Hello guys, I have some info/ideas here that need to be discussed. Some time ago I wrote an article at some of russian IT community sites (habrahabr.ru), which caused a lot of response.
One of commentars was quite sceptic about actual possibility of widespread mesh newtork. He provided two interesting links as a proof: 1) http://www.strixsystems.com/products/datasheets/StrixWhitepaper_Multihop.pdf 2) http://www.belairnetworks.com/resources/pdfs/Mesh_Capacity_BDMC00040- C02.pdf These papers are analyzing current wireless technologies in terms of mesh networking. Authors are saying, that modern single channel AP's are almost unusable in large scale networks, as latency/bandwidth degradation gets more and more valuable. They said that even 5 hops may kill the idea (huh?) even on almost noise-clean environment. Their main argument is half duplex nature of modern WiFi and need of backhaul traffic forwarding along with the user one. Authors conclude that mesh networks may exist only with respect to separation of user/backhaul traffic frequencies/channels. Could you please give a feedback/comment about it? ----- Anyway, whether this is true or not, I may suggest some ideas which probably may be taken in mind: Overpopulated regions with many WiFi hotspots may experience problems with the lack of free channels. Overinterference may cause major slowdown of the whole thing. If we'll try to consciously distribute the channels on our network things may change. My suggestion is to use some kind of map colouring algorithm. Look at the picture on the page http://en.wikipedia.org/wiki/Four_color_theorem Suppose that different colors are representing different channels of WiFi, and different shapes reflecting radiowave interference on the urban relief. Because we know the placement and interconnection of our own spots, we may select the frequencies in a way to minimize interference with other networks, yet achieving good connectivity of our own. Of course this is almost opposite poles and improvement of one will lead to other's degradation. The key may be in multi channel spanning of the network, which needed to be handled by Netsukuku protocols. Suppose we have two nodes: ((( A ))) [[[ B ]]] A operates on channel 6 and B on channel 10. In the current design different channels are treated as completely different networks which may not be aware of each other. Netsukuku daemons running on both nodes could not see each other, thus could not take this into account. Now we adding two more WiFi spots: A2 and B2 linked to their companion by ethernet cable. Operating channels are reversed: [( (A)<-->[A2] )] <~~link~~> [( [B] <--> (B2) )] Now all four nodes may be interconnected into one subnet. Of course, this may be done without any protocol updates. But if nodes will be aware of channels, they may intentionally select the operating channel to increase overall span/quality. On a heavy loaded environment we may intentionally create several layers of Netsukuku using the colouring method described above. By analyzing current state and channel utilization, nodes may vote and decide to migrate to another channel if it will improve the overall quality. On the other hand, such migration may lead to a network split if there are no border nodes linked to a previous channel and fallback internet tunnels are also unavailable. P.S. I was told that recent Linksys APs have two or even three independent transcievers builtin. Such APs may operate on many channels at once and thus may be used as such ‘supernode’. (actually I don't know whether such devices are really exist. maybe this is not true) _______________________________________________ Netsukuku mailing list [email protected] http://lists.dyne.org/mailman/listinfo/netsukuku
