Architecture and Evaluation of an Unplanned 802.11b Mesh Network

J. Bicket, D. Aguayo, S. Biswas, R. Morris, "Architecture and Evaluation of an Unplanned 802.11b Mesh Network," ACM Mobicom Conference, (September 2005).

This paper evaluated ROOFNET, an unplanned wireless mesh architecture involving omni-directional antennas and multi-hop routing. The whole design criteria was based on assuring high throughput and making the entire system very easy to deploy without any constraints imposed by node-placement planning.

In brief, ROOFNET was deployed over a 4 sq km area in Cambridge, MA consisting of 37 nodes each with omni-directional antennas. It assumes that a small fraction of nodes will voluntarily share its wired Internet access and will be called Gateways (The authors' implementation had 4 gateways). Each node further allocated 192.168.1.x IP address blocks to user hosts attached to the node's Ethernet port and used NAT. Roofnet's routing protocol, Srcr maintained partial dynamic database  of link metrics between various pair of nodes and used Dijkstra's algorithm on the database to find routes. DSR-style flooding was used in case the node had to route a packet in absence of a link metric for that corresponding route. Further, the link metric consisted of Estimate Transmission Time (ETT) which predicted the total amount of time it would take to send a data packet along a multi-hop route.

The authors presented an extensive evaluation of the real system and showed that in a random setting, on an average 2.9 hop path formed to the gateways resulting in an average throughput of 627 kbits/sec and 39 ms average latency. Further the authors simulated a denser network and showed that dense network had increased hop count (possibly due to more number of eligible links) as well as higher throughput (possibly because of shorter hops). One important problem between practice and simulations was that in practice the authors observed link collisions which affected the theoretical throughput value calculated at each node for routing however, they claimed that this is not so bad and to avoid this, a scheme involving RTS/CTS will in degrade the performance even further due to additional overhead.

Critique

Overall, I really liked the Roofnet technology. It was definitely simple, easy to implement and robust to failures which makes it ideal as a candidate for deployment. Moreover, the best thing about it was that majority of the claims made by authors were based on actual deployment of a working system. However, I had 2 questions in mind which the paper left unanswered. Firstly, how often does the distance metrics changes? In a continuously working system, I would assume that the metrics would eventually converge to static values... am I missing something here (probably dynamic congestion considerably affects throughput?). Secondly, assuming that this system is a candidate for commercial deployment, I would assume that it will have a single antenna for each building and all the apartments having Internet access through the node's Ethernet ports. In view of such a scenario, I would have liked to see the effect of varying number of end hosts on a particular node. It may well be possible that a node is more congested for some part of the day and less on other time. It would be interesting to discuss these aspects in class.