Switching's new breed.
Many organizations continue to rely upon traditional multiprotocol routers to provide the foundation for their networking infrastructure. The combined effects of a growing number of connected end users, the acceptance of IP as the protocol of choice, and the introduction of the Web and Web-based applications results in a continuation of this reliance. Placed in the center of the network, the router--or more specifically, the backplane of the router--is the collapse point for the entire enterprise. Total network response time suffers because the overburdened router manages all the WAN, as well as LAN, connectivity. How can users overcome this problem without adding new complexities to the existing network?
Multilayer switching, a practical evolution of today's LAN switching and routing technologies, solves this dilemma. It removes the scalability and throughput restrictions that limit network growth, while building the foundation for an emerging generation of gigabit-networked applications.
CHANGE IS GOOD
Four related forces are pressing current network designs to the limit:
* Widespread adoption of LAN switching;
* Increasing centralization of network servers;
* The proliferation of intra nets and IP-based Internet communications;
* Rapid adoption of high-band width, low-cost LAN technologies such as Fast and Gigabit Ethernet.
Over the last couple of years, LAN switching has greatly increased network performance by replacing shared media with dedicated bandwidth. Users benefit from direct access to their networks, and the bottlenecks of shared Ethernet or token ring disappear as point-to-point switching is deployed.
But, as applications arrive to take advantage of switching's improved throughput, bottlenecks emerge at a higher network level. New bottlenecks stem from switching's roots as a Layer 2 bridging technology--switched networks are flat domains that must be subnetted just like bridged networks to alleviate broadcast overhead. Without the subnetting performed by Layer 3 routing, LAN and switching infrastructures do not scale. Large flat, switched networks are subject to the broadcast storms, Spanning Tree loops, and inefficient addressing limitations that brought routers into bridged networks more than a decade ago.
Routing is as important to switched networks as it ever was, and therein lies a predicament: High-performance LAN switches are pumping millions of packets per second (pps) across campus backbones, served by traditional routers that can, at best, handle half a million packets.
Wherever Layer 3 functions are invoked in the switched campus backbone can cause a major bottleneck. Traditional routers remain expensive to buy and manage, so purchasing more router interfaces to handle the overload is not a cost-effective option.
At the same time, switching allows networks to be designed with greater centralization of servers and other resources, helping to streamline network administration and increase overall security. This recentralized topology means that a greater proportion of traffic has to cross the network backbone which entails more traffic being routed beyond a local sub-net. The old 80/20 rule of thumb that previously predicted that 80% of network traffic stayed within a given workgroup or sub-net is being reversed as more applications have to access resources located across the backbone.
Corporate intranets further exacerbate the problem with increased network usage and by granting easy access to resources deployed widely across the enterprise. A common estimate is that roughly one-half to two-thirds of all intranet traffic travels between sub-nets. Wide-area Internet usage has a similar effect, as every Web session has to be routed to the Internet from the user's local network by a router.
In other words, LAN switching and the applications that leverage its performance are quickly arriving at the limits of their improved capacity. Because Layer 2 scalability depends entirely upon Layer 3 routing, the throughput of traditional backbone routers is today's network bottleneck. By some accounts only 10% of today's desktops are connected via LAN switching. As this market continues its rapid growth the Layer 3 bottleneck will worsen.
A COST-EFFECTIVE, HIGH-SPEED SOLUTION
What's needed is a next generation melding of Layer 2 switching and Layer 3 routing functionality which is precisely what multilayer switching achieves. Built on a core of Gigabit Ethernet technology, this solution can switch campus traffic at wire speed, satisfying Layer 3 switching and concurrent wire-speed switching and multiprotocol routing. This combination not only solves today's throughput problems but also removes the conditions under which Layer 3 bottlenecks form. For future growth, it gives networks both gigabits of throughput and an innate capacity to process it on the fly.
TAKING THE LOAD OFF
Now let's examine two aspects of multilayer switching: router offload and switching routers.
One solution is to simply reduce the burden on installed routers by transparently offloading processing-intensive IP and IPX traffic forwarding at wire speed. This is accomplished by placing a multilayer switch in front of an existing router as a "router front end." Conceptually, this approach is much like the front-end processor (FEP) of a mainframe in which the FEP handles straightforward communications while the more expensive central processing unit (CPU) sitting behind it resolves larger matters. In terms of network topology, the router does not know the multilayer switch is there, and other devices think they are communicating directly with the router. This eliminates the need to perform time-consuming configuration procedures.
In many environments, this capability can offload the router up to 80%, freeing the device to efficiently handle topology management and non-IP and IPX traffic. The benefits of such an approach are clear: network performance is improved, there is no need for costly upgrades to the installed router, and no additional administrative burden is incurred.
A BETTER MOUSETRAP
Another multilayer switching approach is to replace the overloaded legacy backbone router with a wire-speed switching router that supports a multitude of protocols, ranging from IP, IPX, OSPF, and Appletalk, to BGP4 and VRRP.
A switching router performs standard Layer 3 routing and Layer 2 switching concurrently at wire speed. It supports Gigabit Ethernet and can switch 10/100 Mbps Ethernet pipes over a gigabit backbone. Installed routers and hosts see it as just another router, while switches view it as either a switch or a router, depending upon the traffic at hand.
This new breed of device delivers the functionality of a traditional router at the price points and performance of a Layer 2 switch. Such compelling price and performance are made possible through hardware integration and advanced silicon logic. Traditional routers use expensive and slow processors to perform Layer 3 functions. Gigabit Ethernet switching routers accelerate routing and switching functions by building these capabilities into Application Specific Integrated Circuits (ASICs), thereby creating a "router on a chip." This provides a manyfold increase in performance and lowers the costs of goods, a savings that is passed on to users in the form of lower per-port pricing.
Switching routers provide the ultimate flexibility and investment protection by allowing users to switch or route on a per-port basis. Such an architecture enables users to maximize their backbone design by deploying either controlled switching or routing wherever it is needed in the network.
HAVE IT YOUR WAY
Traditionally, users have had two choices, put in a switch to switch or a router to route.
With the new breed of multilayer switches, users can now install a single cost-effective device that improves network work performance and provides a foundation for future network growth.
When making a multilayer switching decision, network work managers should choose a product that provides multiple capabilities, including the ability to offload Layer 2 and Layer 3 traffic from traditional routers, support for multiple protocols, and the ability to switch and route on a per-port basis. This ensures maximum flexibility of network design both today and tomorrow.
[Figures 1-2 ILLUSTRATION OMITTED]
Circle 273 for more information from Foundry Networks
Demopoulos is vice president of marketing, Foundry Networks, Sunnyvale, Calif.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||multilayer switching|
|Date:||Feb 1, 1999|
|Previous Article:||Gig-E goes to the head of the class.|
|Next Article:||Cyber scholars build virtual bridges.|