Wiping out Wi-Fi's original sin
See the light and use the blanket?
Cell-planning is the Original Sin of Wi-Fi according to Gideon Rottem, chief executive of Extricom, a company dedicated to preaching a different gospel.
"The sin at the base of Wi-Fi networks is to use the scarce resource of spectrum to increase coverage," he says. In his view, Extricom is there to remove that sin.
Wireless LAN switches, from Trapeze, Motorola/Symbol, Aruba, Cisco and others, use a set of wireless LAN access points to cover a building. Each AP has to be on a different channel to its neighbour - and in the 2.4GHz spectrum used by 802.11b and 802.11g, there are only three non-overlapping channels. The best way to fill a large building with Wi-Fi is a hexagonal array of APs, and that sets up points where three APs' coverage overlaps.
So you can do wireless LANs in 2.4GHz, and that's what those vendors have been doing now, for up to five years in some cases. But now 5GHz is opening up, some of them admit it's been a struggle. "The fact remains that there is simply not enough spectrum available in 2.4Ghz space to ensure a quality service," said Pat Calhoun, CTO of Cisco's wireless business unit, in his blog. "…no matter how creative your channel plan is, any deployment that exceeds three APs will see some co-channel interference - including from devices that may not be under your control (is your neighbor running a network?)."
But Rottem reckons it's been wasted effort. It's better to put all the access points on the same channel, and use processing power to sort data packets and send them to the right access point to reach the client. That's the basis of Extricom's "channel blanket" approach, which we've covered here pretty often.
Just like all the other wireless switch sellers, Rottem reckons that 2008 will mark a major take-off of Wi-Fi, with the arrival of 802.11n, and dual-mode Wi-Fi handsets. He calls it an inflection point, when users will snap up wireless, to run voice, and support multiple applications.
His claim is that, with or without 802.11n, the other architectures can't deliver. They can't do mobility, because they hand-off too slowly between access points, and they can't deliver quality, because of co-channel interference, he believes.
It's a familiar Extricom story, but Rottem comes down pretty hard on rivals - especially Cisco. He visited Cisco to pitch the Extricom vision, shortly before Cisco bought into the cell-planning model, with Airespace in 2004. "I went to show we could do less than 50ms handover," says Rottem. That's what Cisco claimed it could achieve with its standalone Aironet access points, he says: "Dave Leonard [then Cisco's wireless chief] asked 'Why 50ms? Marketing does 50ms, we do 150ms'."
So if even Cisco says 2.4GHz isn't good enough, will 5GHz help? It has a lower penetration than 2.4GHz, says Rottem, and cells have to be smaller. The MIMO element of 802.11n will increase this, but it will be unpredictable, and that might make cell-planning a nightmare.
And 5GHz 802.11n won't have as many channels as it's cracked up to have, he says. Cisco's N access points may only do three non-overlapping channels in the 5GHz band, he believes. That's a conversation stopper for me, because till now, I haven't got past the usual assertion, repeated in Cisco's Aironet 1250 data sheet that the 5GHz band has "up to 24" non-overlapping channels.
Look more closely says Rottem. Use 40MHz channels, and take international regulations into account, and the number goes down. There's a footnote on the Cisco datasheet, that suggests there may only be nine non-overlapping channels in the US, and Extricom has seen Gartner advice that suggests only three channels in the end, he says.
"It's an interesting issue that I'm following up. Wherever we get to with that, 802.11n falls down because it focuses on range and speed," says Rottem, and will be dragged down to poorer performance by the presence of b and g clients. "The promise of MIMO is stability for a single client, but we have multiple clients."
Blankets, by contrast, allow an enterprise-grade N network, even on 2.4GHz, since they can have one 40MHz channel for N and a 20MHz channel for b and g - a scheme which would, I suggest might be just as vulnerable as a cell-planning system, to interference from neighbours' Wi-Fi - though Extricom argues that its architecture offers more resilience because signals can be sent and received form any AP.
But is there a different agenda here, I ask? Is Rottem downplaying 802.11n, simply because it's going to be tricky to implement it on a blanket system? To build a blanket, Extricom has taken the MAC layer off the access point, and re-implemented a central MAC on the switch for all access points. They still have a lot of WLAN smarts (they're not a distributed antenna system) but they all have the same MAC address. "We're actually a MIMO-like array, but at the network level," says Rottem's colleague, marketing VP David Confalonieri.
"The reason we put the MAC at the centre is because it is much better," says Rottem."Anyone who says don't do it, just can't. That's why they say it."
But other vendors can pick up N chipsets from Atheros as they are, while Extricom has to re-write the MAC to put it centrally. That causes a delay, but not much of one, says Rottem. He's determined to ride the wave of N hype that is coming, and that means he's going to at least announce N sometime this year.
"We're going to do it," he says. "We are going to have the best N network in the world. Cisco can have an N access point - we'll have an N network."