Internet protocol in the WAN

The Internet is booming, and expectations are high for electronic commerce and Internet telephony applications. The Internet is expected to reach more than 200 million users and more than 60 million hosts by 2000. Yet the growth, so positive for business, has exposed the limitations of the connectionless, router-based networking paradigm.

Routers were not intended to handle such large-scale networking demands. With limited port capacities and no real mechanisms for intelligent networkwide traffic management, the routers that form the Internet's backbone networks are becoming bottlenecks.

Some industry experts suggest that we can keep the existing router-based WAN philosophy by throwing large amounts of bandwidth into network connections. However, history teaches that applications will eventually occupy all available bandwidth. With this in mind, a WAN networking scheme based on over-provisioning is not the best choice.

The wise service provider manages growth with intelligent traffic management that allows for maximum control over all network bandwidth resources. Although it may not eliminate the need to add larger pipes, this method offers a way to maximize the use of bandwidth resources and realize significant operational economic benefits.

In addition to the need for more bandwidth, industry consensus is growing that a more robust IP networking infrastructure should be created to support the demand for higher quality of service. Networkwide QOS is needed to deliver priority service to higher-paying customers or to deploy voice or video services with data.

Some interpret QOS to mean network bandwidth and user priority control. Others define QOS as controlling network traffic transit delay and delay variation. QOS encompasses these things and more, but whatever the definition, service providers want to use QOS as a basis for offering various classes of service to different segments of end users. By doing this, they can create different pricing tiers that correspond to the QOS levels. That might become one of the best ways to offer new revenue-generating services in public data networks.

To work properly, QOS has to be provisioned desktop-to-desktop. A service provider can provision both the access and backbone aspects of its network to support multiple QOS levels. Providing QOS on an end-to-end user basis, however, is a standards issue, and the jury is still out.

It is easy to become absorbed in the industry debate over what WAN solution best delivers bandwidth and traffic management capabilities along with QOS controls. The most prominent debate in media and industry forums today is Internet protocol (IP) vs. asynchronous transfer mode. In 1993, the major debate was routers vs. switches, and we can see where that led.

It's not the same for IP vs. ATM. For one thing, the Internet growth factor did not become a front-page item until recently. For another, IP and ATM are apples and oranges. The questions should be connectionless vs. connection-oriented networking and traffic flow vs. topology-based connection establishment paradigms where Layer 3 IP traffic meets Layer 2 ATM cells.

The connection question Connectionless networking, born in the LAN, is the capability of having information move between network elements without a preconceived path between source and destination. IP networking became synonymous with connectionless networking by association with the LAN and routers.

It seemed only natural then to extend the same capability into the WAN. But fundamental differences exist between the LAN and WAN that force a different take on the subject.

In the LAN, bandwidth is relatively inexpensive and deployment is typically over a small area. Because the overall distance between elements in the LAN is short, transit delay is typically not an issue. When congestion occurs in the LAN, packets are discarded. Because transit delay is short, the loss of packets is quickly determined. Transmission control protocol retransmits the lost packets to deter application problems. Retransmission, therefore, takes place quickly and is practically transparent to end users.

WANs, on the other hand, are typically deployed over longer distances and therefore transit delay becomes an issue for both controlling QOS and identifying the loss of packets because of congestion. More important, bandwidth is the most expensive recurring operational resource in the WAN. Every bit sent over the WAN has a defined cost. If packet discard takes place, the expense of retransmission can become significant.

The problem with connectionless networking in the WAN is that paths between traffic source and destination are undefined. Information is sent into the network without knowing exactly where it will go or how and when it will get there. If data is discarded in an IP connectionless WAN because of congestion or network-element outages, it takes time before the source recognizes the need to retransmit. Meanwhile, traffic is still being sent into the network over the same link and possibly other links, potentially recreating the same problem. The costs, including the physical cost per bit retransmitted and the impact on the entire network, may be substantial.

Connection-oriented networking, normally associated with ATM and frame relay switching, provides the means to establish a predefined path between source and destination. Predetermined data paths enable greater control of network resources. Over-allocation of bandwidth, or statistical overbooking, becomes a conscious, engineered decision-not an arbitrary side effect.

Connection-oriented networking offers a more deterministic way to respond to changes in network status than connectionless networking does. Policing and traffic shaping, networkwide resource allocation for class-of-service provisioning, deterministic transit delay, and delay-variation measurement and control to establish multiple QOS levels can be achieved on connections in the form of real-time traffic management.

Perhaps the most significant benefit is the capability to perform per-connection,closed-loop, rate-based flow control across the network. This can be accomplished only with connection-oriented networking, where path information is used in reverse to form a feedback loop for congestion measurement and notification. These values have been designed into ATM networking, and thus, ATM has become a prime choice for the WAN.

Using ATM to switch IP traffic will become commonplace. The industry now recognizes a need to integrate ATM and IP tightly to create the most effective WAN environment. Of all the methods, many of which are proprietary, only multiprotocol over ATM and multiprotocol label switching are supported within recognized forums or standards bodies. MPOA comes from the ATM Forum; MPLS, from the Internet Engineering Task Force. MPOA, based on LAN emulation, is seen as more of a campus backbone solution. MPLS is being designed with the large-scale WAN in mind. Because ATM can be used to transport many different traffic types, WANs can be created to consolidate networks.

Converged networking Service providers typically have had to install multiple networks to meet customer needs. A large service provider frequently separates IP, frame relay, X.25 and ATM networks, in addition to a voice network and an entire transmission infrastructure from which it provisions leased lines. Also, to remain competitive, service providers are becoming obligated to change their equipment more quickly. The cost of buying equipment is significant; the impact on services and cost of operations/ownership, is even more so. Some observers suggest that more than 60% of the costs associated with modern data networking lies in cost of ownership.

Demand for Internet, intranet and extranet services is driving service providers to implement large-scale IP networks. But what about voice? The day is coming when voice will be a low-bandwidth "add on" to data traffic. Additionally, new data networking services such as closed-user-group virtual private networks offering consolidated networking options are on the horizon. Service providers can expect to face even more diverse connectivity requirements that encompass not only different types of physical interfaces and speeds, but also multiple protocols and the need for more network and service interworking (Figure 1).

New core networking structures of higher intelligence will be needed to converge data, voice and video and tie mobile/wireless and fixed networks to offer unparalleled networking options.

Because of its virtual path and channel capabilities, ATM is ideally suited for the backbone/core to build highly scalable hierarchical networks. From its inception, ATM was designed to handle multiple class of service and QOS levels. As public networks grow, they will need not only these capabilities but also controlled access. Here again, ATM excels. Surrounding an ATM core with multiservice switches accommodating ATM, frame relay and IP traffic, as well as provisioning for handling packetized voice and video, delivers a switched network infrastructure solution.

This switched network architecture is supported through the use of the intelligent network using SS7. Intelligent networking functions will evolve to support new services for data communications and the Internet in particular. Couple the switched network environment with synchronous digital hierarchy/Sonet and dense wavelength division multiplexing optical transmission networking solutions, and a complete converged networking environment emerges.

Connection establishment paradigms The key to success for large-scale IP networking lies in delivering the flexibility of IP routing with a switched packet forwarding mechanism that offers the highest possible performance and maximum network control.

MPOA, a flow-based connection-establishment paradigm, uses LAN emulation as a basis for its IP routing and forwarding functions. MPOA functions analyze IP data flows and use predefined algorithms to determine if an ATM connection is needed as a shortcut high-speed data-forwarding path.

If a flow qualifies for an ATM connection, and if none already exists to the destination, MPOA uses ATM switched virtual circuit signaling to set up an ATM connection to the other side of the network.

MPOA uses the next hop resolution protocol to map ATM network addresses. In effect, two addressing schemes (ATM and IP) and two routing schemes (routing information protocol/open shortest path first and private network-to-network interface) exist. Also, MPOA, LAN emulation and next hop resolution protocol require servers that must be configured and maintained.

Many issues surround the viability of large-scale MPOA deployment in the WAN. The MPOA specification clearly targets MPOA for the LAN/campus backbone environment. Scaling issues, as well as the capability to easily accommodate IP multicast, may limit MPOA as an Internet backbone choice.

The IETF is developing an MPLS standard that is expected to emerge in early 1999. Although MPLS is intended ultimately to be applied to many link layer protocols, the IETF is concentrating on ATM as the first link-layer application. MPLS is also being designed to work in an intrinsic IP environment.

Tag switching and MPLS are topology-based connection-establishment paradigms that use standard IP routing techniques and establish connections over the defined transport medium before data flow. Tag switching and MPLS use a new protocol for communicating that connections have been established between network nodes. They use tags or labels as the connection identifier. Using ATM as the link layer protocol is a good fit because the ATM virtual path identifier/virtual channel identifier cell address becomes the tag or label in relation to IP address prefixes.

Tag switching/MPLS with ATM allows for the capability to scale from the campus to the largest of WANs. A typical router network requires each node to perform hop-by-hop routing and forwarding techniques (Figure 2). The result is complicated routing tables and scaling and performance problems.

In a network that supports tag switching/MPLS, intermediate switching nodes create a direct connection to the edge of the network, which appears to the routers as a single hop (Figure 3). The tag edge router and the tag switch router could be the same device. Functionally, the tag edge router creates the tag switching connection through the network.

It can perform both normal Layer 3 forwarding functions and establish tag connections. The tag switch router must support only the minimum protocol requirements to establish and maintain connections. Because the intermediate network nodes are operating as an interior network, interior routing protocols can be used to build the routing tables and standard exterior routing protocols. The IP routing table in the backbone is simplified, and the IP packet forwarding performance is boosted significantly.

Topology-based connection establishments could result in many virtual connections. The number of virtual connections to be managed can be limited via virtual connection merging, which allows many such connections to be merged into a single one going to the same network destination. Multiple levels of QOS can be supported by defining tags or labels that have QOS attributes. The mapping from IP-defined QOS levels to ATM QOS levels using tag switching/MPLS is more straightforward because of the easy relationship formed by defining tags or labels.