AutoCell The Self-Organizing WLAN By definition, IEEE 802.11 wireless LANS (WLANs) are constantly in flux. There is no way to predict where a particular client will be at any moment, making it equally impossible to predetermine the load on any wireless access point (AP). How will 802.11 Wi-Fi networks running in adjoining offices or on different floors affect one another? Even something as seemingly trivial as rearranging cubicle partitions or people standing in front of an AP can affect network conditions. Propagate Networks, a startup based in Acton, Mass., has taken on the toughest questions facing WLANs. Its answer: AutoCell TM, an embedded control system for 802.11 WLANs. Makers of wireless LAN switches, APs, and clients simply embed AutoCell in their products, right alongside their chip sets. As a result, their customers should see available bandwidth increase by as much as twenty fold. At the same time, WLAN installation and operating costs the largest expenses related to 802.11 networks should drop by 70 percent. Propagate delivers on these numbers by providing the industry's first selforganizing wireless LAN. AutoCell is based on swarm logic, a natural phenomenon that establishes collective intelligence and decentralizes decisionmaking, rather than deferring to a single, higher authority. It explains why a colony of ants is smarter than any individual member. It also enables simple components to behave collectively with higher-level intelligence. AutoCell is completely automatic: It is a continuous communication system that relies on a lightweight protocol to monitor changes on the wireless domain while keeping overhead very low. Among AutoCell's inherent advantages: Elimination of manual site surveys and channel maps Dynamic load balancing Plug-and-play-implementation Transparent fault recovery and failover Since AutoCell is completely self-organizing, it holds human intervention to a minimum. That reduces the people costs associated with deployment,
management, and maintenance making 802.11 WLANs practical, efficient, and cost-effective. Breaking the Site Survey Cycle Up until now, site surveys and channel maps have been used to manage and maintain 802.11 networks. For any WLAN built with more than a single AP, however, surveys and maps create more problems than they solve. Why? Channel Maps Are Theoretical Consider a company that has decided to deploy an 802.11 Wi-Fi LAN throughout a 10-story building. The process would start with a radio frequency (RF) site survey of every floor. The results would be used to create an initial channel map, a theoretical document that can be considered a best guess of floor-by-floor RF assignments and AP placement. During deployment, however, the WLAN would be tweaked and tuned, necessitating another channel map. That effort would be followed by yet another site survey and, possibly, another channel map. At some moment, the WLAN configuration would be frozen and a channel map drawn to reflect that state. But as soon as there were changes required additional APs added the channel map would be out of date. Channel Maps Are Static A channel map is a static snapshot of a constantly shifting environment. How quickly changes will occur is unknowable, but there is good reason to believe they will be frequent. WLAN gear is cheap: A radio access point now sells for $200 or less, with prices expected to fall further. That economy encourages users to add equipment. Often. Every time they do, they render the existing site survey and related channel map obsolete kicking off yet another survey cycle. Channel Maps Are Rarely Multidimensional Conventional wired networks lend themselves to linear thinking. The understandable tendency is to visualize packets traveling along a wire or through a switch. In contrast, 802.11 networks are multidimensional an inherent characteristic of radio transmissions. This means IT professionals deploying WLANs must think in at least three dimensions to account for interference between adjoining floors or neighboring offices. Unfortunately, interference issues are likely to be discovered only after a WLAN is in use. Addressing the problem means adjusting cell size and transmit power, which then requires a new manual site survey and channel map. To borrow another image from the wired world, performing a site survey and redrawing a channel map every time there is a change on a WLAN is equivalent 2
to reconfiguring the entire network every time a switch is added to a wiring closet or a port enabled. It wastes time and money. It misapplies technology. And it rapidly results in a never-ending spiral. As devices proliferate the problem of managing the RF domain will become acute, acting as a chokepoint to further growth. According to IC Insights, 802.11- enabled equipment will exhibit a compound annual growth rate of 50 percent between 2002 and 2006, reaching roughly 100 million units. Radios are being added to far more than laptops, access points, and WLAN switches. They also are going into IP phones, personal digital assistants (PDAs), tablet computers, and TiVo digital video recorders (DVRs). If wireless is being added to everything, does everything need to be tuned? Clearly, that is impossible without an automated system like AutoCell. Tightly Coupled: Channel Selection and Transmit Power As noted, AutoCell automatically self-organizes a wireless network. Essentially, its algorithm ensures that radio access points transmitting at the same frequency are located as far from one another as possible. It also reduces the transmit power of each cell to minimize or eliminate collision domains, where traffic from APs operating at the same frequency contend for bandwidth, thus degrading WLAN performance. But channel selection and transmit power are highly interdependent. Powering down each cell boosts the number of channels that can be re-used. When transmission power is reduced, cells are smaller thus allowing them to be packed more densely into a given area without risking interference or creating collision domains. By way of analogy, consider a hypothetical game involving colored balls and a large bin. The object is to pack the balls into the bin while keeping balls of the same color as far from one another as possible. To increase the challenge, the balls can change size some shrinking while others grow larger (see Figure 1). Figure 1: The WLAN Challenge 3
Now imagine that the bin is a 10-story building. The colored balls are RF domains. To maintain the parallel, only eight different-colored balls can be used, the number of channels defined by 802.11a. In the case of 802.11g, the number of available channels is only three. Basically, AutoCell not only keeps same-frequency domains as far from one another as possible but also adjusts their size increasing or decreasing their dimensions as necessary (see Figure 2). Figure 2: Coverage Without Contention Closing the Loop on Wireless LANs So how does AutoCell do it? The answer lies in the difference between a closed-loop and an open-loop system. AutoCell is the former. Like all closed-loop systems it employs a feedback mechanism, relying on data supplied by the system itself (the WLAN, in this case) to carry out the appropriate actions. For AutoCell, the feedback mechanism takes the form of the constant communications and exchanges among APs and clients. These are used to respond to changes in the RF environment -no matter how subtle in real time. An open-loop system, in contrast, has no such feedback mechanism. It can be tuned, but such adjustments are based on theoretical considerations not on what is actually happening on the wireless LAN. Again, it is easy to see why site surveys and channel maps are ineffective. They bear little or no relationship to conditions on the WLAN. One of the advantages of AutoCell s self-organizing (closed-loop) approach is that it enables APs to be deployed as needed rather than carefully rationed to prevent collision domains. The more bandwidth required for a particular cell, the more APs that can be installed. The only network design rule for any given cell: If more APs are needed, use them (see Figure 3). 4
Figure 3: Lifting the Limit on APs The same lack of restrictions applies to clients as long as they are equipped with AutoCell. Since Propagate's product is continuously communicating with its peers, when the number of clients employing a particular channel starts to threaten performance, some clients are simply shifted over to another RF channel. Since AutoCell tunes transmit power and channel selection in response to realworld conditions, cell size can be optimized. When WLAN cells are smaller, each user gets a proportionally larger share of available capacity. In fact, capacity can be as much as 20 times higher than on a conventional LAN, ensuring that users get the maximum available bandwidth. There is no way to achieve the equivalent performance when WLANs are tuned manually, even with well-designed network management tools. Decreasing transmit power (thus shrinking cell diameter) increases the number of reusable channels (smaller cells are by definition farther apart). Since transmit power and channel selection are so tightly linked, adding more channels means that transmit power has to be readjusted. That in turn requires channel selection to be re-evaluated. This cycle repeats itself until trial and error, or luck, arrives at something approximating an optimal configuration. And that configuration remains optimal only as long as no changes occur on the WLAN, a short-lived situation at best. 5
A Good-Neighbor Policy for WLANs As mentioned, Wi-Fi LANs are multidimensional. But when companies in the same building are running 802.11 configurations, they may not realize the effect their wireless equipment is having on WLANs in nearby offices or on adjoining floors. That becomes a particular problem when APs and clients are running at full power. Interference and contention are common occurrences, playing havoc with performance. When users complain, the IT administrator blames the vendor's WLAN products or immature technology. At the same time, those overpowered RF channels are leaking through to the WLAN installation next door and on the floor above with predictable results. A similar situation occurs in retail businesses that have no on-site IT expertise. A good example is a coffee shop that offers wireless access to its customers. What happens if it is located next to an airport "hot spot" or a first-class passenger lounge? Which one turns down the transmit power of its APs? Which provides the site survey and channel selection? AutoCell does not simply offer a solution to these problems. It eliminates them. The same self-organizing capabilities that balance power and RF channels can be used to prevent RF interference. And AutoCell takes care of these and other problems automatically, so there's no wasted effort or excessive operating expenses involved in fine-tuning a WLAN. Balancing the Load on the Airlink One of AutoCell s key capabilities, automatic load balancing, allows it to respond equally quickly and efficiently to traffic spikes, power failures, and other unforeseen events. For example, when users are clustered around a single node, available capacity on nearby nodes typically goes unused. AutoCell automatically shifts users onto less heavily loaded APs, ensuring that everyone achieves maximum performance. Similarly, fault handling and failover are natural properties of a self-organizing WLAN. If an AP goes down, for example, AutoCell automatically changes channels or increases transmit power to fill in the coverage gap. On a conventional 802.11 LAN, the same occurrence would likely create a coverage hole, leaving some clients without service. As WLAN users roam, they typically cross cell boundaries. Conventional WLAN software usually does not re-associate clients with another AP until the data rate falls precipitously low or the connection is dropped. There is a historical reason for this so-called sticky roaming: In the early days of WLANs, APs were located as far apart as possible; the goal for clients, meanwhile, was to be able to travel as far from an AP as possible, even if throughput degraded dramatically. 6
Since cells in self-organized WLANs can be smaller and packed far more densely in a given area, AutoCell's roaming algorithm is optimized to maintain high bandwidth rather than to keep a user associated with an AP for as long as possible. In fact, roaming decisions on a self-organized WLAN can be made whenever a higher-bandwidth connection opens up. By rapidly detecting changes and sharing information about available capacity, clients in a selforganizing WLAN can jump from cell to cell without any drop in rate. Recalculating the Costs of Doing Business Prices for 802.11 WLAN equipment are only going to drop. The real expense of building and operating a conventional wireless LAN goes to installation and operation. And the majority of that money goes to people costs rather than technology investments futile fine-tuning, additional site surveys, and endless channel mapping. Virtually all of this money is wasted. Site surveys and channel maps may give an illusion of control, but that is all it is. With AutoCell-enabled APs, switches, and clients, nearly all of the foregoing expenses are eliminated. AutoCell s self-organizing structure keeps data rates high, dynamically responds to rapidly changing conditions, and transparently fails over if there are problems. AutoCell does not force highly compensated IT professionals to waste time conducting useless site surveys and plotting out channel maps that will be obsolete almost as soon as they are complete. The result: AutoCell allows IT organizations and users to put 802.11 APs anywhere they want, and potentially saves as much as 70 percent of the implementation costs of conventional WLANs. 2003 Propagate Networks, Inc. All rights reserved. 7