Wireless networking: 802.11g speeds ahead of IT needs

The evolution of wireless Lan standards can confuse even the most obsessive networks manager.

The evolution of wireless LAN standards can confuse even the most obsessive networks manager.

PC cards and access point hubs based on the 11Mbit/sec. 802.11b WLAN standard were just moving into the mainstream last year when suppliers introduced the 54Mbit/sec. technology known as 802.11a.

Now suppliers are offering a third alternative, 802.11g, which they claim delivers 802.11a speeds over longer distances, while supporting backward compatibility for 802.11b devices -- something 802.11a technology failed to do.

But don't rip out that 11b infrastructure, users and analysts say. PC cards and access points using 11g are still immature and provide few real-world benefits today. By this time next year, though, you'll probably end up buying it anyway because the technology will be built into most hubs, or access points.

The benefits of 802.11g derive from the limitations of its two predecessors. Released first, 802.11b uses a transmission technology called direct sequence spread spectrum, or DSSS. Each access point supports up to 11Mbit/sec. over three channels in the 2.4-GHz frequency range (see table). But the technology has a few drawbacks. Having only three channels increases the likelihood of co-channel interference between neighbouring access points.

And, as with all Wlan technologies, actual throughput is at best only about half the published data rate and drops as distance and the number of users per access point increase. Wlans also face interference from microwaves, mobile phones, Bluetooth devices and even Pentium processors operating in the relatively crowded 2.4-GHz band.

With its higher bandwidth (best case of about 26Mbit/sec.) and up to 13 channels (with 11 more likely to gain regulatory approval in the next two years), 802.11a is a better choice for areas with a high density of users or for applications that require higher data rates. It operates in the less crowded 5.2-GHz frequency band, but it has a shorter range, and its modulation technique, called orthogonal frequency division multiplexing, or OFDM, won't support 802.11b devices. To support both 11a and 11g client types, corporations must buy more expensive, dual-mode access points.

As for 802.11g, it addresses the backward-compatibility issue - sort of. It adapts OFDM to allow 54Mbit/sec. operation in 802.11b's 2.4-GHz band, while supporting 802.11b devices. "It gives you three more channels of 802.11a," says Rich Redelfs, vice chairman of WLAN chip-set maker Atheros Communications. But does that matter? Probably not, he says, because most corporate uses don't yet require the bandwidth.

"We're having this technology burst that is delivering technologies in excess of demand," says Ken Dulaney, an analyst at Gartner. That's certainly the case at Embarcadero Systems, which uses pole-mounted 802.11b access points on its loading bays. "Range matter, while our bandwidth requirements are not high," says chief technology officer John Montgomery.

Montgomery says that in a few areas in the yard where workers tend to congregate, having higher bandwidth might be beneficial. The 802.11g standard promises higher bandwidth than 11b, with comparable range. But with current 11g devices based on a draft specification and lacking Wi-Fi Alliance interoperability certification, the decision was a no-brainer: "We will pass on [802.11g] and use 802.11a," he says.

"If you're trying to put these in an indoor office space, range is not an issue," says Redelfs. Network administrators need to maximise bandwidth for each user. One way to do this is to place access points closer together, but that increases the likelihood of co-channel interference, he says.

The best solution for 802.11b users in this situation is to move to 802.11a/b hubs. "The beauty of 802.11a is you have so many channels you don't have to worry about interference between access points," he says. Another option for 802.11a devices is to "dial down the power" to group access points more closely, he says.

The IEEE is working on a standard, 802.11h that will support adjusting 802.11a power levels and changing channels on the fly to avoid interference with other access points and devices operating in the 5-GHz range, such as radar. But that standard, also called Spectrum Managed 802.11a, is still in committee.

One area where 802.11g could eventually shine is streaming video, says Gartner's Dulaney. But acceptable quality will require solidification of the emerging 802.11e quality-of-service draft standard, which has progressed slowly in committee. And while video streaming over a Wlan connection may work for home users, today's throughput rates still may not be enough to support it in a business environment, where many users share an access point, he says.

For now, Dulaney says, corporate IT shouldn't get too far ahead of the curve. "Buy b now and then switch over to buying a/g at the end of the year," he says.

Why faster G means slower B

Got 802.11b? If so, installing 802.11a/g access points will actually slow throughput rates for existing clients and throttle back the performance of newer 802.11g clients as well, suppliers and analysts say.

How big an impact will this have at the individual client level? The final answer won't come until sometime in 2004, once more mature, second-generation devices arrive and enterprises begin field-testing products.

The performance issue arises from the way in which the 802.11g standard provides for backward compatibility with 802.11b devices. While 802.11g access points can communicate with both client types over the same frequency band, each uses a different communication protocol.

Since a and b devices can't see each other, the default collision-avoidance scheme for managing who gets to talk doesn't work. To solve that, 802.11g access points supporting mixed clients go into "protected mode" and coordinate traffic using a request-based protocol that adds to network overhead. "You're running two different protocols and time-sharing them," says Atheros vice chairman Richard Redelfs.

The design favours 802.11g clients, but throughput for both clients is degraded, he says. "We get frustrated when people say g is backward compatible and as fast as a. If you want to be backward compatible, you're not as fast as a," he says.

Supplier tests of protected-mode performance vary. Tests by Intersil and Atheros show that rate declines could be 40% or higher in some situations. But those aren't real-world tests. Gartner analyst Ken Dulaney expects the degradation of throughput to the client in the range of 10%. "You will slow it down a little bit," he says, adding that he doesn't think most users will notice.

"The bottom line is it doesn't matter," he says, because most clients aren't constricted by bandwidth.

Redelfs is less optimistic. "If you're concerned about bandwidth [for 11b users], you're much better putting an a/b access point up, because you won't touch the b users at all," he says. Alternately, users can force access points to run in 11g mode only. That's technically out of specification and leaves no arbitration for b and g nodes, so collisions will increase with network traffic loads. But devices will probably still be interoperable, Redelfs says.

The throughput limitations of protected mode are also likely to decrease as more mature designs appear, Redelfs claims. Initial 802.11a chip sets, for example, fell short of expectations. Second-generation designs have fared better. "We've dramatically improved the range and throughput of a," he says. He expects similar gains in future 802.11g designs. "There are still huge advances we can make in range and throughput," he adds.

Read more on Wireless networking