This paper for the first time introduces Expected Transmission (ETX) count metric to be used instead of the typical hop-count metric and proves its superiority by experimental tests.
Using the hop-count as the routing metric over a multi-hop network can typically result to choosing a route which is not close to optimum. Minimizing the hop count typically result to using longer links which are inclined to be more lossy. Trivial solutions such as masking errors at link layer or discarding lossy links fail to work as the first one reduces the link throughput and the later disconnects nodes reachable only through lossy links.
With the insight authors had gained from real experiments they designed ETX to address the following issues:
The delivery ratios are estimated by counting periodic broadcast packets emitted from each node. The authors prove the usefulness of ETX by implementing it over DSDV and DSR routing protocols and show the gain of upto 2x or larger when the number of hop are 2 or more.
I found this paper the most interesting paper we have read so far. The authors not only introduced a new concept and thoroughly evaluated it in real setting but also were fair on emphasizing on the cases when ETX does not outperform other schemes as much. They carefully studied the effects of variable packet size and proved that in a more well connected network, decreasing the average number of hop-count will close the performance gap.
I wonder if the issue of packet size can be alleviated by estimating the actual delivery rate from the observed one with a different packet size knowing that the slopes of the curves in figure 14 is quite constant and further using different estimations for forward and reverse directions as ACK packet is typically smaller. I definitely vote for keeping this paper in syllabus.
Using the hop-count as the routing metric over a multi-hop network can typically result to choosing a route which is not close to optimum. Minimizing the hop count typically result to using longer links which are inclined to be more lossy. Trivial solutions such as masking errors at link layer or discarding lossy links fail to work as the first one reduces the link throughput and the later disconnects nodes reachable only through lossy links.
With the insight authors had gained from real experiments they designed ETX to address the following issues:
- The wide range of link loss ratio: many links with intermediate quality exist
- The existence of links with asymmetric loss ratios: 22 of 29 nodes had at list on link with more than 25% difference quality of forward and reverse direction
- The interference between successive hops of multi-hop paths: two hop route route with perfect delivery per link achieves 50% of a single link
The delivery ratios are estimated by counting periodic broadcast packets emitted from each node. The authors prove the usefulness of ETX by implementing it over DSDV and DSR routing protocols and show the gain of upto 2x or larger when the number of hop are 2 or more.
I found this paper the most interesting paper we have read so far. The authors not only introduced a new concept and thoroughly evaluated it in real setting but also were fair on emphasizing on the cases when ETX does not outperform other schemes as much. They carefully studied the effects of variable packet size and proved that in a more well connected network, decreasing the average number of hop-count will close the performance gap.
I wonder if the issue of packet size can be alleviated by estimating the actual delivery rate from the observed one with a different packet size knowing that the slopes of the curves in figure 14 is quite constant and further using different estimations for forward and reverse directions as ACK packet is typically smaller. I definitely vote for keeping this paper in syllabus.
No comments:
Post a Comment