Layer 3 algorithms
Now consider that the operation of an entire mesh network is based on a Layer 3 routing algorithm.
More than a dozen IP routing algorithms have been tried in one form or another in wireless mesh network deployments. None has become standard, for these reasons:
• It is difficult to configure the algorithms to meet the demanding requirements of large-scale wireless networks.
• Use of the algorithms results in a network that is slow to converge and stabilize, does not provide adequate throughput and is low-latency.
• The algorithms cannot adequately support fast mobile roaming. Injecting routing information into a mesh network, regardless of routing implementation, will reconverge part or all of the network when problems such as interference occur. In wireless networks, excessive collisions can cause links to go up and down, which in turn can cause IP routing algorithms to bounce back and forth from link to link on a particular node and from node to node, thus perpetuating an indeterminate route decision process. Under these circumstances, the mesh cannot adapt in time to maintain continuity or recover an active session,such as a real-time voice-over-IP call or a video application.
Worst-case examples are seen in some large-scale single- and dual-radio networks that employ proprietary predictive routing mesh algorithms that split computer-to-server communications over two connections to the same destination. Should that connectivity be disrupted, the impact can be significant, since these algorithms can take twice the time as optimized Layer 2 mesh algorithms to resynchronize the network and can cause retransmissions of millions of packets for data, voice and video.
A predictive IP routing algorithm must be able to reroute and stabilize instantaneously, and the fact is, that just doesn't happen.
Layer 2 mesh implementations based on the spanning-tree and other Layer 2 algorithms have an impact similar to Layer 3 routing, increasing latency and decreasing the ability to support mobile-roaming applications. Some of the common limitations of these algorithms include:
• lack of support for high-throughput routing metrics;
• support for the shortest-hop-count metric, which favors long, low-bandwidth links over short links with high bandwidth;
• inability to discover a route un- til a flow (a change in the network) is initiated;
• lack of efficient route maintenance techniques--an already-discovered route may not be the optimal route at a later time, because of network congestion or the fluctuating characteristics of wireless links;
• absence of a provision to rediscover the new optimal route; and
• route-discovery latencies. Route-discovery latency can be as high as that of a reactive routing protocol, and route-discovery latency statistics can be high in large-scale mesh networks. Latencies of 50 to 100 milliseconds and larger have been measured, limiting the algorithm's effectiveness as a mobile technology.
