 4 megabits per second, AIr is ideal for supporting applications such as LAN access, shared whiteboards, multiperson games and ad hoc networking, in which people can sit down at a conference table and instantly form a wireless team.
A key component of IBM's AIr technology is the transceiver, the outcome of a collaborative design effort by a team at Watson led by Mark Ritter and members of the Infrared Wireless Communications Group in Toronto. The transceiver not only supports the new AIr standard but is backward-compatible with the earlier IrDA standards. To ensure high-quality, interference-free communications required a major system change to allow AIr's variable-rate transmission scheme to work with a well-known radio protocol. The idea, explains Parviz Kermani, who, with Mahmoud Naghshineh, led the Watson team that worked on the protocol and the overall system, is to ensure that each device has fair access to the medium without added design complexity.
Naghshineh explains: "A device that wants to transmit first reserves a short period during which everyone else is supposed to refrain from transmitting. That was a well-known technique in radio communication, but we provided a scheme to do this on top of the variable-rate modem designed by the Zurich team." The IBM scheme not only increased the transmission reliability by incorporating rate adaptation and detection in the protocol, but it increased throughput by means of a new burst reservation protocol.
But even that does not always prevent interference. "The problem is a lack of symmetry," says Gfeller. "In radio systems with a common antenna for transmitting and receiving, symmetry is generally given, so that any station that can interfere with one that wants to broadcast can also receive the request for silence. In IR devices, however, the LED transmitter and the photodiode receiver have different angular characteristics for emission and reception."
In certain situations, that imbalance makes it possible for a device to miss the call for silence even though it is near enough to cause interference. The solution, says Gfeller, involved "parity" -- that is, in each device, the sending power and the receiving sensitivity are matched, so that the protocol can be reliably implemented.
AIr achieves its highest data rates with a clear line of sight between devices. When such a direct connection is lost, a variable data rate scheme allows lower-speed communication with diffuse and reflected signals. The maximum speed for a given condition is automatically determined by counting the errors in a transmission. When the errors exceed a certain level, a signal is sent back to the transmitting device to slow its data rate. The sender then transmits each symbol up to 16 times, overcoming noise with redundancy. "Our goal," says Zurich researcher Walter Hirt, "was to allow
the signal to degrade gracefully under adverse conditions rather than be completely lost."
IBM is the first company to offer AIr products. "That speed to market is a tribute to the superb teamwork between the labs that worked on this project," says Brian MacNeil, manager of the Toronto group. "It has allowed us to provide a significantly enhanced infrared technology that is compatible with over 80 million current IR devices and that will remain compatible with the faster IR devices of the future."
More information
Rowan Dordick is the editor of Think Research.
