September 1998

BY TONY RYBCZYNSKI

Switching, by convention, has been regarded as a Layer 1 (or at best a Layer 2) function. A venerable form of switching, circuit switching, has been put to use for remote on-demand data access. Another form of switching, Layer 2 LAN switching, has been used to change the LAN paradigm. Whereas the LAN was once a shared medium (that is, 10 Mbit/s of ethernet LAN bandwidth shared across, say, 20–30 users), it has become a more manageable switched medium (that is, 10 Mbit/s of ethernet LAN bandwidth dedicated to each user). ATM switching combines some of the elements of circuit and packet switching to deliver low latency and high bandwidth. So what’s all this Layer 3 switching stuff?

LAYER 3 SWITCHING FOR INTER-LAN PERFORMANCE
Layer 2 switching delivers performance within a LAN; however, Layer 2 switching does nothing for the performance bottlenecks between LANs. Thus, a multi-LAN environment demands something beyond Layer 2 switching. One possibility is Layer 3 switching, which is also known as cut-through switching. The basic idea of Layer 3 switching, "route once, switch many," is being developed through various architectures. Layer 3 switching makes use of known packet formats (above the media access control, or MAC, layer) to make packet forwarding decisions at very high speeds. These Layer 3 LAN switches are pushing traditional routers out of the LAN into the WAN.

In wide area networks, Layer 3 switching techniques are also being applied to enhance performance and simplify the network, both in the Internet and in private router and enterprise network switch (ENS)-based networks. In the WAN, some of the objectives are to improve price/performance and enhance the traffic management of IP networks.

Since the late 80s, people have been building multiprotocol router networks, driven by the need for inter-LAN connectivity. These networks have served the industry well; however, as these networks have grown, so has their complexity. Enterprise users are now faced with significant challenges in supporting explosive IP growth ... and not just for data but for emerging multimedia applications as well. IP-based applications have become business critical, and so have the underlying networks.

LAYER 3 SWITCHING SCHEMES
Proprietary Approaches
There are several vendor-specific Layer 3 switching schemes. These architectures all interoperate with the outside world, using a broad range of standard protocols (such as Routing Information Protocol, or RIP, and Open Shortest Path First, or OSPF).

Standards-Based Approaches
Not all Layer 3 switching work is proprietary. In fact, two standards-based Layer 3 architectures have been or are being defined: 1. Multi-Protocol Over ATM (MPOA), standardized in 1997 by the ATM Forum; 2. Multi-Protocol Label Switching (MPLS), slated for standardization in 1999 by the IETF.

The Routing Switch
I should note that there is a third Layer 3 option. This approach, called a routing switch by Bay Networks, significantly changes the price performance of routers, but it does not otherwise change network-wide operation.

A CLOSER LOOK AT STANDARDIZED APPROACHES
Let’s look at the MPOA and MPLS networking architectures. Each has its own value to enterprise networks. In short, MPOA leverages ATM for IP applications, and MPLS enhances router operation.

Multi-Protocol Over ATM — Leveraging ATM For IP Apps
MPOA is a Layer 3 switching architecture that integrates bridging and routing with ATM networking. Router networks scale by imposing order (computable routes) through a collection of flat networks (such as emulated or physical LANs). MPOA respects this necessity, but promotes more efficient communication by allowing direct ATM SVCs (switched virtual circuits) between MPOA edge clients (such as PCs or, more typically, LAN switches and routers) that have found each other via normal hop-by-hop routing. In this way, MPOA provides cut-through switching, reducing the cumulative latency by minimizing the number of points where packet processing must be performed.

MPOA incorporates and expands on the ATM LAN Emulation (LAN-E) standards, which are now widely deployed in campus environments. The LAN-E architecture defines the use of ATM SVCs to emulate a bridged LAN. Like LAN-E, MPOA uses a client/server communication model. The clients establish SVCs to other clients on an as-needed basis to exchange data. The clients and servers are connected via ATM SVCs to exchange MPOA control and routing traffic.

Through a new protocol called Next Hop Resolution Protocol (NHRP), routing entities within a switching infrastructure can communicate with one another to determine unknown IP-to-ATM address mappings. MPOA clients have access to this information to establish a shortcut path to destination clients. In this manner, hop-by-hop processing is limited wherever appropriate, and overall performance is enhanced.

An MPOA client is a LAN-E Client which is also capable of Layer 3 forwarding one or more protocols, but typically only IP. The MPOA server is a Next Hop Resolution Server, a LAN-E Client, a routing engine, and has a little extra MPOA server code. It provides mechanisms to manage SVCs between clients, configuration services, and broadcast management. To existing routers, the MPOA server looks like a traditional router.

The protocols by which a pair of MPOA clients learn each other’s ATM address follow the normal hop-by-hop routing path, so that all routers in that path decide whether or not they wish to allow themselves to be bypassed. If they don’t, they simply return their ATM address as the response to the shortcut request. In particular, the last router before a destination client that is not MPOA capable (that is, lacks a connection to the router via an emulated LAN, or is a LAN-E-only client) will respond with the router’s ATM address.

Multi-Protocol Label Switching — Enhancing Router Operation
The primary goal of the MPLS working group is to enhance router network performance, by integrating the label swapping forwarding paradigm with network layer routing. With label swapping, labels are assigned at the edge of the MPLS network to every packet going to a particular destination MPLS router; at tandem points, the label is used to route the packet, avoiding the need to analyze the address fields of every packet. This base technology of label swapping is expected to improve the scalability, price/performance, and latency of network-layer routing. An evaluation of MPLS is best done by examining the objectives for MPLS as laid out by the IETF, since the standards work is still underway.

MPLS will be deployable as a software upgrade to traditional routers. No specific hardware features will be required that do not commonly exist on routers at the time that the standard is complete, though this does nor preclude additional optional hardware to optimize performance. Like IP itself, MPLS will provide support for multiple transport data links including LANs, PPP (point-to-point protocol) on serial links, frame relay, ATM, and PPP on SONET. Label switching uses IP-based routing protocols, including support for hierarchical networking, for multipath routing and forwarding, and for multiple levels of aggregation.

MPLS supports both topology-driven and flow-driven label assignments, where a flow is associated with an individual application session. MPLS leaves class of service (CoS) support as substantially a Layer 3 function, maintaining compatibility with the IETF’s Differentiated Services Model. This model to some extent replaces the previous work on the IETF Integrated Service Architecture, which included RSVP, the ReSource reserVation Protocol, a Layer 3 technique for CoS signaling and resource reservation. That said, labels can be used in conjunction with CoS classification at the edge of the network to minimize CoS-related processing in the core.

An advantage of MPLS is that it maintains a level of compatibility with IP-based tools, thus supporting operations, administration, and maintenance facilities at least as extensive as those supported in current IP networks. The intent is to allow current network management and diagnostic tools to continue to work in order to provide some backward compatibility. Where such tools are broken by MPLS, hooks must be supplied to allow equivalent functionality to be created.

Comparing And Contrasting MPOA And MPLS
Both MPOA and MPLS are multiprotocol, though in both cases IP operation is the initial focus. MPOA relies on ATM networking, which may be an overly restrictive requirement in many enterprise environments, where the user wants to take advantage of a combination of point-to-point, frame relay, and ATM services. MPLS is transport agnostic.

The flip side is that MPOA can take advantage of ATM CoS/QoS (quality of service) attributes, and allows ATM CoS/QoS to be delivered on a per application session basis; MPOA is said to be flow-driven, with each application session flow being mapped to an individual virtual circuit. This, however, has raised concerns that MPOA may not scale very well. ATM LANE, which is a subset of MPOA, is topology-driven and has been deployed in some very large campus networks. MPLS is also topology-driven and highly scalable, though there is discussion about adding flow-driven label switching to the specification.

If you see that neither standard does it all, you’re right. Each has its advantages. Each (I believe) will endure and win the support of multiple vendors. A major challenge, however, will be to provide interoperability among the various schemes that evolve from today’s Layer 3 switching schemes.

WHAT’S THE BOTTOM LINE ON LAYER 3 SWITCHING?
Conventional router/LAN-based architectures are being replaced by new architectures based on Layer 3 switching. It is all about IP-optimized networks and delivering high-performance networking ahead of application growth. Layer 3 networks set the stage for networks that can offer differentiated services (that is, multiple classes of services) based on user and/or application needs, enterprise policies, and the user’s willingness to pay. It allows the business needs to define the network, and not the other way around. And that’s why Layer 3 switching is a fundamentally important development.

Tony Rybczynski is director of Strategic Marketing and Technologies for Nortel's Enterprise Data Networks business unit, which delivers high-performance data networks globally. This business unit creates new alternatives to increasingly complex data network infrastructures, and offers them through direct and indirect sales channels. For more information, visit the company's Web site at www.nortelnetworks.com.   E-mail questions or comments to the author at [email protected].