I don't buy this. This paper describes a different path metric, ETX or Expected Transmission Count. The concept is that, on wifi links, hop-count is broken. The hops have differing properties, specifically in terms of quality, and we need to take this into account when picking a path. That's a good idea, but transmission count is not a reasonable metric.
The author is attempting to maximize throughput. If all of the wifi links are identical aside from expected retransmission count, then this is plausible. Retransmissions on a wifi drop are why throughput drops. If we expand this to include links with different throughput rates, this metric is totally meaningless. To maximize throughput, we have to know the throughput of the link and the retransmission rate. With these we can compute the expected throughput.
They argued that this could not cause cycles in the network, but I think that depends on their distance vector calculations. Congestion in a wifi-link may be exposed as retransmissions rather than just drops. So, as we route to the lowest retransmission count, this may cause retransmissions in the area, and so on.
I don't think is useful to begin with. Most modern wifi-links are one-hop to the wired network, which has much higher throughput. A network of wifi-routers are generally used for sensornets, which don't often need to maximize throughput.
There's more here, but my keyboard is acting up and it's hellish to keep this post even vaguely comprehensible.
2 comments:
You come out pretty strongly against this... I would just say that the dual of maximizing throughput in this case is minimizing transmissions, which reduces total energy expended and channel noise-- both very good things. The point about multiple bit rates is well taken, though; that's probably why they switch to ETT in the roofnet paper.
Yea what would be the place for Roofnet given that we have pretty good wireless stuff all over the place anyway?
Post a Comment