A Performance Comparison of Multi-Hop Wireless Ad Hoc Network Routing Protocols

Summary
This paper aims to make a fair comparison between four ad-hoc wireless routing protocols. The paper describes each protocol, their simulation environment, the workload they test with, and then the results they got for each protocol.

The protocols tested are:

• DSDV, which does "next hop" routing by maintaining a routing table at each node with a sequence number for each route. Nodes advertise routes with increasing sequence numbers.
• TORA, uses a link reversal algorithm, wherein a Query request is sent though the network, then an Update comes back with the best route. A flow model is used where nodes have height and unreachable nodes are modeled as having a greater height the the local height.
• DSR, which floods the network with Route Request packets when it needs a route. Nodes cache routes and aggressively stop Route Request packets with Route Reply packets for routes they know about. DSR also snoops on traffic to hear or compute better routes.
• AODV, which is a combination of DSDV and DSR. Routes are discovered as in DSR but stored all the way along the path to the destination. Sequence numbers are used as in DSDV.

The results of the paper look at % of packets delivered, routing overhead (i.e. # of route related packets transmitted, and path optimality (i.e. how many more hops did I take than I really needed).

At a high level, DSR performs much better than most of the other protocols, with DSDV performing very poorly, especially with a lot of mobility which basically causes TORA to go into congestion collapse. All protocols perform decently well with low mobility, although TORA and AODV send many more routing related packets. DSDV does have the advantage that it sends a fairly constant amount of routing related information.

Comments
I enjoyed this paper. It gives a fairly clear description of each of the protocols, as well as issues the authors encountered while actually trying to implement each one. The analysis goes to great lengths to try and make the comparison fair, and it seems they do a decent job of that (not using TCP for example as it will vary transmission times and rates based on the current state of the network). However, any real deployment would use some transport protocol and it would be nice to see how each of these performed on TCP, the most likely choice.

It was interesting the DSR, probably the most simple of the protocols, performs so much better than all the others. Perhaps the added complexity of 'clever' routing protocols in mesh-networks simply isn't worth it.