By Dr. Michael Nova
The growing popularity of metropolitan WiFi networks highlights the strengths and weaknesses of using a mesh architecture for WiFi backhaul. More broadband
click to enlarge
Figure 1. Comparison of multi-hop throughput.
Rather than relying on wireless hot spots, cities are starting to deploy metro WiFi to blanket the metropolis, providing WiFi coverage with continuous connectivity, similar to cellular phone service. However, the requirements for a metropolitan WiFi backhaul infrastructure are radically different than those of a conventional wireless office LAN. Wireless LANs are normally indoor, short-range infrastructures that deliver broadband using centralized wireless switch points instead of network cable. When you scale a wireless network to service a metropolitan area, the challenge becomes supporting a backhaul with sufficient bandwidth. Interconnecting outdoor access points (APs) normally becomes the most expensive part of deploying any metro-scale WLAN, which is why most metro WLANs use a wireless mesh approach to share the broadband load among nodes.
Factors that Dictate Mesh Performance
Wireless mesh allows APs to connect to other APs within their range to relay packets to the next access point to form the backhaul. This means a service provider can cover a wide area with no cost for each interconnection. Since the backhaul is wireless, additional access points can be added to improve performance as customer demand and traffic increase; backhaul and user traffic compete in a mesh environment so adding more mesh nodes increases overall bandwidth. However, the performance and scalability of any metro WiFi fixed backhaul network is limited by two key factors:
1) Transceiver range
2) Congestion control.
Transceiver range establishes the maximum distance between nodes. Individual nodes can function as either gateway APs that provide wide area network (WAN) backhaul access or as non-gateway mesh routers that act as APs to local client devices but route to a gateway for WAN access. Both environmental conditions (topology, buildings, trees, node placement, sources of interference, etc.) and the type of radio hardware used (antenna type, radio power, etc.) determine the effective distance between nodes.
Congestion control ensures that no single AP becomes overloaded with data traffic. Congestion is a function of the number of simultaneous users and the type of data
click to enlarge
Figure 2. Multi-client performance over a single hop.
A New Approach to WiFi Data Traffic
The real challenge of deploying a wireless mesh is that the same spectrum has to be used for local data access as well as the backhaul. Sharing the spectrum can reduce overall capacity by as much as 50%. In unconstrained mesh architecture, traffic is routed based on higher level protocols and when a disruption in service occurs, the mesh finds alternative data paths. While this approach is very resilient, it also can consume available bandwidth, which affects performance, particularly for QoS applications.
There are ways to improve backhaul performance, such as using a point-to-point system with dedicated nodes for high-speed backhaul links, or deploying a constrained mesh that uses designated nodes to create a virtual point-to-point backhaul. Of course, controlling the backhaul path offers some control over network performance, but it is more expensive to install and does not completely allow for increases in overall user traffic.
Rather than installing more hardware, a better approach is to modify installed hardware with software that changes the way the MAC layer handles packets, adopting a
click to enlarge
Table 1. Simultaneous # Calls in 6 Hop Network with MOS ≤4.0 Background data traffic (UDP)
Use of KMT offers three benefits for a WiFi backhaul:
1) It increases the amount of usable bandwidth by improving use of the available 802.11x spectrum efficiency.
2) It provides dynamic routing to mitigate QoS issues, such as spatial bias (distance between APs) and preset data rates.
| Estimated Savings from
To demonstrate the potential cost savings realized from upgrading existing wireless mesh hardware with multi-channel TDMA routing software, consider the following:
Estimated network node density
• San Francisco estimates it needs 2000 to 3000 WiFi nodes to cover entire city (46.7 sq miles) or 43 to 64 nodes per square mile.
• Philadelphia’s network planned by EarthLink reported to average 37 nodes per square mile.
• Mountain View, California estimates it will need 300 nodes to cover a portion of the city (the covered area was not quantified but the city is 12 square mile) or about 25 nodes per sq mile.
• Assume average density is 40 nodes per square mile.
Estimated cost per node
• Node units manufactured by Tropos are priced at about $3,300 each.
• A two-man crew costs $1000 per day and can install 5 nodes per day or $200 per node.
• 1 gateway per 10 nodes or 10% cost 23 to install or $400 per gateway (assuming a gateway node is priced the same as non-gateway node).
Estimated capital cost savings per square mile with KMT
• Influenced by level of voice and video traffic
• Influenced by user capacity
• Influenced by number of hops to gateway node
• Sample savings calculation per square mile assuming 30% of nodes are installed to handle capacity concern rather than range limitation: (40 3 30% 3 $3300) + (40 3 30% 3 $200) = $42,000.
Estimated operating and maintenance cost per node with KMT
• Electricity cost per node estimated at 18 W operating 24 3 7 at $0.10/kilowatt hour.
• Service cost estimated at hour per node per month at $75/hr.
• Replacement cost estimated at 2.5% per year.
• Sample O&M savings calculation per node: (18 W 3 24 hours 3 365 days 3 $0.0001/W) + ( 1/4 hour/month 3 12 months 3 $75) + (1 node 3 2.5% 3 $3500) = $329
• Estimated 15 year project life cost per node assuming general inflation at 5% discounted to present value at 10%: $3,300.
Determining the number of nodes needed for congestion control versus transceiver range will depend on project-specific conditions. The ratio will also likely to change over time as consumer familiarity and municipal WiFi use increases, and more demanding multimedia applications become common. Estimated savings per square mile are shown in Table 2 over a range of deployment conditions.
Let’s review each in turn:
1. Improved Spectral Efficiency: Conventional WiFi uses one channel for all communications for all connections, but there are actually three non-overlapping channels available in the 2.4 GHz ISM band (802.11b/g) and many more for the 5 GHz (802.11a) ISM band. KMT makes all non-overlapping channels available for simultaneous use, which increases both the available bandwidth and the effective throughput by a factor of three or more. Consider that with conventional CSMA/CA, three separate WiFi devices would have to be connected to an AP using the same channel. With KMT, each device could have a dedicated channel.
2. Dynamic Routing: Logically, in a wireless mesh APs that are physically closer to gateway APs tend to have higher data rates than APs further away. To mitigate this spatial bias, most mesh hardware manufacturers statically limit the maximum data rate for users in order to reserve capacity for other nodes. Adopting dynamic routing eliminates the need for pre-set data rates for individual nodes. Instead, you can determine traffic conditions in real time and dynamically allocate bandwidth so any node can transmit at full capacity while other nodes are unused.
3. Cross Layer MAC Routing: Wireless mesh routing protocols typically operate at Layer 3 and route based on link quality, rather than node availability. In a standard mesh architecture, some APs may offer better link quality and so quickly become congested during high traffic conditions, even when there are other pathways available.
click to enlarge
Table 2. Per Square Mile Savings Potential through Use of KMT
Applying KMT in a Wireless Mesh
Adopting KMT routing extends both the number of hops available in a wireless mesh and increases the overall efficiency of available bandwidth. Let’s compare the performance of a WiFi AP using conventional CSMA/CA and KMT packet handling in the same device. Figure 1 shows a comparison of multi-hop throughput and Figure 2 shows multi-client performance over a single hop. Clearly, the KMT offers higher throughput and more stable performance, no matter what the number of hops or users/clients.
Improving multi-hop performance (as shown in Figure 1) is extremely important in mobile networks and indoor networks, but in municipal WiFi networks, it also relieves demand for expansion base capacity. By substituting KMT on either single radio or multi-radio platforms, the entire network becomes more scalable since additional nodes can be deployed to extend coverage without adding new backhaul gateways.
Figure 2 demonstrates how congestion control can increase throughput. Let’s assume that a typical municipal WiFi AP is designed to handle 200 users each day, with a diversity factor of 10% (which means only 10% or 20 users are expected to access that AP at any given time). As the number of TCP/IP users increases on the node, throughput declines, as shown by the red line. When the AP peaks with 20 simultaneous TCP/IP users, performance may be adequate for e-mail or web access, but running multimedia or voice may be difficult with so much background traffic.
Since traffic load is not evenly distributed across the network, some nodes are probably strategically placed to manage congestion control (as opposed to obviating transceiver range). Using KMT routing, bandwidth remains high, even as the number of data flows increases, because the TDMA scheme manages the data congestion by prioritizing traffic using multiple channels. And the TDMA software can improve performance for any hardware platform. The results in Figure 2 were achieved by testing on a consumer WiFi card (e.g., Linksys, D-Link AP) and show that with 12 users (six connections), bandwidth is approximately nine times greater using KMT. These values may differ with utility grade APs, but the relative improvement — a multi-fold increase in user capacity — should be comparable. Using KMT can significantly reduce the number of nodes deployed for congestion control.
Optimizing for QoS
The impact of poor QoS and low throughput in a WiFi mesh is a problem for time-sensitive applications, such as VoIP and streaming video, even when there are relatively few users accessing the network. If you have as little as 1 Mb/s of background traffic over a multi-hop connection with multiple users, it is impossible to handle QoS traffic such as VOIP (see Table 1). KMT improves performance for QoS traffic a minimum of three times, and often more. This means mobile users can access VoIP, IPTV, IP radio and other multimedia applications simultaneously over a municipal wireless mesh.
So by utilizing Kiyon’s 802.11 compliant multi-channel TDMA routing at the 802.11 MAC layer, a metro WiFi backhaul can be optimized to handle more data traffic and more types of traffic, from e-mail and HTML right up through high-definition video. It’s a matter of taking advantage of all the data channels available, providing dynamic routing and implementing better QoS for time-sensitive applications.
About the Author
Dr. Michael Nova is founder, chairman and CEO of Kiyon, For additional information, visit www.kiyon.com