The Rewards And Perils Of Meshing With Wi-Fi

Promise: Wi-Fi APs act as self-configuring routers, creating multihop meshes that find their way around radio interference, physical obstructions, and network congestion. Architecture is automated, Ethernet eliminated.

Players: Four start-ups have shipping products: Tropos and BelAir specialize in outdoor meshes, while Strix and Firetide also make indoor systems. Interoperability will have to wait for 802.11s, a standard unlikely to reach shipping products until 2009.

Prospects: The start-ups have already proven that wireless meshes are cheaper, more flexible, and less vulnerable to outages than LAN cables or T1 services. But all use proprietary technology, and their bandwidth will never scale to gigabit applications.

Wireless mesh networks have been a popular idea for years. Visionaries talk about "smart dust"--tiny radios that organize themselves into networks as needed, bypassing service providers and eliminating the need for wires of any kind. New technologies such as ZigBee aim to extend the range of very-low-power networks by making every network node act as a router.Wi-Fi meshes are less ambitious but more real. Because they need to work with ordinary IEEE 802.11 clients, they don't try to set up true peer-to-peer networks where every laptop or phone acts as a router. Instead, the mesh only connects access points (APs) together, replacing the Ethernet cables that would otherwise be needed to link them to the wired backbone.

Six start-up vendors have already shipped proprietary Wi-Fi meshes. The IEEE is working on 802.11s, a standard for inter-AP links. All can make networking cheaper and simpler, but none are a complete replacement for Cat 5. Applications requiring low latency need more expensive multi-radio architectures, and the airwaves will never be able to provide as much bandwidth as cables.

IN AND OUTMeshed Wi-Fi is most useful in places that aren't already wired for Ethernet. For most enterprises this means temporary offices or buildings such as factories and warehouses. But overall, the least wired place is outside, so most vendors are concentrating on outdoor systems (see table below). The market for these is booming, mostly for regulatory reasons. The FCC's new rules about DSL are forcing ISPs to find alternative access methods, which usually means wireless. Simultaneously, an increasing number of municipalities are building out public Wi-Fi networks.

Tropos Networks and BelAir Networks both focus exclusively on metro-area systems. These mainly target carriers and government, but can also be used by enterprises to link large campuses together. Both Strix Systems and Firetide started with indoor systems, but have since expanded into the outdoor market as well. Like all Wi-Fi standards, 802.11s will officially cover only indoor systems, the theory being that the access market is better served by other technologies such as WiMAX (see "Wi-What?").

The existing Wi-Fi vendors initially stayed out of mesh, saying only that they would eventually support 802.11s, the planned standard. The growth in the municipal Wi-Fi market has persuaded them otherwise. A year ago, Nortel Networks launched an outdoor system intended to compete with BelAir and Tropos. Last month, Cisco Systems did the same, using technology it acquired with Airespace in January.Motorola has also spent its way into the market, buying the generically named MeshNetworks, a company that specialized in true peer-to-peer networks. These can achieve very high reliability, but require proprietary (non-Wi-Fi) clients, limiting their appeal to all except the military and public safety agencies.

Cisco is hoping to sell its outdoor mesh to enterprises, pitching unified management as the main advantage: The outdoor nodes are controlled by the same Airespace box as indoor Aironet APs. But customers who want to avoid wiring a building are still limited to Firetide and Strix.

HOPPING IN THE AIR

In the simplest Wi-Fi meshes, all APs and clients share the same frequency channel. This makes the APs relatively cheap because each one only needs to contain a single radio. The problem is that because the channel is shared, so is the bandwidth. The APs act like hubs, so the mesh performs similar to a flat network built entirely of hubs. All clients contend for access to the same bandwidth.

Worse, a single-radio system can't transmit and receive at the same time, introducing a noticeable delay for every hop. And because all hops use the same channel, interference means each extra hop halves the total available bandwidth. The result is that after a few hops voice and data come to a crawl.Despite these disadvantages, single-radio systems are still popular. They provide a low-cost way to cover a wide area, exactly fitting the needs of most municipal Wi-Fi plans. However, enterprises more interested in the quality of coverage will need dual-radio or multi-radio systems.

A multi-radio system uses one radio for ordinary Wi-Fi links to clients and the rest for meshed links to other APs (see figure). In most architectures, the client links are based on 802.11b/g, the 2.4GHz standard supported by most Wi-Fi hardware. The mesh itself is usually based on 802.11a because its 5GHz frequencies are less crowded, so there's less risk of interference between the mesh and Wi-Fi LANs. However, 802.11 doesn't natively support meshes, so every vendor needs to run its own proprietary technology on top of 802.11a. The 802.11s standard aims to replace those proprietary technologies, covering single-radio and multi-radio systems.

Currently, metro-area vendor BelAir is the only player to sell both single-radio and multi-radio systems, designating the multi-radio gear for the network core and the single radios for the edge. Firetide sidesteps the issue entirely by including an Ethernet hub instead of an AP for client connections. This lets users choose their own AP vendor (usually Cisco), or connect to non-Wi-Fi devices such as security cameras. Strix has a modular system that supports up to six radios and can also use Bluetooth for client connections.

It's possible to provide high-quality coverage using single-radio systems, but only if the mesh is architected so that each node is within a hop or two of an uplink to the Internet. For example, the city of Chaska, MN, uses 230 Tropos nodes to cover a 16-square-mile area. The network is very successful: It allows the city to offer all its residents mobile Internet access at DSL speeds for $16 per month and is set to pay for itself within three years at current subscriber levels. However, because the Tropos APs contain only a single radio, the network needs 36 uplinks at T1 speeds distributed throughout the city.

In contrast, the city of Tempe, AZ, uses about 400 Strix nodes to cover a 40-square-mile area. The network is able to get by with only seven T3 lines because the Strix nodes all have at least two radios. This means additional hops don't significantly increase latency or reduce bandwidth. Packets take less than 40ms to travel 10 hops--a delay too short for human listeners to notice even when the network is used for VoIP.MESSY MESHES

Compared to the existing products, the IEEE's aims with 802.11s seem rather modest, targeting meshes of only 32 APs. Its ETA is similarly unambitious: The official timeline says the final specification will be published in July 2008. Even when standards are finalized, it takes at least another six months for the Wi-Fi Alliance to check that different vendors' products actually work together.

This timeline assumes that vendor infighting doesn't get in the way. At the IEEE's September meeting, there were nine rival drafts (which actually represents progress over the 15 in July and 35 last year). The main disagreement is over how much functionality should be built into the standard, and how much should be left to optional or proprietary extensions.

Intel, Nokia, and Motorola are promoting a relatively concise specification they call the Simple, Efficient, and Extensible Mesh (SEEMesh). Its main competitor is the Wi-Mesh Alliance, a Nortel-led group aiming for tighter integration between 802.11s and 802.11e, the nearly complete standard for QoS. The principle behind Wi-Mesh is that APs should be able to reserve bandwidth or a specific route, creating virtual circuits for the meshed backbone links. Compared to SEEMesh, this improves reliability but adds protocol overhead.

Regardless of which proposal is chosen, some extensions will still be needed. There are even more possible routing algorithms than 802.11s proposals, and researchers continue to develop more as they experiment with different network architectures. Routing within a wireless mesh is a much harder and newer problem than routing on the Internet backbone, as interference means links between nodes can fluctuate randomly in bandwidth and reliability.At present, all vendors use their own proprietary algorithms, optimized for their particular hardware. 802.11s will likely allow these to be used, but mandate support for at least two standard routing methods. For relatively stable networks, each node will keep a complete map of the others in memory. For networks where the APs themselves are moving, each AP will be aware only of its nearest neighbors.

Click here to read Andy Dornan's posts. Write to him at [email protected].

Wi-What?

The popularity of outdoor meshed Wi-Fi systems hides an awkward technical detail: 802.11 is intended as a short-range LAN technology, not a replacement for DSL. To Wi-Fi's fans, this doesn't matter. Extending 802.11 to access lines is simply innovation at work. Rejecting long-range Wi-Fi on that basis would be like rejecting VoIP because IP was designed for data.

The IEEE isn't convinced. The 802.11s standard concerns only indoor systems, and vendors are developing plenty of other technologies aimed at the outdoor market. Most prominent among them is 802.16, which is the basis of WiMAX. Its main rival is 802.20, a broadband mobile technology developed by Flarion Technologies, now a subsidiary of Qualcomm. Then there's 802.22, a proposal to transmit data in unused TV frequencies. And that's just the IEEE. The cell phone industry is developing various 3G-derived technologies, though like 802.20 these face the barrier of requiring licensed spectrum.So should outdoor network architects wait for these other technologies? The overwhelming answer from city governments is no.

WiMAX's BROTHER

Compared to alternatives, Wi-Fi is cheap and ubiquitous. A basic 802.11 NIC can add less than $10 to the price of a PC, while 802.16 terminals are bulkier and cost at least 10 times as much. There's also a lot of justified confusion about the new technologies themselves.

WiMAX is the clear leader, but it tries to do so many things that interoperability may be difficult--a 3G-like mobile network and a point-to-point microwave link can both claim WiMAX compliance, for example. To narrow its scope, the South Korean government is developing its own 802.16-based system, called WiBro. Likely to be a subset of WiMAX, it's already proving popular, and not just in Korea. In September, Sprint Nextel announced a U.S. trial using Samsung hardware.

The mesh vendors themselves claim to be technology-agnostic. Though they're all focused on Wi-Fi now, their secret sauce is in the proprietary protocols used to link APs together, not the commodity 802.11 links that cover the last few feet. All can easily switch to WiMAX or whatever their customers demand.0