Bill Mccormick wrote: > It does look like the NM code will select APs over mesh... I bet this > plays havoc with IP changing between link local addresses and DHCP > addresses. >
This is partly because of the scalability limitations of the existing collaboration model in a simple mesh. However, I think we can greatly improve situation in a simple mesh scenario and that may simplify the decisions that NM needs to make. > Did you expect over half of the packets in your data file to be > broadcasts? Specifically 11754 out of 21587 packets were sent to the > broadcast address. This is true because all bulk traffic is sent over broadcast _and_ is reliable. Sounds weird? Let me provide a brief explanation. All bulk data traffic (by "bulk" I refer to traffic that is _potentially_ interesting to all nodes, such as profiles, files etc) is sent to the broadcast address, but the mac address of the node that originally requested the data (or the node to which this data is actually meant for) is added in the actual payload, so only one will be replying with acks to the sender (which are necessary to ensure reliability), but all other nodes will be receiving the data too because it's sent to broadcast. This actually happens anyway on your network interface, so this scheme imposes _no_ additional cost in terms of traffic, other than moving the decision of whether a frame is meant for the current host to a higher layer. This scheme allows sizeable bandwidth savings (http://lyme.media.mit.edu/cerebro/index.php/Experimental_results), especially in dense wireless networks. Pol -- Polychronis Ypodimatopoulos Graduate student Viral Communications MIT Media Lab Tel: +1 (617) 459-6058 http://www.mit.edu/~ypod/ _______________________________________________ Devel mailing list [email protected] http://lists.laptop.org/listinfo/devel
