White Paper Overview This paper focuses on wireless mesh infrastructure systems used for creating large Wi-Fi access networks, and examines three different approaches currently available for implementing them. It examines the strengths and weaknesses of each approach with a particular focus on the capacity that is available to users. Can wireless mesh infrastructure systems deliver enough capacity to support broadband services for a large number of users? Mesh is a type of network architecture. Originally, Ethernet was a shared bus topology in which every node tapped into a common cable that carried all transmissions from all nodes. In bus networks, any node on the network hears all transmissions from every other node in the network. Most local area networks (LANs) today use a star topology in which every network node is connected to a switch (switches can be interconnected to form larger networks). Mesh networks are different – full physical layer connectivity is not required.As long as a node is connected to at least one other node in a mesh network, it will have full connectivity to the entire network because each mesh node forwards packets to other nodes in the network as required. Mesh protocols automatically determine the best route through the network and can dynamically reconfigure the network if a link becomes unusable. There are many different types of mesh networks. Mesh networks can be wired or wireless. For wireless networks there are ad-hoc mobile mesh networks and permanent infrastructure mesh networks.There are single radio mesh networks, dual-radio mesh networks and multi-radio mesh networks.All of these approaches have their strengths and weaknesses.They can be targeted at different applications and used to address different stages in the evolution and growth of the network. Capacity of Wireless Mesh Networks Understanding Single Radio, Dual Radio and Multi-Radio Wireless Mesh Networks
Transcript
White Paper
Overview
This paper focuses on wireless mesh infrastructure systems used for creating large
Wi-Fi access networks, and examines three different approaches currently available for
implementing them. It examines the strengths and weaknesses of each approach with a
particular focus on the capacity that is available to users. Can wireless mesh infrastructure
systems deliver enough capacity to support broadband services for a large number of users?
Mesh is a type of network architecture. Originally, Ethernet was a shared bus topology
in which every node tapped into a common cable that carried all transmissions from all
nodes. In bus networks, any node on the network hears all transmissions from every other
node in the network. Most local area networks (LANs) today use a star topology in which
every network node is connected to a switch (switches can be interconnected to form
larger networks).
Mesh networks are different – full physical layer connectivity is not required.As long as a
node is connected to at least one other node in a mesh network, it will have full connectivity
to the entire network because each mesh node forwards packets to other nodes in the
network as required. Mesh protocols automatically determine the best route through the
network and can dynamically reconfigure the network if a link becomes unusable.
There are many different types of mesh networks. Mesh networks can be wired or
wireless. For wireless networks there are ad-hoc mobile mesh networks and permanent
infrastructure mesh networks.There are single radio mesh networks, dual-radio mesh
networks and multi-radio mesh networks.All of these approaches have their strengths
and weaknesses.They can be targeted at different applications and used to address
different stages in the evolution and growth of the network.
Capacity of Wireless Mesh Networks Understanding Single Radio, Dual Radio and Multi-Radio Wireless Mesh Networks
The first wireless mesh networks were mobile ad hoc networks – with wireless stations moving aroundand participating in a peer to peer network. Mesh is an attractive approach for wireless networking sincewireless nodes may be mobile and it is common for a wireless node to participate in a network withoutbeing able to hear all of the other nodes in the network. Mobile peer to peer networks benefit from thesparse connectivity requirements of the mesh architecture; and the combination of wireless and meshcan provide a reliable network with a great deal of flexibility.
The popularity of Wi-Fi has generated a lot of interest in developing wireless networks that support Wi-Fi access across very large areas. Large coverage access points (AP) are available for these scenarios,but the cost of deploying these wide area Wi-Fi systems is dominated by the cost of the networkrequired to interconnect the APs and connect them to the Internet— the backhaul network.
Even with fewer APs, it is very expensive to provide T1, DSL or Ethernet backhaul for each access point. For these deployments, wireless backhaul is an attractive alternative and a good application formesh networking.Wireless connections can be used between most of the APs and just a few wiredconnections back to the Internet are required to support the entire network.
Wireless links work better when there is clear line of sight between the communicating stations.Permanent wireless infrastructure mesh systems deployed over large areas can use the forwardingcapabilities of the mesh architecture to go around physical obstacles such as buildings. Rather thanblasting through a building with high power, a wireless mesh system will forward packets throughintermediate nodes that are within line of sight and go around the obstruction with robust wireless links operating at much lower power.This approach works very well in dense urban areas with many obstructions.
There are many different types of mesh systems and they often get lumped together. Since early
wireless mesh systems were focused on mobile ad-hoc networks, many people assume that wireless
mesh systems are low bandwidth or temporary systems that can not scale up to deliver the capacity
and quality of service required for enterprise, service provider and public safety networks.That is not
the case. Engineered, planned and deployed effectively, wireless mesh networks can scale very well while
still offering a cost-effective evolution strategy that preserves the network investment. Understanding
the strengths and weaknesses of single, dual, and multi-radio mesh options is the first step.
Single-radio Wireless Mesh In a single-radio mesh, each mesh node acts as an AP that supports local Wi-Fi client access and forwards traffic wirelessly to other mesh nodes.The same radio is used for access and wireless backhaul.This option represents the lowest cost entry point in the deployment of a wireless mesh networkinfrastructure. However, because each mesh AP uses an omni-directional antenna to allow it tocommunicate with any of its neighbor APs, almost every packet generated by local clients must berepeated on the same channel to send it to at least one neighboring mesh AP.The packet is thenforwarded to another node in the mesh and ultimately to a node that is connected to a wired network.
This packet forwarding generates a lot of traffic.As more mesh APs are added, a higher percentage of the wireless traffic in any cell is dedicated to forwarding.Very little of the channel capacity is availableto support users.
There is debate in the industry about the impact of mesh forwarding and actual throughput that ispossible in this scenario.The capacity analysis is somewhere between 1/N times the channel capacity and (1/2)^N times the channel capacity where N is the number of wireless hops in the longest pathbetween a client and the wired infrastructure.
Figure 2 shows AP capacity estimates for a single-radio Wi-Fi mesh network using these equations. Usercapacity available at each AP declines as you add more APs to the network and increase the number ofwireless hops.The starting capacity is 5 Mbps because the network is a single channel of 802.11b, whichhas a raw data rate of 11 Mbps and useful throughput measured at the TCP/IP layer of about 5 Mbps.This throughput is shared between the access traffic and the backhaul traffic in a single radio mesh.
Throughout this paper, the vertical axis can be scaled to reflect the radio capacity. Some good rules ofthumb are; 5Mbps for 802.11b only mode, 11Mbps for 802.11b/g mixed mode and 22Mbps for 802.11gonly mode.The latter is not typically deployed in public environments due to backwards compatibilitywith the large pool of 802.11b devices.
It doesn’t really matter which of the equations is closer to real world behavior. 1/N is more optimistic,but neither scales to support large networks. Capacity available in each cell declines rapidly as more APs are added.
There are mesh protocols that optimize the forwarding behavior and eliminate unnecessarytransmissions. But the best these optimizations can do is to bring the network closer to 1/Nperformance, which is inadequate for most permanent infrastructure applications today. Single-radio mesh systems will not deliver broadband performance to the user population throughout a very large coverage area.
This analysis may seem harsh, but it is actually oversimplified. It assumes perfect mesh forwarding, nointerference and perfect coordination of the Wi-Fi channel access.That will never happen, so real worldthroughput and capacity will usually be even lower.
Figure 3: Single-radio Mesh Architecture, String of Mesh APs
To illustrate this point, consider a linear string of mesh APs arranged so that each one can hear only oneadjacent neighbor on either side (Figure 3).This is not a likely real world deployment, but it simplifies theanalysis and we will use this example to compare each of the wireless infrastructure mesh approaches.Throughout this paper we will also assume that client access load is evenly distributed across the meshAPs. In this string of APs with the wired connection on the end, N the number of hops from figure 4, issame as the number of mesh APs.
The total channel capacity is 5 Mbps.You can see that 1/N performance is basically not achievable. N=5,so each AP should have 1 Mbps of capacity.All of the traffic from the entire mesh network will have toflow through AP5 to get to the wired network. If each mesh AP accepts a load of exactly 1 Mbps oftraffic from its clients, then AP5 will have to forward 4 Mbps of traffic from APs 1, 2, 3 and 4; and hasexactly 1 Mbps of capacity left for its local clients. For this to work, there would have to be perfectcontention, interference and collision management. The mesh APs would have to coordinate theirtransmissions with each other and perfectly control the transmissions of all their respective clients.That is not how Wi-Fi works.
In a single-radio Wi-Fi mesh network, all clients and mesh APs must operate on the same channel anduse the 802.11 Media Access Control protocol. As a result, the entire mesh ends up acting like a single,giant access point—all of the mesh APs and all of the clients must contend for a single channel.Thisshared network contention and interference reduces capacity further and introduces unpredictabledelays in the system as forwarded packets from mesh APs and new packets from clients contend for the same channel.
The configuration in Figure 3 has minimal connectivity required to complete the mesh and minimuminteraction between adjacent APs for a 5-node mesh AP network.APs 2, 3, and 4 can hear two otherAPs; and AP1 and AP5 hear one other AP each. Each time AP3 transmits,AP2 and AP4 must defer andhold off their transmissions since they are using the 802.11 MAC protocol, which is essentially “listenbefore talk”.Whenever that hold-off doesn’t happen, collisions and retransmissions occur resulting inmore congestion and lower capacity.
A capacity analysis of these systems should include both the effects of the mesh forwarding and theeffects of the shared network backhaul, which can be significant.
Consider the string of Mesh APs in Figure 3. If we move the wired backhaul from AP5 to AP3, whathappens to the capacity?
N, the number of forwarding hops, is reduced from 5 to 3, so we might expect the capacity to be higherthan the N=5 capacity shown in Figure 4. However, due to the shared network behavior and the factthat AP3 can hear more mesh AP neighbors than AP5, the capacity is actually lower as shown in Figure 4.(Note:The x axis in Figure 4 is the number of Mesh APs, not the number of wireless hops in the longestpath through the mesh.)
Figure 4: Single-radio Mesh String,Wired Connection in the Middle
The 1/N equation we used earlier predicts that per-AP capacity will be 1.67 Mbps when N=3. However,when we factor in the effects of contention and interference when the wired connection is in the middleof a string of 5 APs (Figure 3 with the wired connection at AP3), the estimated capacity is .58 Mbps.Thismatches the (1/2)N prediction of .56 Mbps when N = 5.
The string of mesh APs that we have described so far is not a typical mesh configuration.The cluster ofmesh APs shown in Figure 5 is a more common example of a small mesh network.
In this case, contention and interference would reduce the capacity available for client access beyondwhat we have described in the string of APs examples previously discussed. Large coverage mesh APs in these systems have high power radios and high gain antennas.The mesh APs can hear each other at a much greater range than they can hear the clients they support, because most Wi-Fi client devices are low power with low gain antennas.
In this cluster,AP3 can hear all the other APs except for AP5.All traffic for the entire mesh networkflows through AP3 so it will frequently hold off the other APs, limiting their ability to handle traffic fromtheir local clients.A more complicated formula is required to characterize the impact of neighboringmesh APs in a shared backhaul network as well as the mesh forwarding.
The capacity in a single-radio mesh is limited by both access and backhaul issues. Optimizing the meshforwarding protocol will not solve the problem.The basic capacity is too low and adding more meshnodes makes it worse—no matter how perfect the mesh protocol.
Single-radio solutions offer the lowest cost entry point in the deployment of mesh networks. In aninfrastructure network, single radio mesh systems are best used for small mesh clusters of a few nodes.Larger systems may be created by providing wired backhaul to one of the nodes in each cluster or usingwireless backhaul links to aggregate multiple clusters. Single radio mesh solutions can also be the rightapproach for mobile, ad hoc peer-to-peer wireless networks where the emphasis is on basic connectivityor used for large sensor network and meter reading networks where the data rate is very low.
Dual-Radio Wireless Mesh The capacity and scaling ability of wireless mesh infrastructure networks can be improved by using APs that have separate radios for client access and wireless backhaul.
In a dual-radio mesh, the APs have two radios operating on different frequencies. One radio is used forclient access and the other radio provides wireless backhaul.The radios operate in different frequencybands, so they can run in parallel with no interference.A typical configuration is 2.4 GHz Wi-Fi for localaccess and some flavor of 5 GHz wireless for backhaul. Since the mesh interconnection is performed by a separate radio operating on a different channel, local wireless access is not affected by meshforwarding and can run at full speed.
However, in a dual radio mesh the wireless backhaul is a shared network so it is subject to the samenetwork contention issues that hamper the single radio mesh.The contention on the backhaul networklimits capacity and creates additional latency making the dual radio approach inappropriate for voice traffic.
The backhaul mesh in dual-radio mesh architectures is usually a shared network running the 802.11MAC protocol.With only one radio dedicated to backhaul at each node, all of the mesh APs must usethe same channel for connectivity to the backhaul mesh. Parallel operation on the backhaul network isnot possible, as most of the APs hear multiple other mesh APs. So they must contend for the channeland at the same time generate interference for each other.The result is reduced system capacity as the network grows.
As noted earlier, the useful capacity of each Wi-Fi access coverage cell is 5 Mbps. In dual-radio meshsystems, the access radios of adjacent cells can use different channels.There are three non-overlappingchannels in the 2.4 GHz band, so they will be able to operate independently in most cases (Figure 6).
Figure 6: Dual-Radio Wireless Mesh, String of Mesh APs
The most likely shared backhaul network protocol is 802.11a, which has a raw date rate of 54 Mbps and useful throughput of approximately 20 Mbps in this type of network.
Capacity is limited because of the behavior of the shared network used for the backhaul. Contention and interference vary depending on the placement of the APs.All of the APs must operate on the samechannel for the wireless backhaul and they must be able to hear at least one other AP in order to bepart of the mesh.
Typically, each AP will be able to hear at least two or three other APs.Those with more adjacentneighbors will have more contention and generate more interference than isolated mesh APs at the edge of the network. It is difficult to predict the system capacity without making assumptions about AP placement.
Figure 7 compares the capacity for the minimal overlap string of mesh APs shown in Figure 6.Thebackhaul network offers 20 Mbps of capacity, so per-AP capacity is good for a few nodes.After three orfour nodes, the per-AP capacity drops off because of the shared network effects. In a more typical meshcluster with more overlap between mesh APs, useful access capacity could be worse than shown here.
Figure 7: Per-AP Capacity in a Dual-Radio Wireless Mesh System,Wire at End.
Dual-radio systems are a big improvement over single-radio mesh designs and represent a logicalevolution in the growth of a mesh network. However, dual-radio systems alone don’t scale to metrodimensions and the high and unpredictable latency on the shared backhaul network makes them a poor candidate for voice over Wi-Fi.
Multi-Radio Wireless Mesh Like a dual-radio wireless mesh, a multi-radio wireless mesh separates access and backhaul.It goes a step further, however, to provide increased capacity by addressing the shared backhaulnetwork issues that limit the dual-radio mesh architecture.
In a multi-radio wireless mesh, multiple radios in each mesh node are dedicated to backhaul.The backhaul mesh is no longer a shared network, since it is built from multiple point-to-pointwireless links and each of the backhaul links operates on different independent channels.
This type of multi-radio design is called a multiple point-to-point mesh. It is possible to createvery rich mesh topologies with this multi-radio approach and just a few backhaul radios at eachnode. (Figure 8)
Figure 8: Multi-Radio Wireless Mesh, String of APs
When used for backhaul in this fashion, the performance of a multi-radio mesh is similar to switched,wired connections.The mesh radios operate independently on different channels so latency is very low.There are only two nodes per mesh link, so contention is very low. In fact, it is possible to run acustomized point to point protocol that optimizes throughput in this simple two-node contention-freeenvironment.These dedicated point to point links are usually in the unlicensed 5.8 GHz band and basedon 802.11a chipsets today. In the near future this will be a good application for 802.16d WiMAX.Thesepre-WiMAX wireless links have a potential throughput of approximately 25 Mbps.
Performance in a multi-radio mesh is much better than the dual-radio or single-radio mesh approaches.The mesh delivers more capacity and continues to scale as the size of the network is increased—asmore nodes are added to the system, overall system capacity grows.
Figure 9 shows the capacity per-AP for the multi-radio configuration shown in Figure 8.We assume achannel capacity of 23 Mbps for each of the point-to-point wireless backhaul links. In this string of APs,without the direct link between AP1 and AP5, total system capacity is limited to a single channel of thebackhaul link because there is only one backhaul link connecting the node to the wired network.Thisdelivers maximum per-AP capacity for up to five mesh APs and then declines with each additional AP.
Figure 9: Per-AP Capacity in a Multi-Radio Wireless Mesh String of APs,Wired Connection at the End or Loop.
Adding the wireless link between AP1 and AP5 doubles system capacity and delivers maximum per-AP capacitythrough 10 APs, as shown.
So the bottleneck in a multi-radio architecture is not in the wireless mesh. System capacity in this architecture islimited by the wired backhaul. System capacity will increase and per-AP capacity will remain stable as more meshAPs are added to the network—as long as there is enough wired backhaul support. Capacity increases beyondthose shown in Figure 9 are possible if there are multiple wired network egress points supplying the mesh.
A more typical multi-radio mesh configuration is shown in Figure 10. In this design there are multiple pathsthrough the network, and a mesh protocol would eliminate the forwarding loops and minimize the number ofhops to the wired backhaul. Larger networks would typically have additional wired egress points to increasecapacity and offer more redundancy in the system.
The estimated capacity of multi-radio mesh is compared to single-radio and dual-radio designs in Figure11.This chart shows the capacity of the different approaches when deployed in a linear fashion around awired connection in the middle.
The capacity of each node in these graphs has been scaled to reflect 802.11 ‘b/g’ compatibility modeoperating at an over the air rate of 54Mbps.The delivered peak tcp/ip rate in this mode is 22-25Mbps.The normal operating scenario of ‘b/g’ compatibility mode delivers approximately 11Mbps tcp/ip capacityas extra packets are transmitted to enable ‘b’ client radios to determine the presence of ‘g’ packets on air.
Figure 12 plots the performance of the different mesh modes in terms of the overall system capacity,for a cluster of nodes connected to a single egress point. It is interesting to see that the single and dual radio mesh systems asymptote to a system capacity close to the capacity of the medium (the air)whereas the capacity of the multi-radio system rises until limited by the capacity of the node at theegress point. In practice this means that in single and dual radio systems, adding more nodes to a cluster does not increase the overall system capacity.
Multi radioDual radio, realisticSingle radio, realistic
There are many other advantages of the multi-radio mesh approach.
• Co-existence—Most large wireless meshes today are designed to support Wi-Fi clients in the 2.4 GHz band.There are many other Wi-Fi devices out there. It is important for largeinfrastructure to fit into the RF environment.A single-radio mesh must use the same channelthroughout the system. (Similarly, the backhaul mesh in a dual-radio system uses the same 5 GHz channel for the whole system.) It is unlikely that this channel will be the best at eachlocation in a large network.A multi-radio mesh is much more flexible. Each access radio canbe assigned a different channel, so the co-existence problem is isolated to the coverage area of a single mesh AP—not the whole system. Multi-radio meshes fit into their environment and share the unlicensed spectrum better.
• Interference—Multi-radio meshes are very flexible in terms of channel assignment on the access or backhaul radios.They can adapt to interference because each access radio can be set to the channel that is least used in a given area.The backhaul network consists of point-to-pointlinks.They use directional antennas that have high gain, but they project their signals in a narrowpattern in a specific direction.This minimizes the impact of the multi-radio backhaul mesh onother systems in the area. Multi-radio meshes have very little self interference because of flexiblechannel assignment and multiple radios operating on different channels at the same time. Bothdual-radio and single-radio meshes cause self-interference, since all the nodes in the mesh mustshare a common channel for backhaul. In addition, interference from external networks in onelocation will disrupt service across and entire dual- or single-radio mesh network.
• Latency—The dedicated point-to-point links in the multi-radio mesh keep backhaul latency lowand predictable. Single-radio mesh and dual-radio mesh approaches have a shared backhaulnetwork using a contention based protocol with unpredictable latency. Multi-radio mesh is suitable for voice applications, the others are not.
Conclusion Capacity in a wireless mesh infrastructure is affected by the mesh forwarding performance, sharednetwork contention and self interference of the mesh APs. It is important to consider all of these issues when analyzing these systems.
Single radio wireless mesh, representing the lowest cost entry point in the deployment of a meshnetwork, is low capacity and will not effectively scale to implement a complete large network.Single radio mesh is best used in small mesh clusters at the edge of a network.
The dual-radio mesh architecture represents the logical evolution in the growth of a mesh network.Dual-radio systems alone don’t scale to metro dimensions.
Multi-radio mesh systems separate wireless access and backhaul, and use dedicated point-to-point links to form the wireless backhaul mesh.This eliminates both in-channel mesh forwarding and sharedbackhaul network contention overhead.The result is a high capacity system that can scale to supportlarge networks with broadband service for many users.
In the real world, large wireless networks require an integrated combination of the three meshapproaches described. It is possible to deploy a very low cost, low capacity network based mostly on single-radio mesh with some multi-radio mesh nodes acting as aggregators for single radio meshclusters. Over time, the network can be upgraded with more capacity or better QoS by replacing single-radio mesh nodes at the edge with multi-radio nodes. Network design should be customized to meet the application requirements and budget by using the appropriate mix of the different wireless mesh approaches.
