News

World wakes up to true speed of WiFi - just as it gets faster

by Guy Kewney | posted on 23 May 2003


This week, at the WLAN Event in London, chip maker Intersil is demonstrating - live - a mixed mode WiFi service, running at full speed. That is, the 802.11b (legacy) system is running around 3-5 megabits per second, while the just-approved 802.11g circuits are running around 15 to 20 megabits.

Guy Kewney

The final "rubber stamp" of full IEEE approval of the 11g standard comes in June, and the WiFi Alliance says it will start testing rival products at that point, to ensure they are genuinely compliant, and will genuinely work together with each other.

The news appears to have astonished some observers, who seem to have believed that the old "signal speed" classifications were in some way a reflection of true throughput. That is to say, the world is full of people who honestly think that 11b runs at 11 megabits per second, and who were expecting 11g to run at 54 megabits.

That never was the case - any more than any motorcar can deliver the miles-per-gallon figures that advertising claims. Theoretical maxima are one thing; the real world is quite another.

In the real world, the mess of "pre-G" standards is tidied up by the new official standard. The main beneficiary will be 11g, which will be able to work very nearly at its full data throughput rate even when 11b nodes are operating in the area.

Intersil's implementation of the new standard is called Nitro, and the company's top technologist, Dennis Eaton - who is also chairman of the WiFi Alliance - demonstrated the Nitro technology yesterday in Olympia. He emphasised that this technology was built into all new-g standard

Back in April Intersil claimed a three-times speed up for 11g in "mixed mode" environments.

This was based on the observation that normally, the 54 megabit signal speed of 11g would allow the entire network a data throughput of around 22 megabits maximum - that is to say, no matter how many users were online, the total amount of bits "in the air" at any time would never exceed 22 megabits.

But when an 11b device came into this environment, it "choked" the 11g throughput. "Some solutions on the market ... do not exhibit the ability to support even the minimum protection mechanisms required by the latest 802.11g draft standard," Intersil remarked at the time.

"These systems allow .11b and .11g radios to transmit packets without mutual awareness. This causes collisions, transmission errors and resultant retries that degrade overall system performance. This problem is addressed in the current 802.11g draft standard, which incorporates protection mechanisms that effectively control the traffic over the network."

One primary cause of slowdown was the simple fact that it takes a lot longer to send a data packet over 11b than over 11g. So if an access point is sending equal numbers of packets to two clients, one 11b and one 11g, then it's going to spend most of its time talking to the slow one. Nitro uses the provisions of the new 11g spec to send several packets in a burst to the 11g client.

The discovery that the old 11g spec was flawed seems to have eluded several observers, who have now discovered it inside the publicity for the new.

One report this week actually said that the new spec would suffer from the very flaws that it was devised to fix.

The report quoted Randy Conklin, director of operations for Broadband Central, a wireless Internet service provider based in Draper, Utah, saying that "the 10Mbit/sec. data rate for 802.11g isn't good enough for advanced applications such as voice over IP or video."

Conklin claimed that to support those applications, Broadband Central would need at least 20Mbit/sec. data rates. "As a result, the service provider will look to deploy pure 802.11g service offering the faster data rates," added the report.

In the real world, there is no way to offer a pure 11g service for the foreseeable future; too many potential customers simply don't have it.

But the news that you need 20 megabits per second for VoIP will come as a considerable shock to those of us who have been doing it successfully over modem links, for years; and anybody planning to run uncompressed 20 megabit video over a wireless link had better plan to be the only user, and to be sitting within 15 feet of the access point.

The expectation of US Robotics, which has announced its own proprietary enhancement to 11g (a system which will work in standard 11g mode as well, of course) which will provide a theoretical signal speed of 100 megabits, is that even this "100 megabit" nominal system, will probably provide around 500 kilobits per user in most typical offices.

If that sounds slow compared with 100 megabits, it's worth reminding ourselves that most people still can't get broadband Internet at 512K over copper wire.

First fully-approved, WiFi compliant, tested and stamped 11g devices will start appearing on the market in July, barring last minute disputes about the standard.

There is still a slight chance that the standard could be held up to August/September, some sources insist. But the momentum for the new standard is such that even if someone does find a flaw, they are likely to be outvoted by the need to get product into the market.

Broadcom's chipset may be flawed, but Linksys and other customers have sold five times as many 11g systems month on month compared with the sales of the older 11b standard, and the rate is climbing. Further delay would simply give more of the market to Broadcom, which is not acceptable to the many other players in the market.


You can discuss this article on our discussion board.