I will contribute to one of the Freifunk projects; nodewatcher, via Google Summer of Code this summer and I wanted to keep you updated on my progress as well as exchange thoughts about my ideas.
First of all, nodewatcher is an open source, modular community oriented platform used for network planning, node deployment, node monitoring and maintenance. nodewatcher was initially developed to be primarily used by the wlan slovenija project. With 1336 nodes, it’s really successful and a great example for community networks. As nodewatcher gets deployed elsewhere with even more nodes, it’s natural to ask ourselves if we can be smarter about allocating spectrum to our wireless nodes – these nodes are mostly inexpensive wireless routers but it’s natural to extend the meaning of the term to dedicated wireless access points (i.e. Unifi AP).
The theoretical foundation for this problem is fascinating by itself: Each node has a different amount of noise in each channel (the 2.4GHz band allows 3 non-overlapping channels where each channel is 20MHz wide) and each node wants to maximize its SNR (signal-to-noise ratio). I will term this as the greedy approach, which is already used in enterprise level devices. However, in an urban setting, nodes are close enough to each other for their signal to act as noise to other nodes. The greedy approach is no longer optimal as it bears a high price of anarchy. Instead, our goal is now to maximize the sum of channel capacities (under a power constraint). I will have to devise an algorithm to solve this problem and the algorithm does not seem trivial since the number of combinations is increasing exponentially with the number of nodes in the system. Even with only 10 nodes, we haveover 59000 possible allocations on 2.4GHz band and over 95 trillion on the 5GHz band.
Traditional networking literature tackles this optimization problem with Lagrange multipliers. An alternative is to look at approximate graph coloring schemes and compute chromatic numbers. I hope to experiment across various settings and approaches.
Over the course of the project, I hope to experiment with a real network which consists of at least 10 nodes and measure the improvements. One exciting thing about real life experiments is that nodewatcher was mostly used inside wlan slovenija’s network and I get to run it independently! This will probably allow me to fix some bugs on the way and contribute to nodewatcher in this aspect as well.
The algorithm will initally be developed as a nodewatcher module, but I hope to eventually port it to openwrt (possibly after the summer ends). The main difficulty is that nodewatcher can act as a central level planner, whereas the openwrt scenario requires negotiation among nodes. So it’s harder to convince a node to decrease its TX power to benefit other users. But imagine a network where nodes can communicate and achieve a socially optimal point of spectrum allocation! A glorious future awaits us.