Big bandwidth is simple and cheap. Managing bandwidth using quality-of-service (QoS) techniques - while a more elegant approach to network design - suffers from daunting complexity. Right or wrong, that’s the general perception among network managers, and it’s the perception that is influencing most buying decisions.

The argument for QoS has always been that certain types of traffic, such as voice and video, need to get through a network in a timely manner. So those real-time packets should be given a higher priority than the rest of the data, to make sure they have access to the available bandwidth.

The argument against QoS has been that if gobs of bandwidth are available, then prioritization isn’t necessary. All packets will get the bandwidth they need.

The LAN side

The latter argument makes the most sense in the LAN, as the prevalence of Gigabit Ethernet at rapidly falling prices is giving desktops and servers more bandwidth than they could possibly use.

It’s difficult to find network equipment that doesn’t provide excessive bandwidth, says Stephen Hope, network consultant at Energis Integration Services, in Carrington, England. "Even entry-level equipment is wire speed."

"Most of our customers are going ‘the bigger the better,’ " says Brian Stengel, executive vice president of IntelliNet, a network management service provider in Cleveland. "Most of them are finding [big bandwidth] much easier to sell to management, plan with the applications teams, design and implement with the network teams, and operate with their management service providers or internal operations teams."

The WAN side

The argument for QoS continues to make more sense in WANs, where bandwidth is scarcer and more expensive than in the LAN. But even this is changing.

Buying more WAN bandwidth rather than a QoS device served the purposes of Gliatech, says Dave Bujaucius, the Cleveland biotech firm’s senior IT manager. "We found that the difference between a fractional and a full T-1 was less than $100 per month," he says. "A QoS product would have taken several years to pay for itself, so we just purchased the extra bandwidth."

Of course, that’s nothing compared with what may become available in the WAN at low cost in the next few years. The bandwidth available for the WAN is growing extremely fast, thanks to the "optical network revolution," says Dave Passmore, research director at The Burton Group.

SONET speeds over a single channel of fiber are quadrupling every two years, while the number of channels vendors can squeeze onto one fiber-optic line is doubling in that same time.

That means prices are bound to fall rapidly, as bandwidth gets less expensive and carriers start to undercut each others’ prices. "It’s basically an optical arms race," Passmore says.

This will likely happen everywhere except remote areas, which won’t get access to fiber as quickly as metropolitan areas.

A matter of quality

But even if plenty of bandwidth is available, chances are that file transfers will at times encroach on voice and video traffic if QoS isn’t in place. That’s because TCP is designed to use all the bandwidth available to get the job done, says Mike Bass, principal network design engineer for a large interexchange carrier.

"No matter how much bandwidth you have, you will hit saturation peaks several times a day that last maybe only a few seconds each," he says. If voice calls are running at the same time, callers will likely hear skips or silence for those seconds.

Many users are choosing to take those chances. After all, cell phone users already put up with such poor quality.

But some network managers aren’t considering QoS simply because they aren’t ready to trust data networks with voice calls. Energis’ Hope says a customer once posed the question, "Why would I want voice with the reliability of data?"

"This kind of - probably justified - paranoia is common and will persist for a long time," Hope says.

Making it work

For organizations insisting on putting voice or video and data on one network, the most prudent course of action might be to combine the two approaches. Instead of creating eight QoS levels for traffic of varying importance, create just two: one for voice and video, and the other for everything else. At the same time, continue to overprovision bandwidth somewhat. This approach adds only a little complexity, while making sure that skips and gaps don’t creep into phone calls.

Visalia, Calif., is taking this approach. The city is linking its sites in a giant LAN, says Michael Allen, the city’s IS manager. Voice will run over the LAN, and in the future, so will video-based training.

"The plans are to implement QoS for the voice side and not differentiate the data side with QoS," Allen says. "If it were data only in this high-bandwidth environment, we wouldn’t bother with QoS."