The 802.11 wireless-local-area-network (WLAN) market continues to enjoy exceptional growth. According to IC Insights (www.icinsights.com), this year's 802.11 silicon shipments are projected to reach 35 million chips. That number translates into 80% growth for 2003 alone. As this trend continues, radio density will dramatically increase in multiple environments: enterprises, public hot spots, and—eventually—home-entertainment systems.
This radio-density explosion will be a mixed blessing, however. Today, every cell in an 802.11 WLAN is actually a shared network equivalent to the early days of wired-Ethernet LANs. In addition, factor in the limited number of radio-frequency (RF) channels combined with large coverage zones. These elements limit the number of shared networks that can be deployed in any given space.
Some people think that these problems could be solved with better network management or more sophisticated site surveys. Other individuals believe that these issues can be mitigated with tools that manually measure and configure the RF environment. But these proposed solutions are work intensive and expensive. Plus, their static factors make them problematic as the number of WLAN users increases.
Once they become operational, 802.11 WLANs constantly change. As more end stations are added, WLANs must rearrange themselves or expand for better performance and routine fault management. As the network grows in size and complexity, each of these tasks becomes geometrically more difficult. The necessity to automate some or all of the daily-operations management becomes increasingly important.
Remember that the environment is subject to constant change. People and devices move, furniture and partitions are re-arranged, or a variety of bandwidth-demanding applications is added. The image of a team of WLAN managers following users around in order to readjust and fine tune network operation is logistically unsound and economically doomed.
The ideal solution is a completely automated, self-organizing WLAN system. It should monitor, configure, and optimize the RF domain wherever and whenever it's needed. To be truly automated, however, the RF control system must address certain requirements. Installation, for example, must be simple plug and play. All changes must be self compensating.
In addition, the load must be dynamically and optimally distributed across available bandwidth. When it's possible, roaming must be at the maximum data rate. Fault handling and failover have to be automatic and predictable. Lastly, the network's adaptation behavior must always be stable and deterministic—not unpredictable and oscillating.
The control system should be continuously running so that it rapidly adapts to changes without user intervention. To allow for installation in a variety of devices (clients, access points, and switches), the control system must be lightweight and computationally efficient. It also must be able to work across multiple network architectures.
When it comes to purchasing an 802.11 network, many users find that the cost of installation is the number one stumbling block. Compared to the cost of purchasing the WLAN hardware, it typically costs more to survey the site; develop an engineering plan; hire consultants; simulate the environment; install the equipment and software; and follow up with tuning. After the initial installation is complete, thousands of dollars may still be needed to manually reoptimize placement and reconfigure channel maps.
In future environments, such as home entertainment, planning won't be an option at any level. Consumers won't tolerate anything less than being able to plug in devices wherever they want them. This means that the networks must eliminate configuration limitations so 802.11-based products can be installed without any special skills. For these goals to be reached, a continuously running automatic system must gain control over several elements. For instance, the self-organizing WLAN network must control frequency/channel selection and transmit power control.
A self-organizing network should constantly evaluate and recompute optimal channel allocations for the entire network. It doesn't matter if this evaluation is done at initial startup or on an ongoing basis. The vital point is that this allocation should be handled automatically. Otherwise, the WLAN will need reconfiguration every time a change is made to either the network settings or the RF environment.
Upon closer inspection, this problem reveals other dark sides as well. The elements of channel selection and transmit power control are highly interdependent. If both elements aren't simultaneously coordinated, the physical placement of the APs is restricted. For example, say the transmit power isn't controlled. By default, all of the cells would operate at full power. The APs would then have to be spaced very far apart to avoid co-channel interference. If they're placed far enough apart, it's safe to say that two APs—on the same frequency—won't create overlapping coverage zones.
If two APs on the same channel do have overlapping coverage zones, they create a large shared collision domain. This domain diminishes bandwidth (FIG. 1). For the users within it, traffic from either cell adversely impacts the performance of the conjoined overlapping cells. Essentially, access points on the same frequency are competing with one another to serve the users in the shared-collision domain.
This condition of conjoined overlapping cells occurs in WLAN installations that have reached "ChannelMax." This state is reached when all of the available non-overlapping channels have been allocated in a given space. In a typical 200-by-200-ft. office space with 802.11-based APs supporting eight channels, ChannelMax is reached when only eight APs are installed (FIG. 2). In this example, adding more APs beyond ChannelMax doesn't increase available bandwidth. Instead, it creates more shared collision domains. The available bandwidth is therefore reduced.
When ChannelMax is reached, a well-designed self-organizing-network algorithm chooses channels such that the distance between cells on the same channel is maximized. The algorithm also continues to optimize the network by reducing power in these cells. It minimizes or completely eliminates shared collision domains while maximizing overall network performance (FIG. 3). By coordinating the control of transmit power and channel selection, coverage zones can even be placed closer together without incurring interference.
In the resulting installation dynamic, arbitrary configurations can be the norm. The WLAN network now self organizes. As a result, clusters of APs can be installed without any consideration for special design rules, additional site surveys, and density (FIG. 4).
If more bandwidth is required for a given zone, more APs can be installed. This possibility exists because the network dynamically self constructs. The network-design rule now becomes simple: Deploy APs in densities that are proportional to the bandwidth requirement in a given zone. If that's not good enough, just add more APs. In all cases, a self-organizing network allows an 802.11 WLAN to break through the ChannelMax barrier.
A number of IT managers think that they can control transmit power and frequency selection manually. Even if they have a relatively well-designed network-management tool, they're sorely mistaken. Changing transmit power actually increases the number of channels that can be reused. But when the cells shrink, the number of frequencies that can be squeezed into the space increases. If the channels have been changed, the power also can be changed. Plus, the frequencies or channels that previously interfered may not be overlapping anymore. In other words, the result of turning one knob (power or frequency) causes the IT manager to go back and readjust the other knob (power or frequency).
Without an automatic system that calculates the correct interdependent settings, the manual trial-and-error process can go on forever. By using the computers that are already embedded in its network devices, a self-organizing WLAN can calculate the optimal configuration. It doesn't have to rely on the trial-and-error process faced by the IT manager.
The amount of load on a network is rarely distributed evenly across a given area. To anticipate current user clusters, a self-organizing WLAN allows for the higher-density placement of APs. But the users' needs may change over time. WLAN users are mobile by definition. Without automatic load balancing, many APs may go unused—even though they have available capacity. For example, think of users who may be clustered around a single AP in conventions, conference rooms, classrooms, or dense office environments. In these situations, there may be adjacent APs nearby with virtually no load. User loads should be automatically distributed.
It isn't just the high-density situations that need load balancing, however. Users also may cluster in break areas or other unpredictable spaces. With square footage at a premium in many offices, break areas can even turn into work areas. A network that cannot respond to a location-related traffic spike may provide inadequate bandwidth on some APs. Using a heuristic approach, a self-organizing network distributes clients (phones, laptops, and PDAs) across available APs. This method combines range, data rate, and AP load.
In a self-organizing WLAN with densely deployed cells, roaming should be designed to maximize the bandwidth per user. It shouldn't remain on a potentially inadequate AP. In a self-organizing WLAN, clients can roam whenever a higher-performance connection is available. Those clients can even maintain high data rates across roaming events. To retain such high data rates, clients and APs must be proactive in their understanding of the nearby environment's available bandwidth. By rapidly detecting motion and sharing information about that bandwidth, clients in a self-organizing WLAN can jump from cell to cell without any drop in rate.
The software's decision to roam—or re-associate to another access point—is usually based on one of two triggering events. Either the data rate drops significantly or connectivity is lost altogether. This roaming algorithm is a remnant of the early days of WLAN deployments, when cells were widely dispersed and needed to be large. Typically, the end users that moved across those large cells experienced different data rates as they moved further away from the AP. While the data rates were dropping, roaming algorithms needed to maintain connectivity. A self-organizing WLAN produces higher-density overlapping cells on different channels. Now, it's possible to implement roaming at the full rate of 54 Mbps.
When an AP fails, however, it's likely that a coverage hole will be created. This situation may leave some clients without service. Fault handling and failover are natural properties of a self-organizing WLAN. As previously discussed, self-organizing APs are constantly optimizing their choices of power and channel selection. When an AP fails in the neighborhood, self-organizing APs will automatically change channels or increase transmit power to fill in the coverage gap. Meanwhile, clients will naturally roam to the best AP.
A WELL-BALANCED SYSTEM
Note that there is a fine line between an adapting system and an unstable one. If a system is constantly changing without reason, it's inherently unstable. The best example is a system that only listens to itself and pays no attention to outside influences. The opposite extreme is a system that doesn't react to any outside change or influence.
In designing automatic control systems—especially in large configurations—stability is paramount. To avoid the problem of oscillations and over-compensations, numerous feedback loops must be implemented in self-organizing WLANs. These loops measure various system parameters. They're then used as inputs to a holistic algorithm that makes control decisions. To ensure stability, a properly designed self-organizing algorithm includes dampening and hysteresis in its control loops.
Without automatic, self-organizing WLANs, 802.11 deployments will surely falter. Consumers and IT organizations will quickly exceed their tolerance for the time and expertise that it takes to manage increasingly dense RF environments. For an equivalent wired scenario, imagine that installing another Ethernet switch in the wiring closet would mean reconfiguring the entire network. Thankfully, the WLAN market is ripe for the emergence of self-organizing, self-optimizing WLAN networks.