Printer Friendly

What's all the noise about switching layers?

Not too many years ago, a lot of vendors were getting a lot of hype over a great new technology: Layer 2 switching for IP. Ipsilon, since acquired by Nokia, called it IP Switching; everyone else had their own flavors and versions. Cisco rolled out tag switching, 3Com had FastIP, Ascend offered IP Navigator, and so on and so on. A flurry of Layer 2 activity seemed to peter out a year or two ago, and now no one seems to want to talk about Layer 2 switching anymore. What happened?

Whither Layer 2 switching? Mostly, it's still around, but in different forms. Vendors used ASICs (application specific integrated circuits) to make their Layer 2 switches fast: instead of building router software, the switching smarts are burned into silicon. And a funny thing happened when vendors realized they could use ASICs for their routers: performance skyrocketed, and relatively dumb Layer 2 switches didn't seem like such a bargain anymore. In very simple terms, what vendors are now calling Layer 3 switches are, at least functionally, nothing more than routers that use ASICs instead of software to figure out routes.


The layers we're talking about are OSI reference model layers. Layer 2 represents the data link layer: this means Ethernet or ATM or whatever network medium transport protocol is being used. Switching at this layer means looking at a field in the ATM cell header or a shim between the Ethernet and IP headers for a hardware network interface address of the next hop node. It's fast, because the switch doesn't need to do much processing of the frame, just look at the destination field in the header--or the attached label on a frame of switched IP. The switch looks at the label, looks it up in a table, and then switches the frame. There is little processing to be done, and what processing the switch does happens on silicon, on an ASIC.

Switches can move frames at wire speed, don't have to do any time-consuming processing of packets in software like routers, and are also considerably less expensive than routers--as much as one-tenth the cost per port. So, what's not to like? For one thing, switches may be fast and cheap, but they aren't terribly smart about routing. Which means they don't always do the right thing with IP broadcast and multicast. Instead of realizing that broadcasts shouldn't be propagated across subnets, switches forward these packets across all connected subnets, clogging them with irrelevant traffic.

Switches also lack the smarts to understand what to do with multicast packets. Switches are particularly attractive for high-volume ATM networks. Because ATM uses virtual circuits (VCs) and small network transport units (53 bytes), ATM cells don't have much room for IP headers, let alone IP packet payloads. If every ATM forwarding device reconstructed IP packets from ATM cells, determined a route, built a new VC for the next hop, and then re-fragmented the packet for forwarding, the devices would be hard pressed to keep up with the traffic.

But routing occurs at OSI Layer 3--the network layer. This is where IP operates, independent of any raw network protocols, and where IP routers have to process Layer 2 frames up the protocol stack. Only then can the router start to work on processing the IP headers.

Though you can't do switching without any routing at all, you can switch packets once they are inside an autonomous system if you do the routing once as the packet enters. (An autonomous system is an aggregation of networks that behave as a single network to external observers--it may be a high-speed backbone or may be an organizational internetwork.) The edge switch/router accepts inbound packets, figures out how to route them, figures out how to encode the route into labels that can be added to the network frames, and then switches the frames into the autonomous system. Switches inside then switch the frames until they arrive at their destination.

So, what's the difference between a Layer 3 switch and a Layer 3 router? Not only do the Layer 3 IP/ATM switches use ASICs, but they also do the smart thing with ATM VCs. And what about this Layer 4 switching that vendors talk about? Layer 4 is the transport layer--where TCP and UDP do their things. So-called Layer 4 switches actually examine, not just the Layer 2 network address and the Layer 3 IP address, but they may also look at the TCP or UDP port associated with each packet so that different streams of packets can be treated differently. Each TCP virtual circuit is uniquely identified by the IP addresses and TCP ports of the endpoints.


Layer 2 IP switching technology is still waiting on ratification of open standards. Which means that, if you decide on 3Com's FastIP solution, you must migrate a lot of your networking infrastructure to support FastIP, and you may have trouble getting it all to interoperate. IP routing may be slow, but at least it is interoperable; and any router should forward packets from a host without problem using any kind of network interface card to any other router. The IETF is working on protocols relevant to Layer 2 switching, and it is moving closer to publishing RFCs for Multiprotocol Label Switching (MPLS), Multiprotocol over ATM (MPOA), the Next Hop Resolution Protocol (NHRP), and related protocols. Once vendors have some standards to which they can design their products, interoperability should follow.

MPLS, derived in part from Cisco's early work on tag switching, as well as from work by IBM and Hitachi, uses labels (very much like Cisco's tags) to stand in for actual routes. When an IP packet enters an autonomous system through an edge router/switch, that device figures out the packet's proper route, as well as the appropriate next hop for the packet, and assigns it a label--and then forwards the packet to the next label switching device. Rather than looking at Layer 3 routing information (the destination IP address), the label switch just reads the label of the cell, checks its switching table, and determines the appropriate next label switch, as well as a new label for the cell. When the packet hits another edge device, the label gets stripped off and discarded; and the edge router just forwards the packet normally.

The label switching approach focuses on the routing and switching infrastructure, allowing network managers to build up switched networks where traffic requires them and leaving the rest of the network as is. Another approach is exemplified by the Multiprotocol over ATM (MPOA) specification. Championed by the ATM Forum, MPOA uses NHRP to act as a network application used by nodes to negotiate an MPOA switched route. When a node starts communicating with a remote node, it may need or want a fast, switched route, perhaps to guarantee some degree of quality of service (QoS). In that case, the node sends an NHRP request to the remote node asking it if there is a Switched route available; if the remote node agrees, then they can continue to communicate bypassing unnecessary routers. The network solutions based on this approach tend to require the addition of special network interfaces as well as special software.

However, Cabletron senior architect Trent Waterhouse says that "some people are saying MPOA will be dead on arrival." The reason, according to Waterhouse, is that network vendors figured out how to apply to new Layer 3 switches the lessons of ASICs in Layer 2 switches. By implementing routing routines into silicon, vendors can approach switched Wire speeds for moving packets--without losing any of the benefits of routing.


Cisco aggressively supports MPLS, contributing to the IETF workgroups and submitting numerous Internet drafts outlining how MPLS should work and how it can be used. Surprisingly, however, Ranjeet Sudan, Cisco's product manager for tag switching and MPLS, says, "We don't stress the performance increase in tag switching any more." Sudan explained that, although you can get similar performance improvements from Layer 3 switch/routers, tag switching turns out to offer unique benefits for doing "traffic engineering." For example, network engineers can use tag switching to do load balancing by defining two or more equivalent routes across a backbone.

And Cisco is not the only vendor backing away from exorbitant claims for pure Layer 2 switching performance improvements. 3Com product manager Glen Gibson says "FastIP is not being positioned as the best solution for new installations, areas which are now being addressed by Layer 3 switches, which offer higher performance and more features."

Loshin is a freelance writer living in Mass.
COPYRIGHT 1999 Nelson Publishing
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1999 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:Technology Information; OSI reference model layers
Author:Loshin, Pete
Publication:Communications News
Geographic Code:1USA
Date:Jan 1, 1999
Previous Article:New edge routers outshine all-purpose gear.
Next Article:Latest routing and switching products.

Related Articles
Global view favors packet move.
Tips for testing on X.400 networks.
Go for automatic.
LAN switches gear up for Web explosion.
GPRS challenges mobile network operators.
Cicada Semiconductor enables low cost Gigabit Ethernet networking.
How to secure switches and routers: security-in-depth philosophy marries traditional network security technologies with implementations. (Special...
Pressure control comes from a new source.
Atmel licenses industry-leading Gigabit Ethernet Physical Layer Core.

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters