This project evaluates the effectiveness of Mesh networks with unplanned placement, omnidirectional antennas and multi-hop routing in bringing last mile connectivity to communities. The lack of planning might increase the rate of deployment but might result to seriously faulty topologies and low performance. The authors make their case by experimenting with Roofnet which is a network of 37 802.11b wireless nodes with roof mounted antennas spread over 4 square kilometers.
The nodes are installed by volunteers who get to use the network for their own personal usage. The network contains 4 gateways to the global internet. For routing a combination of link-state and source-routing techniques are used and bit-rates of wireless cards are set with an optimized algorithm called Sample Rate. Broadcast probes are used in order to gather the required metrics the routing and bit-rate selection algorithms. The routing algorithm tend to choose links with throughput higher than 2Mbps which are usually not too long and discard the low quality links. The roofnet network achieves 627kbps average throughput and 39ms average delay.
The authors show through simulation results that to achieve good connectivity and performance, dense deployment is a must. The simulation results show that network is not too sensitive to deletion of individual links but can be very sensitive to deletion of individual nodes. It takes the two most important nodes to fail in order for the network performance to drop by 43%!
Even though the authors tried, I believe it is very hard to compare this scheme with the case of single hop access point scheme. In the case of single hop access point architectures even though you might need more gateways to cover the whole area, it is more likely that individuals experience a better throughput and robustness. Another advantage of single hop access points is that they can be set to orthogonal channels in order to minimize the interference and thus will not suffer as much as the Roofnet mesh network.
I also recommend another paper published for the same project which studies the physical layer effects in more depth.
The nodes are installed by volunteers who get to use the network for their own personal usage. The network contains 4 gateways to the global internet. For routing a combination of link-state and source-routing techniques are used and bit-rates of wireless cards are set with an optimized algorithm called Sample Rate. Broadcast probes are used in order to gather the required metrics the routing and bit-rate selection algorithms. The routing algorithm tend to choose links with throughput higher than 2Mbps which are usually not too long and discard the low quality links. The roofnet network achieves 627kbps average throughput and 39ms average delay.
The authors show through simulation results that to achieve good connectivity and performance, dense deployment is a must. The simulation results show that network is not too sensitive to deletion of individual links but can be very sensitive to deletion of individual nodes. It takes the two most important nodes to fail in order for the network performance to drop by 43%!
Even though the authors tried, I believe it is very hard to compare this scheme with the case of single hop access point scheme. In the case of single hop access point architectures even though you might need more gateways to cover the whole area, it is more likely that individuals experience a better throughput and robustness. Another advantage of single hop access points is that they can be set to orthogonal channels in order to minimize the interference and thus will not suffer as much as the Roofnet mesh network.
I also recommend another paper published for the same project which studies the physical layer effects in more depth.
No comments:
Post a Comment