Building on IBM's leading high-speed networking switch
technology, researchers have proposed a new way to avoid
congestion and keep traffic flowing across networks.
In Brief:
As network computing grows at an ever-faster pace, the router technology for directing packets of data to their destinations has come under increasing pressure. IBM researchers, building on the company's advanced asynchronous transport mode (ATM) switch technology, are developing new means of integrating switching and routing to overcome communication bottlenecks. By creating special end-to-end connections that minimize processing, data packets can be switched onto dedicated high-speed links that carry them straight to their destinations.
In many areas of our lives, there is a level of detail most of us tend to ignore unless our interest is severely piqued. It is that willingness to enjoy a certain measure of ignorance that keeps us from prying too closely into the innards of the many mysterious objects that we use on a daily basis. All that changes, of course, when one of those objects does not work as satisfactorily as we would like it to. Suddenly, we want to know what is happening behind the scenes, even if there is little we can do to fix it.
Such is very likely the general state of mind of the average computer user in regard to the networks that enable us to send email, surf the Web, shop, bank, check our credit card balances, exchange files and data, and even carry on a phone conversation. As long as the bits and bytes come and go quickly enough, it is easy to overlook the process by which any single message gets from point A to point B. But as almost everyone connected to a network can
attest, there are times when even our minimal expectations are not met.
There are reasons for that, of course. And, while knowing about those reasons doesn't entirely make up for the delays, there is a certain pleasure to be had in knowing where the problems lie and what is being done to alleviate them.
So, what is the problem? In fact, there are two separate issues. The one that is most often discussed is that of bandwidth, which is basically a measure of how many bits one can transmit over a communications channel, such as a telephone line. There is another issue, however, that is less visible but ultimately more critical. For, even if one had unlimited bandwidth, it would be useless without a means to ensure that messages reach their intended recipients. The technology to do that is based on what are called routers. When the traffic through them becomes too heavy, the routers themselves become bottlenecks that can result in delays and even lost data.
Reducing the traffic jams on the Internet represents a major business opportunity. Not surprisingly, several companies have joined the hunt for a suitable solution, including IBM. "We are looking at ways to treat telephony and video data - which require a greater continuity of delivery - differently from email and data files," says Ton Engbersen, manager of the Advanced Networking Laboratory at IBM's Thomas J. Watson Research Center.
Researchers at IBM have proposed ways to overcome the routing problem by exploiting the company's lead in asynchronous transfer mode (ATM) technology. As a result, IBM is well positioned to provide the most effective integration of switching and routing for the networks of the early 21st century.
How the problem came to be
As computer networks have evolved, so have the means of directing the data sent over them. The Internet, which is really a collection of many interconnected networks, grew out of an effort to develop a communications network that could survive nuclear attack. As such, it has no vulnerable central switching node but rather provides many paths over which information flows.
A set of procedures for handling the data that covers everything from the high-level application programs to the bits actually sent across the wires is embodied in Transmission Control Protocol/Internet Protocol (TCP/IP). Messages or files sent across the Internet are chopped into pieces called packets, each of which carries an address, or header, based on the unique IP address of the recipient's computer.
The packets are not sent directly but instead make their way by a series of shortest-path hops. When they reach the final destination, the packets are reassembled in the correct order to reproduce the message for the recipient. In the early days of the Internet, general-purpose computers served as the forwarding centers. When a packet took the wrong path, explains Engbersen, the nearest computer checked its address and redirected it. By the mid-1980s, however, network traffic was beginning to overwhelm these systems.
It was also in the mid-80s that local area networks (LANs), such as the IBM Token Ring and Xerox's Ethernet, began to take off. These generated additional network traffic.
Software written by two computer scientists at Stanford University - who subsequently founded a company, Cisco Systems Inc., to exploit their work - provided a solution. The software formed the basis of routers - specialized computers that pick the best next hop among the many paths available to get a group of packets to another router closer to its destination. Routers ultimately replaced the general-purpose routing computers in the Internet.
The routing approach
When it receives a packet, the router first identifies the packet's destination from the headers attached to it. Then it looks up the destination address in a table, which allows it to determine the next nearest router that will move the packet toward its destination. Finally, it sends the packet on its way. To perform their mission effectively, routers must maintain current tables, which they do by continually exchanging, across the network, updated information on new addresses and address changes.
Soon after it arrived on the scene, routing became the dominant means of avoiding bottlenecks over the Internet and other networks. In response to the rapidly growing traffic, designers came up with specialized high-performance router computers with large numbers of input and output ports. And Cisco rapidly emerged as the dominant market leader, with 80 percent of the router market.
However, as use of the Internet increased exponentially, bottlenecks started to emerge within the routers themselves. "A router has to read the address on every packet coming in, find the address on a listing of 50,000 to 100,000 entries it keeps, then send every packet out on a link that takes it closer to its destination," says Engbersen. "It takes only millionths of a second to process each packet, but that finite amount of time adds up and the largest commercial routers start to choke when throughputs push them beyond their rated capacities."
Scaling up capacity by adding routers, including faster ones, helps. But such improvements will soon be unable to keep up with the rate at which network traffic is increasing. The problem is compounded by the requirements of emerging multimedia applications for differentiated service. In any case, the need for packets to pass through a series of routers significantly affects the speed with which packets proceed to their destinations.
Beyond routing
In light of that limitation, researchers at IBM and other corporations are taking an entirely different approach to overcome router-induced bottlenecks: avoiding routers as far as possible. The means for doing that is by integrating switching and routing in so-called switch-routers. A switch-router reads the destinations of incoming packets of data at the edge of the network and, in effect, directs them onto dedicated fast lanes along which they cruise uninterrupted to those destinations, avoiding any routers or other bottlenecks en route.
Fortunately, a high-speed packet-switching technology known as asynchronous transport mode (ATM) has been slowly emerging as a faster way of sending data (see "ATM Leadership"). ATM was initially conceived as a way to handle data-intensive applications - such as multimedia - that combines text, images, voice and video. Although the market did not develop as fast as expected, the technology is proving to be just what is needed for solving the router crunch.
Ipsilon, a networking company, was one of the first to propose a combination of routing with ATM switching. Called IP switching, the idea involves looking at the address data in the packets and trying to establish when a long flow of data between a source and destination is under way. At that point, software requests a connection through an ATM switch for that stream of packets.
Given its advantages, it is natural to wonder, Why not get rid of routing altogether and just use ATM switches? The simple answer is that switching is faster for some things but too costly for all things. "Switches must agree in advance about what needs to be done with different packets, and that involves an additional control overhead," says Roch Guerin, manager of broadband networking at Watson. "When the network includes millions of people all over the world, you quickly run into a scalability problem with switching to provide that control everywhere. Routers, though slower, are very good at what they do - handling addresses and deciding on next-hop strategy."
Nevertheless, Guerin and his Watson colleague Dilip Kandlur have taken the integration of switching and routing a major step forward by using a feature that is present in ATM but absent in IP networking. In an IP network, packet forwarding is typically done using a software-based lookup of the full IP address carried in the packet header. In ATM networks, this lookup is performed in hardware because of the much shorter connection identifier appended to each packet.
Once an ATM connection has been set up, the associated connection identifier not only permits faster forwarding, but it also enables easier service differentiation between packets belonging to different connections. Such a capability is key to supporting quality of service (QoS), which is crucial for voice and video. The result is that Guerin and Kandlur's approach to the integration of switching and routing does not have to try to determine which packets need a clear path at each router; rather, certain data streams can be accorded priority and assigned a high-speed route at the point where they enter the switched network.
The value of switching and routing
In fact, IBM's ATM technology already provides a first step toward such an integrated approach to switching and routing by virtue of its multiprotocol switching system (MSS). A card plugged into an 8260 ATM hub, MSS can handle packets in conventional router fashion or identify packets that require express delivery and send them nonstop through the ATM switch to their destination.
"The MSS provides quality of service for different packet flows, as requested, by establishing an appropriate switched connection for those packets. Those requests, which are expressed in ReSource reserVation Protocol (RSVP) messages, may be generated by your Internet service provider when you request high-level service or by an application such as your Web phone," says Kandlur.
"The MSS is a key innovation that supports switched virtual networking, multicasting and quality of service," says Erich Port, manager of networking at IBM's Zurich Research Laboratory. "Today's MSS product can handle 70,000 packets per second, which represents a throughput of about 35 megabits per second."
Even at that speed, the MSS processors could potentially become a bottleneck because of the time needed to determine whether the data packets are QoS packets or not and then forward them in the appropriate fashion. Guerin and Kandlur have therefore taken the technology one step further. They have put smaller MSS processors between each link and the switch. That distributes the load and allows for a far greater amount of data to be handled in this fashion.
Work with a prototype suggests that this form of integrated switch router can easily be scaled to very high throughputs with existing ATM technology by establishing multi-point-to-point connections. At the same time, following the lines of IP switching, efficient paths for all packets through a network built with ATM hardware can be achieved, further enhancing the throughput. Together with Cisco, IBM is working with the Internet Engineering Task Force to make this technology, invented in Research, a standard. And IBM's Networking Hardware Division is working on means of incorporating the advance into its first integrated switching and routing product.
"Compared with commercial routing, we believe ours is a much less expensive approach," says Engbersen. "The integration of switching and routing constitutes a game change from software-intensive, relatively expensive routers to a market with multiple players offering products with increased capacity and at lower costs. We are on a path where processing power is no longer the limiting factor because now you can combine routing and hardware-based switching."
Michael Sinclair is a freelance science writer based in North
Guilford, Connecticut.
More Information:
ATM Leadership
Making the Internet Click
ATM Leadership
"The way to beat conventional routing is by doing things at the ATM level because it is inherently faster," says Doug Dykeman, a scientist at the Zurich Research Laboratory. "There's less delay, less processing involved, and it's going to cost less if done right."
Over the past five years, the Zurich laboratory has carried out significant development of ATM (asynchronous transfer mode) technology and switching products. "We invented the mechanism that establishes the path between ATM switches," says Zurich's Erich Port. "Our work evolved into the Private Network-to-Network Interface. This allows you to network multiple nodes together to find the right route automatically. And standards that we helped develop permit interoperability between ATM
switches from different vendors."
Effective ATM switching means more than shuttling cells at top speed. It needs to cope with different types of traffic, at different data rates under different loads. "The whole reason ATM was developed was to support different types of application - video, telephone, computer data, whatever you will," says Dykeman.
The key to the necessary flexibility is PRIZMA, a high-speed ATM switch invented by a team led by Ton Engbersen (see Making the Internet Click) and colleagues at Zurich. PRIZMA sits at the center of a commercially available ATM hub, the 8260, designed at IBM's LaGaude Development Laboratory in France. Able to switch at a rate of some 6 gigabits per second, four PRIZMA chips can be linked to provide a capacity of more than 24 gigabits.
Making the Internet Click
In the beginning of his 16-year career in IBM Research, Ton Engbersen worked on signal processing and chip design. Successful efforts in such areas as debugging chips and switching signals led him inevitably to asynchronous mode transfer (ATM) technology - an area of specialization that, he asserts, "will be the underlying technology for the future of the Internet."
For the past year, Engbersen has taken on a new role: helping IBM's Networking Hardware Division to exploit the company's synergy. "We are actually at the forefront of network-centric computing and developing Internet hardware," he explains. "We must ensure that IBM has the right products to go into the intranet and Internet environments." A significant part of that mission is defining and building the next-generation Internet switch/router, a task that is now well under way.
Engbersen's interest in switching applied to networks started in 1987, when he headed a group that helped to implement what is now known as the tree switch. Based on the requirements of potential users of that switch, the team came up with an idea that eventually became the PRIZMA, short for Packet Routing in Zurich Modular Approach. Zurich transferred the ATM switch technology to IBM's Networking Hardware Division for commercialization two years ago.
Since then, researchers have worked on ways of further scaling the technology, by creating faster PRIZMA chips that can be combined to attain much greater throughput. Says Engbersen: "I think the biggest thing you achieve with PRIZMA is that you have a building block with which you can engineer a switch."
In his new role, heading a group of about 60 people, Engbersen faces the challenge of making intranets and the Internet click with ATM. "We have a couple of very good ideas," he declares. "The work is not simple, but we're getting there."
For more information see: http://www.networking.ibm.com/nsw/aris.html
