Thursday, October 9, 2008

XORs in The Air: Practical Wireless Network Coding

Ah, this paper that has a mechanism true to its name! Routers use the broadcast medium in wireless networks to estimate packet reception probabilities and optimally encode the packets to minimize transmissions. The idea is clearly explained in Figure 3 of the paper. Given the knowledge of which nodes have received which of the packets, the router can accordingly encode the packets and rely on the nodes’ ability to decode the packets. The routing algorithm, COPE, uses the ETX metric to estimate the probability of which of the nodes have received the packets. I didn’t quite understand their explanation of why reception reports won’t work. The authors mention that during periods of high congestions, reception reports are likely to get lost. But arguably, the potential for any smart encoding is going to be limited under high contention as the ETX values are going to very low too (estimating that dynamically is another major headache!).

The coding+MAC gain of COPE was interesting. MAC protocols aim to achieve fairness among all the nodes though the load on the routers and the end-nodes are vastly different. COPE reduces the effects of this “unfair” fairness (!) and also helps in draining the router queues faster.

COPE increases the throughput of UDP flows significantly, but not much for TCP.

I like COPE better than ExOR as it seems applicable in a more practical scenario. We should definitely keep it in the reading list.

1 comment:

Randy H. Katz said...

Implications of maintaining packet pools to admit of coding gains? How are packets flushed or aged out of the pool? Inquiring minds want to know.