For Peter Ecclesine, mesh networking is the best thing to happen to Ethernet since full duplex. That's because meshes essentially open up a whole new dimension that could stoke growth of any wireless net, from 802.11 to Zigbee. And, theoretically at least, the cost of admission can be low, requiring just a half megabyte of code and some tables to handle a store-and-forward process, according to the technology analyst for Ethernet giant Cisco Systems Inc.
Indeed, mesh networks open up opportunities that range from smarter home networks in suburbia to cheaper last-mile links in big cities and distant rural villages. In between are a host of applications, from easier-to-manage business networks, strategic battlefield and antiterror communications for the military and public-service agencies, and an emerging class of uses still being defined for sensor networks.
But the broad set of opportunities is also opening up a Pandora's box of technical problems-not the least of which is how to set standards for a technology headed in so many directions at once. Latency, battery life, quality-of-service and security all become thorny issues, as is the need to set at least
a baseline for performance. "We have a broad range of market scenarios, so it will be important to find common ground among the proposals in order for .11s to be successful," said W. Steven Conner, a wireless architect in Intel Corp.'s communications labs, who edits the IEEE's 802.11s spec.
Overall, "the problems [with mesh networking] are in how to do this taking into account quality-of-service and security, while getting efficient use of limited bandwidth of the shared wireless media," said Donald Eastlake, a distinguished member of the technical staff at Motorola Labs, who chairs the 802.11s effort to standardize mesh technology for Wi-Fi.
Setting a baseline for performance is one of the critical challenges, said Ecclesine of Cisco. Systems could eat up a lot of processing power handling messages and calculating routes as nodes come and go in the net, especially given the relatively long range of links for 802.11a. Latencies for nodes on a mesh might range from 100 milliseconds to 100 microseconds, Ecclesine added. Given those realities, some say the management problems constitute mesh technology's biggest technical hurdles.
Meanwhile, hot debates rage over a number of implementation issues, such as whether to define meshes at the media-access or network layer and whether meshes should be enabled in all client devices or only on access points. Others debate what slices of scarce spectrum are best suited to meshes, with proposals ranging from 700 MHz to 60 GHz.
The issues are so varied and so daunting that Microsoft Corp. researcher Victor Bahl has concluded that although meshes are viable, today's technologies remain inadequate.
Bahl convened a summit last June of more than 25 researchers from a dozen universities and a handful of established and startup companies to discuss mesh networking. Nine months later, Bahl is considerably more upbeat than he was then. The summit sparked a number of papers and collaborations on mesh networking. Microsoft Research has developed key enabling software that it is trying out on separate 30- and 65-node test beds running near its Seattle labs.
"We are engaged with many universities, giving them source code and funding to deploy their own test beds. In a year or two we could have a great solution working in multiple test beds," Bahl said.
The hunt for the optimal spectrum for meshes leads to a host of regulatory and business issues. Typically, 802.11 nets use .11a for the mesh and .11b on clients, in part because the .11a spectrum at 5 GHz is more widely available. It has relatively poor propagation through walls, however.
Ecclesine noted that a 5-GHz mesh designed to cover the 298 square miles of New York City might require tens of thousands of nodes, whereas Verizon's cellular network there uses about 1,000 basestations. By contrast, Flarion Technology Inc. (Bedminster, N.J.) has demonstrated its wireless technology covering a 70 x 50-mile swatch of Research Triangle Park, N.C., using a 1.9-GHz frequency and just 110 towers, Ecclesine said.
The Cisco analyst said the 23- and 60-GHz spectrum areas are relatively available for communications of less than a mile in distance, but whatever is chosen will have to contend with business hurdles around putting up new towers. Bahl of Microsoft suggested that low-frequency communications in the 700- and 900-MHz range provide the best distance characteristics and may become more widely available once the Federal Communications Commission gives back some of the analog TV spectrum.
Whether to implement mesh at the media-access (Layer 2) or network (Layer 3) level may surface as a point of contention. "This could become a cultural debate over whether the routing protocol belongs with the IEEE or the IETF [Internet Engineering Task Force]. I don't know how this will work out," said Bahl.
Intel wants to implement mesh at the media-access control (MAC) layer to avoid difficulties assigning Internet Protocol (IP) addresses to devices that cannot be directly "seen" at the network layer. Some existing products already implement mesh at Layer 3, however, said Intel's Conner.
"We created a Layer 2.5 where our routing software sits," said Bahl, referring to Microsoft's Link Quality Source Routing, a variant of Dynamic Source Routing. "It takes into account what's happening at the link, although this violates some of the principles of layering."
Another hot issue is whether to implement mesh only at the access point (AP), making it easier to embrace existing systems, or to enable it across all clients, creating a richer potential of network routes. Home networks prefer to enable meshing for all clients, while many of today's metro networks implement meshing only at the AP.
"Doing the mesh work only at the AP is good for legacy systems, while a peer-to-peer architecture is more futuristic, and I think that's a fine position," said Bahl. "But how would you get such products into the market?"
Developers seem to have reached a consensus that an optimal mesh uses multiple radios, something Bahl has advocated for some time. Executives from startup MeshDynamics (Santa Clara, Calif.) make the case for their three-radio architecture in a contributed article now online at CommsDesign.com.
"Radios are becoming low-priced, and we have lots of good scheduling algorithms already," Bahl said. Using multiple radios, he said, "we think we will be able to go seven to nine hops with VoIP [voice-over-IP] quite nicely."
But Microsoft has also developed a technique called slotted seeded channel hopping, which boosts the spectrum efficiency of single-radio meshes using fast switching among three 20-MHz channels.
"No one has built NICs [network interface cards] with the idea of switching channels fast, so today it takes several milliseconds," said Bahl. "But moving forward, channel switching can be done faster-say, in 80 microseconds-so that you can use the spectrum much more effectively. This really enables the single-radio mesh."
The .11s standard will require a single-radio mode but will also support multiple-radio options. Similarly, .11s aims to support encryption security handled at each link as well as architectures that handle security only at the ends of a link and not at intermediate hops.
The .11s task group, which has attracted more than 70 attendees, expects to start reviewing as many as 34 standards proposals at its July meeting. BelAir Networks, Motorola's MeshNetworks group, Nortel Networks and Tropos Networks are among the top stakeholders expected to submit proposals.
"These are the four leading players deploying citywide networks with some intellectual property they would probably like to see get into the standard," said Cisco's Ecclesine.
In addition, companies such as Intel, Nokia, NTT Docomo, Samsung, Sony, STMicroelectronics, Texas Instruments and Thomson have expressed interest in proposing their mesh technologies. The .11s group could put forward its draft pick to the broader 802.11 organization within a year. A final standard should be filed by the fall of 2007.
Intel's Conner believes that standard should define ways to discover and set up secure mesh links, extending the .11i standard. It may also need to define a default method for finding an optimal mesh path, then let nodes negotiate for better path selection techniques if they are available.
The .11s work also needs to fold into Wi-Fi a host of past developments in wired Ethernet standards, including support for virtual LANs, 802.1x security, and existing standards in discovery protocols and operations-and-management techniques.
Adding to the complexity, .11s should build bridges to other mesh efforts such as the Intelligent Transportation Systems work on 5.9-GHz meshes, linking cars and smart highways. Separately, the IETF has been working on Layer 3 routing for mobile, ad hoc networks under its Manet program for many years.
Unlike the broad Manet effort, the .11s group is tightly focused on what it considers a sweet spot of networks of up to 32 basestations and 1,000 clients. That solutions segment aims to embrace home, neighborhood and city meshes.
Home, neighborhood, city
Microsoft initially considered suburban neighborhood networks a hot emerging market for meshes. Researchers envisioned ad hoc communities creating shared Webcams for neighborhood security, pooled storage for caching and backup, and local libraries of TV shows and movies. "You don't have to go all the way up to the cloud to talk to the person three houses down," said Bahl.
But in the past year, the company discovered that users are averse to some apps, such as sharing hard-disk space. So now Microsoft is shifting its focus to metro and rural mesh nets that could bring broadband access to more than 32 million homes in the United States alone.
Cisco Systems believes that with the .11s technology, unlicensed Wi-Fi will beat WiMax in many last-mile scenarios. "New York City wants 2-Mbit distributed Net access across the city. How can you backhaul all of that?" said Ecclesine. "It's not likely we'll have clear line of sight, but when you have mesh you don't need line of sight."
Splitting its bets, startup BelAir Networks Inc. (Kanata, Ontario) recently announced it will implement its mesh technology on Wi-Fi as well as WiMax networks.
Home, office and industrial scenarios are also hot targets for mesh. Wi-Fi meshes would let mobile office workers move from one cubicle to another without the need for running new wires or even making any manual network administration changes, Ecclesine said.
Intel demonstrated that a home equipped with mesh-enabled Wi-Fi clients could help solve two key problems. Wireless home nets are currently weak on whole-house coverage, but meshes could let traffic span a home by making multiple room-to-room hops. In addition, meshes can negotiate for optimal routes, opening the door to better A/V wireless networking.
Backers of Zigbee, meanwhile, are focused on industrial and control applications, including sensor networks. In another article in this section, a Freescale Semiconductor Inc. engineer describes how that low-power, low-cost network technology handles mesh (see page 43).
A rich handful of startups are targeting sensor nets for home, factory floor and military uses. Among them, UC Berkeley spin-out Dust Networks has a unique routing protocol that allows systems to synchronize with each other on a set time schedule. Basically, the nodes in a network talk with one another at set time intervals. If data isn't available to pass between the nodes, the system turns off, saving power. And if information needs to be transmitted, the transmission occurs on the predetermined clock cycle.
- Robert Keenan contributed to this report.