|
|
|
|
Are Layer 3 switches all they're cracked up to be? And do they really constitute a threat to the established world order?
In internetworking, nothing has generated more buzz than Layer 3 switching. Touted as the killer of routers and Cisco Systems' profit margins, Layer 3 switching has probably done more to generate discussion and debate than relieve enterprise networks of congestion. But let's cut through the hype and get to the point: Layer 3 switching is a marketing term for low-cost, wire-speed routers. Period. End of discussion. But not end of debate. Layer 3 switches do not kill off routers because they are routers. They do threaten Cisco's fat router margins because this new breed of hardware-based routers is as little as one-tenth the cost of a Cisco 7500. But Cisco seems to have a remedy for that. Cisco markets its new Catalyst 8500 as a Layer 3 switch even though the box is priced more like a router. Layer 3 switches usually cost around $650 per 100M bit/sec port. But the low-end Catalyst 8510 costs up to $2,000 per 10/100M bit/sec port. Cisco rationalizes this pricing by claiming its 8500 is more of a "wire-speed switch router" than a Layer 3 switch. Come again? "We're not trying to fight the terminology; we're going with that flow," says Jayshree Ullal, vice president of enterprise marketing at Cisco. "But we don't believe the 8500 is the same type of Layer 3 switch as a Bay Networks Accelar or a 3Com [CoreBuilder] 3500." Ullal may have a point. The Cisco 8500 supports multiple protocols and several Cisco IOS software features typically found in the company's 7500 router, which costs about $5,000 per 100M bit/sec port. Layer 3 switches usually support only IP, or IP and IPX, and do not include most of the software feature sets included in high-end routers. Still, Cisco's competitors are anxious for a bake-off. "I'm not sure if Layer 3 will kill Cisco. However, it is very clear that they are late to market with a far inferior product offering," says Frank Hayes, director of LAN switching programs at Cabletron. "Although reading some of their material leads you to believe their product is gold, they have been very, very reluctant to have a head-to-head comparison of their 8500 technology vs. the competition." And reluctant to come down to Layer 3 switch price points. Is the 8500, at $2,000 per 10/100M port, Cisco's way of protecting its handsome router margins in the age of low-cost, high-speed routing? "That's a really good question," says Scott Heritage, an analyst at investment firm Warburg Dillon Read in New York. "Layer 3 switching is not going to be a positive impact on Cisco's business model, but I don't think it's going to be severe enough to hurt the company."
Cisco's pain is minimal"I believe customers are willing to pay a premium over existing Layer 3 switches in the market because if they're going to deploy this in the backbone, they want the functionality, and they want the robustness of knowing that their production backbone will work," Ullal says. "Cisco can certainly build and offer customers an IP-only, highly optimized, no frills, no functionality Layer 3 IP switch at price points that our competition is offering, but we don't believe the market has matured enough to get there. We believe that people are still fundamentally working with collapsed routed backbones and are looking at how to retain that functionality and increase the performance," she says. Sales of Cisco's 7500 router have not fallen off since the emergence of Layer 3 switching, but the company's Catalyst 8510 only started shipping in June. Nonetheless, Ullal says switching is growing between 30% and 50% year after year, while router growth is between 10% and 15%. Shipments of the 8510 to date have met Cisco's expectations, Ullal says. The higher end Catalyst 8540 is now shipping, Ullal says. Though Layer 3 switching appears to be taking business away from traditional routers in the enterprise, Cisco's greater emphasis on sales to service providers is intended to buffer some of that impact, Heritage says. "That's an opportunity to make up for some of the negative things that are happening in the LAN market," he says. Cisco concurs. "The WAN growth has offset the LAN shift to switching," Ullal says. That shift has users pushing their collapsed backbone routers to the WAN edge of the network and front ending them with Layer 2 and Layer 3 switches. As a result, fewer $5,000 100M bit/sec router ports are being sold to segment and subnet LANs and to serve as collapsed backbone devices. "In the enterprise, slowly but surely, the routers are going to be more peripheral, more aimed at the wide area, more aimed at making sure the legacy protocols are managed and routed with whatever infrastructure is put in place," says Craig Johnson, an analyst at The PITA Group in Portland, Ore. This phenomena has resulted in just single-digit growth for routers over the past few years, Johnson and Heritage say, disputing Ullal's figures. Johnson predicts flat growth for routers by year-end. Yet the impact on Cisco will be due to whatever Cisco does, not the market, according to Johnson. "Will it hurt Cisco revenue-wise? I don't think so," he says. "Cisco understands very well the transitions that are taking place. It's just a question of how Cisco manages its own transition." Users are just beginning to manage their own transitions from traditional routers to Layer 3 switches, analysts agree. But they disagree on whether Layer 3 switching is ready for prime-time play. "There are some networks where Layer 3 switching is definitely anchoring the network, but it's still so early in the life cycle," Heritage says. "If you look at the sales wins, they've all been onesy-twosy-threesy deals," Johnson says. "To me that's people kicking the tires, solving immediate bandwidth problems." Both analysts agree that the use of Layer 3 switching will take off when some known quantity vendors - including Cisco, 3Com and Xylan - begin shipping high-end products this fall. But whether Layer 3 switching is ready for enterprise use depends on the enterprise. "A Layer 3 switch represents a subset of a full-fledged router," Ullal says. "A router ends up doing a lot of LAN aggregation, as well as WAN aggregation. Most of the Layer 3 switches fundamentally focus on the LAN." Some enterprises also still depend on different types of routing features found only in traditional routers, such as DHCP Proxy, debug commands and access control lists, Ullal says. And many Layer 3 switches sacrifice data loss for performance. Ullal says some Layer 3 switches drop 50% of their packets. "I don't believe the first generation of Layer 3 switching has been ready for prime-time enterprise use," she says. "They've mostly been toys that people played with in the lab." Nonetheless, Cisco is positioning the Catalyst 8500 as a LAN aggregation switch in enterprise backbones, assuming half the duties of a 7500. The 7500 will now function as a WAN aggregation device, Ullal says. Cisco plans to add WAN interfaces to the 8500, but it will not support as many as the 7500, Ullal says. As for Layer 3 switches being toys, don't tell that to Fujitsu Network Communications (FNC) in Richardson, Texas. The maker of telecommunications equipment believes Layer 3 switching is ready for the enterprise and that its enterprise is ready for Layer 3 switching. FNC is installing 3Com's CoreBuilder 3500 Layer 3 switches to avoid the latency that's inherent in traditional routing. Previously, the company connected Bay Networks' System 5000 hubs in each building wiring closet to a Bay Backbone Concentrator Node (BCN) router in the network core. The BCN slowed transmissions by 1 to 5 msec, Fujitsu officials say. The Layer 3 switches forward packets at wire speed, which officials say eliminates latency. Another reason FNC chose Layer 3 switching was to control traffic flow by distributing the routing function around the network. Layer 3 forwarding at the edges of FNC's campus network helps contain broadcast traffic. Previously, all traffic had to go through the BCN router in the core of the network. But the CoreBuilder 3500s in the wiring closets learn the routes from Address Resolution Protocol (ARP) tables, instead of performing IP lookups on each packet. ARP "snooping" can also be done at wire speed, which makes it practical to employ Layer 3 routing at the edge, Fujitsu officials say. Another factor for Fujitsu selecting Layer 3 switches was capacity. Even with up to 5 msec of latency, Fujitsu was satisfied with the BCN's performance. But the company had filled up every BCN port. Lastly, FNC chose Layer 3 switching to experiment with the latest in routing and switching technology, company officials say. But the latest and greatest in technology does not fit every enterprise network. Financial services giant and FNC neighbor USAA in San Antonio, Texas, looked at Layer 3 switching once and decided against it. USAA believes its huge network - 30,000 users in 200 virtual LANs - is too complex for Layer 3 switching. The network is anchored by 78 Cisco 7513 routers and divided into 17 "electronic communities," each composed of 2,000 to 2,500 users and 12 VLANs. Electronic communities are largely segregated from each other to isolate faults and broadcast storms and to allow for unique naming schemes. "It's very much a routed network," says Michael Sjolander, manager of network LAN services at USAA. Each VLAN within an electronic community is a separate subnet, and the 7513s do a lot of routing of IP and IBM Advanced Peer-to-Peer Networking (APPN) traffic within electronic communities and across the WAN to regional offices. The VLANs are configured with Cisco Catalyst 5000 Fast Ethernet departmental and wiring closet switches and LightStream 1010 ATM backbone switches. Even though the backbone is switched, it is a Layer 2 switched backbone; all Layer 3 duties are handled by the Cisco routers, Sjolander says. USAA tried building a switched Layer 3 network at one time but ran into implementation and application problems. Also, a switched Layer 3 network would not be able to handle the considerable amount of APPN traffic in the USAA net. APPN isn't the only protocol Layer 3 switches can't handle. Most wire-speed Layer 3 switches are IP-only; some of those that support other protocols, such as IPX and AppleTalk, forward those packets at the same rate as traditional routers, or slower. Indeed, multi-protocol routing and WAN access are two reasons some users still keep older, slower routers around. "If you weren't running AppleTalk or IPX, you wouldn't need routers," says James Wiedel, director of networking at the University of Southern California in Los Angeles. Community First Bancshares in Fargo, N.D., formerly used Novell servers to route IPX. The company then flattened out the network with Layer 2 switches and implemented Layer 3 switching to subnet LAN segments. The company installed Bay's Accelar Layer 3 switches. Community First Bancshares still has a router in place to route between its internal network and wide-area segment, where the company has its server farm, says Randy James, vice president of development. That router is also handling the IPX traffic. "I'm not aware of any Layer 3 switches right now that do IPX routing," James says. Cisco claims its Catalyst 8500 supports IPX, and Cabletron claims its SmartSwitch Router does also. But protocol support is only one factor to consider when sizing up Layer 3 switches. Another is how deep into the network Layer 3 capabilities should go. Usually, Layer 3 switching is isolated in the core of the network, at the backbone switch. But some vendors, such as Bay (now a unit of Nortel), propose distributing Layer 3 functionality out to the edge of the LAN in wiring closets. Bay claims Layer 3 in the wiring closet will deliver quality-of-service capabilities, such as application and bandwidth priority, and policy, closer to desktop systems. Yet analysts question whether users will pay the price for Layer 3 functionality - and complexity - at the edge of the network where switches are low-cost and featureless. "There's a lot of reluctance out there to extend OSPF [Open Shortest Path First] across the entire network," says Esmeralda Silva of International Data Corp. in Framingham, Mass. "It's not a straightforward thing. I think users want to keep as much of the complexity in the backbone as possible." Another factor to consider is ATM. If users are implementing a pure, native ATM network, Layer 3 switching is irrelevant because ATM's Private Network-to-Network Interface (PNNI) routing protocol operates at Layer 2. But native ATM enterprise networks are rare. Users looking for the performance boost from ATM's virtual circuit switching yet seeking to retain the familiarity of IP and IP routing protocols will most likely run IP over ATM. The question then becomes whether to use PNNI as the routing protocol - which FORE Systems naturally endorses - or Routing Information Protocol or OSPF. Users looking to run multiple protocols over ATM can use the ATM Forum's Multiprotocol over ATM (MPOA) specification, which some vendors are now shipping. MPOA essentially provides Layer 3 switching for ATM networks. But again, cost, complexity and network penetration of MPOA are issues users must consider when evaluating Layer 3 switching with ATM. ATM hardware generally costs more than Ethernet, Fast Ethernet and Gigabit Ethernet gear, and MPOA adds cost and complexity on top of that. These factors aside, no matter how you slice Layer 3 switching, it all ends up being the same thing: routing. And if a product performs routing, it's a router. Perhaps reformed router basher Cabletron sums it up best:
"Routing is alive," says Cabletron's Hayes, "much like TV was alive as it transitioned from black-and-white to color."
|
 |
Forum: Cutting through the hype
Contact Senior Editor Jim Duffy
What exactly is a switch?
The many uses of the word "switching"
Switching grows up
Can Layer 3 switching increase bandwidth?
Cisco preps Catalyst 8500 switches
Router and Layer 3 switching
Fujitsu looks to Layer 3 switches for speed, backup
Cabletron smartens up switches
Layer 4 switching could relieve saturated servers
Network World Fusion Focus on High Speed LANs
Where are they now?
Market size:
"While routing switches as a class of product have only been available for
nine months, they have already been installed in a wide range of
applications. And while some customers are only scratching the surface of
the products' capabilities, others are running full OSPF routing, filtering
and prioritization successfully in their networks today. The customer base
has spoken . . . adoption of these routing switches has already been beyond
expectations."
- Basil Alwan, Bay Networks
| 
Copyright, 1995-2001 Network World, Inc. All rights reserved. |
By Jim Duffy