October 2001
 

The Future Of Packet Switching

BY TONY RYBCZYNSKI

AlohaNet was a satellite-based packet radio network, which tested some concepts that later formed the foundation of Ethernet. There were also several projects in Europe, such as one at the National Physics Laboratory in the United Kingdom and the Cyclades network in France.

These had some common technical attributes. Information was fragmented into packets of some maximum length, with a header containing an address field, and an error checking protocol and a checksum trailer to protect the data from corruption, or more precisely to identify when the packet was corrupted. Packet switching included the concept of speed conversion. A packet could be sent at say 1,200 bps and received by a high-speed device at 4.8 Kbps (that's not a typo!). An intermediate node (a commercial minicomputer) would receive a packet, check that it was received correctly, look up the destination address and a next hop table, and transmit it on the appropriate high-speed link.

From Theory To Dollars
Meanwhile, the commercial world (i.e., mainframe manufacturers such as IBM and the "seven dwarfs") had developed techniques to allow more than a single device to use expensive long-haul circuits and mainframe ports. To this end, they developed proprietary statistical or packet-multiplexing techniques, whereby the mainframe would poll each "dumb" device (think of it as a PC with a modem and display, but without processing and memory) to see if it had something to send and then authorize transmission of "blocks" of information in an orderly fashion.

In the first half of the 1970s, several service providers led by Bell Canada, France Telecom, and Telenet (acquired by what is now Sprint) identified a business opportunity to leverage this packet switching research work to establish commercial packet network services. They recognized that standards were key, if the mainframe manufacturers were to develop systems that would allow enterprises to use these public services. This culminated in the development of the X25 packet networking standard in 1976. In fact, at the International Computer Communications Conference in Toronto in 1976, Nortel Networks demonstrated the first purpose-built, standards-based packet switch that was designed for carrier networks and supported trunks at a state-of-the-art 56 Kbps.

The significance of these developments is that they established the technical and business viability of packet switching technology. They established the architectural principle of layered designs, codified in the Open System Interconnection standard, or OSI for short. The layering principle states that common communications functions are grouped within one of seven layers, each of which provides well-defined services to the layer above and uses well-defined services of the layer immediate below. This single industry-accepted principle enabled a string of technological advances in packet networking, with new technologies replacing the existing technologies thus preserving the investment in applications on the one hand, and lower layer networks on the other.

MAKING THE CONNECTION WITH IP
A lot of the networking developments over the years have been at Layers 1, 2, and 3. In the last five years, though, Layers 4 to 7 functionality has found its way into network devices. Layer 1, the physical layer, has evolved from analog to digital, from wired to wireless, from copper to fiber, and from 1.2 Kbps to 10 Gbps and beyond. Layer 2, the data link layer and the domain of frames, has the job to compensate for physical layer characteristics by ensuring that Layer 3 packets cross the physical layer without error. As the error performance of the physical layer improved, the needs for Layer 2 functionality diminished from error correction to error detection, leaving recovery to higher layers. Layer 3, the network layer and the domain of packets, is where a lot of the debates in the industry have been centered.

The early experiments in packet switching were based on a connectionless mode of operation, also known as datagrams (realized today as IP), whereby each packet carried the full address of the recipient. In fact, back in the mid-70s, Bell Canada proposed that the first packet standard be based on datagrams. However, commercial reality dictated that connection-oriented operation would be more amenable to the mainframe-centric networks and enterprise users, who basically wanted a better private line. Connections were also a smaller step from circuit switching for which service providers knew how to manage and bill. Hence was born the virtual circuit, which survives in frame relay and ATM (asynchronous transfer mode), and arguably was the first VPN (virtual private network) technology. Virtual circuits could be pre-configured or "permanent," or set up dynamically or "switched." A virtual circuit packet did not carry a full address, but rather a relatively short label that simplified the routing process, the label to be used being defined before transmission started (e.g., during call set up). An important attribute of virtual circuits (then and now) is that they are order preserving and don't duplicate packets in case of rerouting, both deemed important to minimally impact existing applications.

But there was more to this debate than just a question of connectionless or connection-oriented. The other key factors were all about addressing and more generally where the intelligence lies. The network layer is the layer that includes end-to-end addressing. Through X25, frame relay, and ATM, the service providers hoped to establish and control the numbering plan, much the way telephone numbers are used. But this never happened, one reason being that permanent virtual circuits were the dominant mode used for VPNs. VPNs were "dumb" pipes over which enterprises laid their private computing and communications environments, which had their own addressing and control environments (e.g., using IBM's System Network Architecture and TCP/IP). The operative words were "dumb" and "control." For example, even though X25 was designed as a 3 Layer architecture, VPNs running over X25 effectively used X25 virtual circuits as a Layer 2 replacement within the mainframe architecture, preferring to perform addressing, routing and flow control at the edge of the network under control of the user. The connectionless proponents likewise continued to pursue end-point intelligence running over a simple connectionless network, as in TCP sessions running over IP.

In fact, switched virtual circuit operation generally never became mainstream, though it was used by service providers to simplify virtual circuit provisioning and recovery. Three noted exceptions were: the first surfers dialed into X25 networks in the late 1960s and switched between various databases (e.g., National Library of Medicine) and early public e-mail systems; ATM switched virtual circuits were the foundation for ATM campus LAN operation; and switched voice used switched ATM virtual circuits.

Frame relay emerged as the second-generation wide area packet networking technology that was effectively a higher performance, much simpler X25 protocol. It provided VPN-permanent virtual circuits purely at Layer 2 and could better support the burstiness of LAN-based traffic sources. Switched virtual circuits were part of the standard, but have again seen little use. Frame relay also took advantage of the improved error performance of digital transmission, and the intelligence embedded in PCs. Being simpler than its predecessor, it could operate at speed initially up to T1 and today up to T3.

The evolution towards yet higher speed came with third generation systems based on ATM, a Layer 2 technology. Through the development of service interworking standards, it has become an extension of frame relay services serving larger sites requiring broadband access at OC3 and OC12 speeds (155 and 622 Mbps respectively). With ATM, all traffic is converted into short cells and again transported over virtual circuits. One of the founding principles of ATM was support of a broad range of classes of service, including voice, data, and video. When used as an extension of frame relay, much of the class-of-service richness of ATM has not played a role since frame relay was the lowest common denominator. ATM brought speed, but the cost was a hit in increased complexity. ATM has evolved to serve two distinct environments: a VPN offering as discussed above, and a networking technology.

As a major networking technology, ATM has been deployed by carriers within their access and backbone networks. Service providers have leveraged switched virtual circuits to simplify virtual circuit provisioning and recovery. Service providers have also leveraged ATM's rich QoS capabilities to support IP, transparent LAN services, frame relay/ATM, long-haul public voice, video, and even traditional private line services. In fact, some larger enterprises have followed suit, using ATM as a robust infrastructure for converged private enterprise networks. In the early 1990s, ATM was also used by enterprises as a campus backbone technology, but switched Gigabit Ethernet networks have since become the campus backbone technology of choice due to its superior price performance, scalability and simplicity.

FOURTH GENERATION LAYER 2 VPNS
Today, following the Internet revolution of the 1990s, the industry has come full circle, having established connectionless IP as the dominant networking protocol and IP addresses as the addressing standard. As a Layer 3 packet protocol and following the OSI model, IP can run on a range of Layer 1 and 2 infrastructures including Ethernet, physical pipes, and virtual circuits. In fact, there are two addressing schemes used by IP: 1) the public IP space used by service providers (each subscriber has a unique IP address), and 2) a private space used by enterprises (IP addresses are unique in an enterprise but not between enterprises). And of course, communications between these two environments is a requirement, to allow connectivity to mobile employees, telecommuters, partners, and remote sites via Internet-tunneled VPNs.

While technically feasible, many enterprises shy away from using the public Internet as the backbone technology for site-site connectivity, citing reliability, security, and performance concerns. In fact, all the enterprises really want is high performance connectivity for their private IP packets over a simple Layer 2 network. Given that IP is connectionless, nothing could be more simple than using a universal connectionless Layer 2 standard: Ethernet. Using a connectionless Layer 2 protocol avoids the complexity of having to configure and operate a mesh of virtual circuits to connect enterprise sites. In fact, Ethernet is becoming the fourth generation wide area Layer 2 packet technology providing what we can call Virtual Private Ethernets. This is not your father's Ethernet based on shared media 10/100 Mbps operation. This Optical Ethernet combines the simplicity and cost of switched Ethernet with the reliability and performance of optical technology, at speeds up to 10 Gbps. Labels are used to ensure virtual-circuit-equivalent isolation of customers. Optical Ethernet represents the first time that LAN/MAN and WAN technology is based on a common Layer 2 architecture, and therefore represents a significant opportunity for enterprises to rethink their deployment of routers, servers and storage in an extended campus network environment that goes across multiple sites and even to remote offices and branches.

So what's a packet? From a Layer 3 perspective it's IP. From a Layer 2 perspective, it is a frame relay frame, ATM cell or more recently an Ethernet frame. And what's the future of packet switching within the enterprise? IP and Optical Ethernet -- it's that simple, it's reliable, and it's fast.

Tony Rybczynski is director of strategic marketing and technologies for Nortel Networks' Enterprise Solutions unit. E-mail questions or comments to tonyryb@nortelnetworks.com.

[ Return To The October 2001 Table Of Contents ]