Voice on Wi-Fi - the architecture question
So, will we change our Wi-Fi for voice or not?
By Peter Judge. Techworld | Published: 10:00, 06 November 2006
Aruba is claiming a major breakthrough in voice on Wi-Fi (VoFi). But the claims come after other vendors have argued that Aruba's kind of centrally-managed wireless LAN is simply not capable of scalable VoFi.
We put those criticisms to the company's vice president Keerti Melkote, and product director for voice Peter Thornycroft - and got more detail from them about the company's VoFi announcements.
The trouble with VoFi Wi-Fi was originally conceived around a single access point acting as a hotspot, and that model has had huge success in the home. In the office, or on a campus, access points have to be linked to cover an area, and enterprise switches' main job is to manage that.
Voice on Wi-Fi throws another spanner in the works. It's real-time, unlike the data applications currently in use, and it's going to be delivered in handheld devices (single-mode Wi-Fi handsets, or dual-mode phones. So users will roam more from cell to cell, at the same time, while running an application uniquely sensitive to breaks in service.
Issues like this led radical vendors such as Extricom and Meru to argue that WLANs must eliminate roaming with a "blanket" architecture that sets all wireless access points to the same channels, while more traditional Wi-Fi switch companies like Trapeze (as well as newcomers such as Colubris) argue for decentralising enterprise Wi-Fi architectures to make them scalable.
Aruba's Thornycroft claims the existing opportunistic key-caching technique, which makes access points ready to talk securely to any phone nearby, gives a handover time that less than 20ms. He also says its scalable, claiming that Ohio University will soon have the world's largest campus WLAN, with around 10,000 access points.
Why blankets aren't a comfort zone Melkote dismissed "blanket" architectures: "It may solve the roaming issue, but you can't solve that problem alone," he said. "You must solve it in the context of multi-service WLANs. Making an architecture optimised to solve one problem alone at the expense of the others is the wrong way to do things."
Blanket architectures may also lag on standards, such as the Wi-Fi security specifications, because they use different technology, he warned: "They've built a single channel model, but it's still not WPA2 certified," he said. "They have to rev their silicon every time a standard changes. We're using standard silicon." More significantly, he claimed that blanket architectures increase the cost of WLANs by using proprietary technology. If that means there's a price war coming in business WLANs, that could be good for everyone.
He also suggested these architectures may require security keys pushed to all access points: "That will limit the number of devices you can support, because of the table size on the APs. Suddenly, it doesn't scale well."
Blanket have downplayed the complexity in their networks, said Melkote. While the client doesn't have to make a roaming handover decision, the issues are deferred into the network, where flows of traffic must be de-duplicated and assigned to clients and applications. Aruba says users that have looked at blanket Wi-Fi have reported issues at the margins of cells: "Filtering out duplicate traffic in the uplink is not too tricky, but downlinks are harder."
Decentralising is nothing new? Melkote joined with other providers, in claiming to have delivered elements similar to Trapeze's more-distributed architecture before. "The distributed forwarding model is not new," said Melkote. "We launched a remote AP with distributed forwarding, so to me they are just responding to our message - late."
This is a WAN product, not a LAN one, he agreed, but distributed forwarding is only necessary on the WAN, he said. "Inside the campus, there is adequate capacity in the LAN - it's never heavily utilised."
This capacity and scalability is an issue Aruba has been criticised for. Melkote concedes that, Aruba systems require all wireless traffic flows to pass through the central controller, because they apply encryption centrally: "But carrying traffic to a central point is not an issue from bandwidth point of view." There's plenty of scalability for Aruba, even when faster Wi-Fi systems come on-stream, he said: "Our controllers can scale to 7.5 Gbit/s and we have a roadmap to 40 Gbit/s. We don't see it as a bottleneck.".
"Ultimately all networks are centralised, and the bottleneck is the pipe into the data centre." He reiterated Aruba's often-made suggestion that wired network should be sent through the controller as well as wireless, in order to apply security consistently.
The WLAN's role in convergence By 2007, Melkote says Aruba's switches will have a role in handing calls between switched and IP networks, with an open API to interface to deliver this service to companies' PBXs. A telephony gateway will set up a switched call when the phone moves out of Wi-Fi range, and the Aruba switch will decide when to drop the Wi-Fi leg and cut across to the switched network. "We are in the best location to determine when you have moved outside the Wi-Fi zone," said Thornycroft.
In converging the two, Aruba will support the telco's UMA specification, which Thornycroft reckons will have a long life, as full SIP/IMS implementations for convergence will take longer to produce. "The cellular world is looking to IMS, but there are different opinions on how quick it will be to get there," said Thornycroft. "We are SIP and IP already, so we are further on than the cellular carriers. The cellular carriers are pushing IMS further out and UMA is the primary technology to fill the gap before we get to nirvana. The longer that gap is, the more successful UMA is likely to be."
Simplifying the handset Dual-mode handsets are still evolving, and Aruba wants to help that along with a specification for them. "There are eleven portions of the IEEE 802.11 standard that need to be implemented in any dual-mode handset," said Thornycroft, citing 802.11e, 802.11k, 802.11r and others. Aruba will publish version 1.0 of the client behaviour specification, a profile for wireless VoIP handsets that should guarantee QoS, performance and battery life, as well as making sure handsets all work properly with other vendors equipment.
"Most vendors use a low cost Wi-Fi handset chipset," said Thornycroft. "There's very little enterprise class behaviour in that. We will write reference code for a dual mode handset, that will be implemented next year."
This sounds a lot like a proposed standard, and as such it ought to have multi-vendor involvement and be put through an official body, Melkote agreed, "They are already working on this, but we see a need to move a little bit faster." Nearly everything in the client behaviour document is based on things published in 802.11, and is already going to be included in a Wi-Fi Alliance brand - but that won't be out till the middle of 2007.
"[the Wi-Fi Alliance brand] will be dated when it comes out, because of the way standards bodies work," said Melkote. "We're absolutely trying not to invent our own stuff," he assured me.
VoFI coming into its own? All this complexity begs the question, why would anyone do voice on Wi-Fi? Aruba, like all sophisticated vendors, has moved on from the early arguments. "Early discussion centred around cost-savings, gained by offloading traffic from licensed spectrum to unlicensed spectrum," said Melkote. "We don't believe that is a primary driver. The driver is greater mobility." If all calls go to a single phone, there will be increased productivity, he said.
VoFi has taken off more slowly than Aruba expected, said Melkote, but the main part of the market was always going to wait on dual-mode phones: "Now we've seen them come out, it gives me a lot more confidence."
"As in the cellular market, the handset is the slowest component - the one that gates things," said Thornycroft. "Handsets have arrived, on the Wi-Fi side, but they are still built against the model of an isolated AP, not an office network."