Many organizations recognize the value that a wireless mesh network can provide, but are unsure whether they should implement one that is managed or ad hoc. The decision revolves around how to implement an authentic network without wires: one that is scalable, secure, managed and architected for performance. With these goals in mind, the many differences between the two approaches require the designer to consider a variety of factors. That said, the need for mesh networking and its many inherent advantages has led to a call for standards in ad hoc networks that should solve many of the outstanding issues.
Mobile-communications networks prompted the desire for ad hoc information exchange without the need for fixed infrastructure or a centralized switching station. Ad hoc mesh networks allow for a pure peer-to-peer information exchange with heterogeneous radios, allowing an assortment of different links to be present in the network. Those links might include, for example, ultrawideband and Wi-Fi in the home or code-division multiple access and Wi-Max in the wide-area network.
Mesh protocols address link discovery and route selection. Examples include the Mitre MMLDP and MMRP protocols within the ad hoc sector, and MMBDP to tunnel through a wired network. These protocols are illustrative: They use a single hello message type to identify neighbors by Internet Protocol (IP) address, an unspecified mechanism to report link events and a metric to compute least-cost paths. This latter, unspecified metric may be a value representing signal quality, data rate or other appropriate choices. These protocols are simple and effective for their intended purpose, but they are hampered by the fact that broadcast messages consume bandwidth.
Multicasting can be used to conserve bandwidth. In the case of wireless devices that are part of a multicast group-such as users in a convention center, classroom or similar work group-driven site-a multicast transmission to a single destination address can be used. Multicast extensions to the Ad-hoc On Demand Distance Vector routing protocol illustrate this case.
The AODV protocol creates a multicast tree shared by all the nodes updated by periodic hello or route request messages. Thus, AODV provides a single path between nodes. A similar protocol, the On Demand Multicast Routing Protocol, uses forwarding groups to further limit the broadcast messages updated by periodic route refreshes from a source node. ODMRP provides redundant routes within a mesh topology. Both of these protocols discover routes only when they need to deliver data traffic.
Dynamic-source routing improves upon these protocols by embedding the full route information in each data packet, eliminating the need for intermediate nodes to store a routing table. DSR typically performs best in a stable network at relatively low data rates (sub-1 Mbit/second), making it less suitable for large, dynamic networks.
Networks built with these mesh protocols may be unarchitected, but they are unpredictable and insecure. A managed mesh network overcomes these issues by being structured and optimized. The network nodes adopt specific roles instead of operating in a complete peer-to-peer fashion. Certain nodes provide access for wireless devices at the network edge and transport the packets into the network core. Backbone nodes may also provide access, but their primary purpose is to form the engineered mesh. Wired nodes include ingress/egress connections to the wired LAN, and they may also provide access to wireless devices.
The entire network system is optimized to maximize performance (maximum throughput with minimum latency) at the lowest cost (number of required nodes and overhead from control signaling). Today the radios in a managed mesh backbone are homogeneous-for example, all 802.11a or 802.11g-enabling free substitution of links while assuring performance. Access radios may be heterogeneous.
A managed mesh goes beyond simple link discovery and routing, as nodes must discover their roles. Once roles and links are discovered, paths must be selected, often using a combination of values such as received signal strength, latency, throughput and error rates. Since many 802.11 chip sets vary power and link rate with observed error rate, the selection algorithm must also be robust enough to compensate for these changes.
Routing in a managed mesh is usually accomplished by Layer 2 switching to achieve low latency over multiple hops. Since switching is distributed throughout the nodes, decisions can be local and switching tables small. This lets the network scale up without depending on a large routing table-either a central table, or one that is periodically distributed throughout the network with attendant overhead costs.
Managed mesh networks include a number of interacting subsystems that control discovery (of both infrastructure and devices); distribution, meaning path selection and re-selection based on network and traffic changes; management, i.e., configuration and monitoring; and security, which includes authentication and encryption. Typically, a managed mesh relies upon several state machines to enable this automatic operation. Such networks are well-architected, with bottlenecks analyzed during system design, and are able to automatically deal with radios, best routes and failures.
See related chart |
A new node joins the mesh using a discovery protocol to broadcast its presence. Existing nodes reconfigure the network to incorporate it.
To date, more than 60 mesh networks and their variants have been proposed, and that number is growing. A taxonomy for such networks includes those that use table-driven, on-demand, hierarchical, geographical, power-aware, multicast or geocasting techniques. As mesh networks enter the mainstream, one can readily see the desire for an overarching standard. And, as one considers the variables among networks, one can readily understand the reluctance of vendors to become enmeshed in such an undertaking.
A network designer needs to consider much more than protocols. For example, since the goal of an ad hoc mesh network is for each device to be a router, one can readily envision the laptop computer as both an end node and an intermediate node in a mesh. For the network designer, this raises the question of security (Who else is transiting my computer?) and bandwidth (Who else is sharing my limited-link throughput?). While not insurmountable, these issues demonstrate the need to think through the entire implementation and not just the technical elegance of the mesh protocol.
More to the point, a network designer should examine the desired use of the network and then establish the benchmarks. These might include the choice of radio technology for the application; the throughput (not link rate) available to each simultaneous wireless user; interfering sources and RF obstructions; the degree of redundancy desired; and the management information required to assure optimum performance of the system under varying traffic loads.
At a higher level, the designer should consider installation requirements, and any anomalies caused by the particular protocol in use. Interoperability among radios-or said differently, multi-RF capability in the network nodes-is another consideration, tied closely to the intended application and its likely extension over time. Multi-RF capability also determines the ability to use less crowded spectrum for some tasks, such as a 5-GHz backbone combined with 2.4-GHz access.
Finally, the designer should consider a plurality of networks. The wired and wireless networks typically connect using Ethernet frames and IP packets, making interoperability a nonissue. A designer might want to use ad hoc mesh techniques to connect disparate equipment, as in a personal-area network, or in the future to connect disparate radios in a mesh within a home network. It is here that standards work would prove beneficial.
It is much less likely that a designer would want to connect nodes from different vendors over wireless mesh links in a commercial network. While this is conceivable in a mobile ad hoc environment, it is more difficult to foresee the benefits in an architected or managed mesh scenario. The trade-off would be performance vs. mixing and matching equipment. To the extent that wireless interfaces are standard, the benefits from further standardization are unclear. More likely, various mesh networks would coexist, with internetwork connectivity based upon a common denominator such as Ethernet and TCP/IP.
Ad hoc networks may be appealing for their ability to readily link many devices as peers. But managed mesh networks deliver the engineered performance and stability of a wired LAN and the dynamic advantages of an unwired mesh topology.
Bob Jordan is vice president of marketing at Strix Systems Inc. (Westlake Village, Calif.).