Network Computing

NEWS BLOGS FORUMS NEWSLETTERS RESEARCH EVENTS DIGITAL LIBRARY CAREERS

IMMERSE YOURSELF:

Hardcore ATM Switches for the WAN

By David Willis All enterprise networks have something in common: Growth, with a capital G. Wild, organic growth that makes you wonder if you're managing the network or if it's managing you. Although new campus technologies spit out data streams at alarming rates, the old stuff still demands reliable plumbing. How can you deliver the network capacity to sustain growth, without losing stability and blowing the budget?

Carriers have learned that the best way to attack skyrocketing network costs is to reduce the number of networks they support, and they're using ATM as the mechanism to do it. Multiservice switches take on multiple personalities, exposing varied interfaces to end equipment over a single infrastructure. Like the carriers, larger enterprise managers can consolidate legacy networks into a common core, while providing new service features that are going to be required at the campus, or even desktop, level in the near future.

It's important to understand where these switches fit in the enterprise network, since they fall into a completely different category from ATM campus and LAN switching products. Instead of consolidating data traffic over a high-speed switching infrastructure where bandwidth is essentially free, they provide consolidated trunking for a variety of WAN interface and protocol types.

To view the Report card.
Thus, trunk speeds generally achieve a DS-3 (44.736 Mbps) level. Carrier services at higher rates are often too expensive, and ATM inverse multiplexing over T1 circuits is only now being standardized to provide more affordable links. It's also important to note that falling T3 circuit tariffs are making T3 more viable for the enterprise. In some areas, T3 costs have dropped to only three to four times the current cost of T1 service.

We asked the major enterprise WAN switch manufacturers to submit their multiservice switches for a hands-on test. To qualify for the test, the switches had to support ATM, frame relay and circuit emulation, and have some provision for handling voice traffic. Cascade Communications Corp. opted out, citing its carrier-only focus; Newbridge Networks did not have strong voice support (such as compression, echo cancellation and bandwidth optimization) available; Alcatel Data Networks and Cisco Systems could not muster up the equipment.

This left us with three switches that have the right mix of multiservice capabilities for the enterprise WAN, backed by vendors with the cojones to offer them up for a public performance test: General DataComm's APEX MAC, IBM Corp.'s 2220 Nways BroadBand Switch Model 500 and Norther n Telecom's Magellan Passport 160.

Unlike other performance tests, ours was not an attempt to "break" the switches by mindlessly blasting them with cells. Just as you wouldn't buy a car based on the temperature at which the engine liquifies, sending unrealistic traffic levels over switch ports doesn't tell you much about how the switch behaves in the real world. Instead, our test involved sending a mix of traffic types at reasonable levels while measuring loss, delay, circuit quality and reliability (see "How We Tested," page 58).

Of course, there is much more to these switches than just performance. When you put all your traffic over a single infrastructure, your network management system (NMS) had better be top-notch. The switches must support standard interfaces to your existing equipment, so you're not forced to invest in new edge gear. They must also supply a cost-effective migration path when your needs grow beyond current capacity. To o ffer any hope of interoperability, they must support the r apidly evolving set of ATM standards now being ratified by the ITU Telecommunication Standardization Sector (ITU-T) and the ATM Forum.