Related Links Networkers session: "Designing High-Performance Intranets with Multilayer Switching" White Paper: "Enterprise Campus Solutions"	Packet™ Magazine Archives, Fourth Quarter 1998 Multilayer Switches New Breed of Network Devices Supplies Needed Performance Boost to Campus Backbones As graphics-intensive applications devour bandwidth, and more traffic crosses the campus backbone to tap into server farms, users are bringing Layer 3 switching capabilities into the aggregation layers of their intranets to maintain application performance. Companies that have been ramping up their use of graphics and multimedia applications might suddenly find their campus backbones straining to maintain network performance. To support the bandwidth requirements of such applications, high-performance devices called Layer 3 switches are making their way into campus networks. Their incorporation is becoming a particular necessity in organizations running centralized server farms for security and manageability advantages. "The server farm approach is causing an increasing volume of traffic to venture off of LANs and cross the backbone as users tap into centralized or remote resources," notes Geoff Haviland, a Cisco Network Design Engineer. Shifting Traffic Patterns Such traffic patterns deviate from the long-standing tendency for 80 percent of LAN traffic to remain local and traverse the backbone just 20 percent of the time. In fact, common observations are that the ratio has flip-flopped so that 80 to 100 percent of traffic is now actually leaving the LAN. This shift has become a major driver toward giving backbone equipment and links faster switching capabilities so that network managers can continue to ensure application performance. An optimal way to handle the traffic loads is to create a Layer 3 switched "distribution layer" in the intranet that aggregates LAN traffic across the backbone at high speeds. The Layer 3 equipment designed specifically to comprise this layer combines ultra-fast switching capabilities -- implemented in hardware -- with intelligent software-based routing functions. This specialized equipment, which includes Cisco's Catalyst® 8500 (announced in April 1998) and its Catalyst 5000 and 5500 models, leverages the wire-speed packet switching achieved in hardware and then adds the intelligence of Layer 3 routing protocols. The widely installed Cisco 7500 core router can continue to perform distribution-layer functions in environments where protocols such as AppleTalk, which are not supported in the specialized equipment, are in use. Like the Catalyst 8500, the Cisco 7500 performs Layer 3 packet forwarding as well as routing of protocols, but trades off some switching performance to be able to support more protocols and a variety of WAN interfaces. Routing protocols in Cisco IOS® software allow Layer 3 devices to exchange important information with one another about the status of the network so that they can determine the optimal paths that packets should take to reach their destinations. Some common routing protocols in use are the industry-standard Open Shortest Path First (OSPF) and Cisco's widely implemented Enhanced Interior Gateway Routing Protocol (Enhanced IGRP).
A modular, hierarchical campus design that incorporates multilayer switching eases scalability and network management while enabling a redundant architecture.
	Hierarchical Design Boosts Scalability, Manageability A common scenario for optimally arranging one of these high-performance, switched backbones is to use a structured, hierarchical design model. "Such a design boosts the scalability and manageability of networks," says Haviland. In addition, users should attempt to switch traffic wherever possible. "Today's multilayer devices take a switch-centric view of the network," notes Fred McClimans, Chief Executive Officer of Current Analysis, Inc., a network consulting firm based in Sterling, Virginia. "They enable switching wherever possible for optimum performance and leverage smart Layer 3 features as needed. This keeps network topologies simple, which is the key to maintaining application performance," he says. Haviland, in fact, advocates a hierarchical design that includes several layers of switches. He advises installing an "access layer" of Layer 2-only devices to connect LAN nodes to one another. These are devices such as Ethernet switches that dedicate 10 or 100 Mbps of bandwidth to each network node. The Layer 2 switches can then be aggregated in wiring closets at the "distribution layer" through high-performance multilayer switches. Depending on the port density and throughput needed, as well as on the network-layer protocol support required, the distribution layer can be formed from equipment such as the Catalyst 5000, 5500, or 8500, or the Cisco 7500. The Catalyst 8500 series, for example, has been optimized for IP, IP Multicast, and IPX environments, while a device such as the 7500 will be necessary for environments running other Layer 3 protocols and for providing WAN connections. The Catalyst 8510 campus switch router (CSR) can connect up to 32 Fast Ethernet pipes and forward up to 6 million packets per second (pps). The Catalyst 8510 also supports Fast EtherChannel® and Gigabit Ethernet for even higher-capacity trunking. The Catalyst 8540 CSR offers still higher throughput and higher interface density. Users can connect up to 128 Fast Ethernet or up to 16 Gigabit Ethernet trunks. The Catalyst 8540 offers nonblocking Layer 3 switching at up to 24 million pps. Users can determine which model to run at the distribution layer by calculating the number of links that must be connected and the anticipated traffic load coming in. Often, says Haviland, users employ a 20:1 oversubscription ratio to determine the required speeds of links to the wiring closet. "This is a good, conservative rule of thumb," he observes. Haviland has seen bandwidth oversubscriptions of an order of magnitude higher than the 20:1 ratio "with no problems at all," and suggests that network managers monitor traffic loads to determine the proper oversubscription ratio. Catalyst 5000, 5500, and 8500 switches can also form the "core layer" that front-ends the remote server farm with very high-speed links, such as Fast Ethernet (100 Mbps) or Fast EtherChannel technology, which can double or quadruple that speed depending on implementation. If the Catalyst 8500 forms the core, Gigabit Ethernet or 155-Mbps ATM connections are also optional server connections. Modularity Eases Growth Using a modular approach that combines Layer 3 switching in the distribution and core layers with Layer 2 in the wiring closet allows the basic module to be repeatedly reproduced to scale the campus. The fundamental module "building blocks" also deliver redundancy and deterministic recovery from failures. Wiring-closet switches in each building block can be connected to two separate switches at the next layer in the hierarchy for redundancy. Users should then take advantage of Layer 2 features in the wiring closet, says Haviland. For example, Per-VLAN Spanning Tree (PVST) allows load balancing on both uplinks from each Layer 2-only switch to the wiring closet without spanning tree blocking. Another helpful feature is Uplink Fast, which enables the Layer 2 devices to recover from a link failure in less than three seconds. Cisco IOS Features Complement Raw Performance In addition to running high-speed links and performing fast packet-forwarding, using the hierarchical model subnets the backbone to contain broadcasts, provide security, and enable proxy services as required. Networks with Layer 3 switches can also take advantage of Cisco IOS software features to help them scale. For example, each module can take advantage of Hot Standby Router Protocol (HSRP), a Cisco IOS feature, for fast recovery from failures. Interior routing protocols such as OSPF and Enhanced IGRP continue to provide sophisticated load balancing and fast, deterministic recovery from failures across the campus backbone when necessary. Cisco IOS management and troubleshooting features such as Network Time Protocol (NTP) and protocol debugging also add value to the basic switch performance. Combining these functions with Layer 3 switching -- which uses a small amount of Layer 3 network information when performing packet switching, but charges no performance penalty for it -- attempts to offer a "best-of-both-worlds" approach to backbone networking. "Using Layer 3 switching in the campus is the fundamental tool for achieving scalability and manageability. The intelligence is in the Cisco IOS software," Haviland explains. ATM Alternative Handles Voice Traffic ATM LAN Emulation (LANE) can be used to build a scalable campus backbone. LANE is an ATM-standard scheme that allows more traditional LANs such as Ethernet and Token Ring to communicate over an ATM network. Using the same basic multilayer module design, LANE can simply replace Ethernet links among campus sites for connecting the modules together. ATM backbones also are useful for environments that trunk real-time voice and video traffic along with data traffic.

Geoff Haviland, a Cisco Systems Network Design Engineer, created the session, "Designing High-Performance Intranets with Multilayer Switching." Haviland presented this session at both 1998 US Networkers events. To contact him, e-mail haviland@cisco.com.

Top of Page Table of Contents Posted: Thu May 13 10:46:28 PDT 1999 Copyright © 1998 Cisco Systems, Inc.