Building hybrid backbones

George Yeager had a critical decision to make: He wanted to help his company save money on a major WAN upgrade without jeopardizing mission-critical applications.

The main data center location in Columbus, Ohio, for the Columbia Energy Group required seven T-1 ports on AT&T's frame relay network just to handle the data traffic arriving from about 140 sites nationwide.

Yeager, manager of architecture and design for Columbia Energy's Enterprise Multimedia Communications unit, knew that was an inefficient setup because some access lines could lie fallow while others were all used up. He also knew that four of the company's other locations across the country were generating enough traffic to justify higher speed connections.

Yeager had the following choices to upgrade the network:

Add more T-1 frame relay ports at the high-volume sites and pay the extra access costs.

Pack more circuits from remote sites onto each T-1 data center frame relay port and hope the network wouldn't get congested.

Migrate to ATM to get T-3 connectivity and convert all the branches to T-1 or slower speeds.

Yeager's decision: None of the above. Instead, he chose frame relay-to-ATM interworking, a type of hybrid network that requires no change in configuration at branch offices while upgrading data centers.

"We thought an all-ATM network was premature and too expensive," Yeager says.

Columbia Energy, a large natural-gas pipeline and distribution company, will now have T-3 ATM connections at the heavy-duty locations while the branch offices retain their frame relay links. "The conversion is entirely accomplished in the AT&T network," Yeager says.

Yeager is following an increasingly popular path in which users are choosing to migrate only the locations that need broadband access to ATM in a two-tier WAN design (see graphic).

But before you consider implementing frame-to-ATM interworking, make sure you know exactly what the carrier is offering. There are three different flavors of interworking standards:

The first, Frame Relay-to-ATM Network Interworking, is a Frame Relay Forum standard that's nice to know about but doesn't really produce extra user benefits. The standard allows frame relay switches to chop variable-length frame relay packets into fixed-length ATM cells and put those cells back together into frames. Also known as Implementation Agreement FRF.5, the technology is usually installed on carrier switches. Deploying FRF.5 lets the carrier sell an all-frame relay network even though the carrier cloud is really ATM.

The second flavor, Frame-based User-to-Network Interface (FUNI), is an ATM Forum standard. FUNI defines a datagram that looks a lot like a frame relay packet but with certain variations in its header. The changes enable the frame to match the way ATM cells perform functions such as multiprotocol encapsulation and class of service (CoS). FUNI interfaces are designed to be installed at branch offices while the data center upgrades to a private ATM switch or T-3 access into the carrier's ATM net.

The third standard, Frame Relay-to-ATM Service Interworking, was ratified by the Frame Relay Forum in 1995 and is also known as Implementation Agreement FRF.8. In addition to chopping frames into cells, FRF.8 maps a company's frame relay permanent virtual circuits to its ATM PVCs. That way, users can install standard frame relay links at some sites and standard ATM links at others.

Of the three standards, many frame relay users already enjoy network interworking because their carriers use it on their backbones. But the technology does not provide additional bandwidth to any of the customer locations. That leaves users seeking additional WAN horsepower to choose between FUNI and service interworking.

FUNI gives users the benefits of ATM at relatively low-traffic sites without forcing them to accept ATM's big disadvantage - the fact that 5 bytes out of every 53 are lost to overhead. But FUNI interfaces and frame relay interfaces aren't exactly the same, so a lot of customer premises equipment software has to be swapped out.

Service interworking, by contrast, requires no changes at all at the frame relay sites. Yet it can have a positive impact on overall network performance. Consider a situation in which an end user at a frame relay site is sending traffic to an ATM-based data center, explains Tom Siracusa, general manager of AT&T's Data Network Design & Performance Analysis Group. When the traffic hits a switch on the AT&T network it sends the traffic along. The switch doesn't wait until it has received the entire frame relay datagram, which might run 1,500 bytes.

But the failure of FUNI to take off leaves some gaps in functionality. Unlike FUNI, service interworking does not support many of ATM's CoSes back to the frame relay site.

Siracusa concedes that the only ATM service class that works perfectly under AT&T service interworking is ATM variable bit rate nonreal time. Yet many users say that's acceptable because their initial reason for installing service interworking was to cost-effectively boost network speeds and feeds.

Columbia Energy concentrated on boosting its data capacity rather than worrying about CoSes. Largely because its frame relay/ATM network carries mission-critical traffic that regulates the flow of gas through the company's pipelines, Yeager chose not to rely on the technique of oversubscription. That's the practice of assigning, say, twice as much frame relay bandwidth coming off remote sites to a single data center port, which would result in an oversubscription ratio of 200%.

Because Columbia Energy is only willing to run a maximum 125% oversubscription, the T-3 ATM links in the center of the network were key to obtaining the needed bandwidth cost-effectively, Yeager says. Even better, the new T-3 access link can also replace a mishmash of additional T-1 and subrate dedicated access lines for Columbia's call center and outbound voice traffic.