A Performance Comparison of Multi-Hop Wireless Ad Hoc Network Routing Protocols; Broch, Maltz, Johnson, Hu, & Jetcheva

This paper presents performance comparisons of 4 routing protocols for ad hoc wireless networks: DSDV, TORA, DSR, and AODV. It simulates each in ns-2 for use in a network of 50 moving nodes.

Destination-Sequenced Distance Vector (DSDV) is a distance vector protocol that must frequently broadcast its route updates. It uses sequence numbers for choosing new routes such that if a route comes in with a greater sequence number or equal sequence number and lower metric, it selects the new route. Temporally-Ordered Routing Algorithm (TORA) works to quickly establish a route with little overhead using a "reverse link" algorithm, which can sacrifice route optimality.

Dynamic Source Routing (DSR) places the route a packet should take into the actual packet's header, keeping intermediate nodes from actually needing to maintain any routing info, though nodes can cache routes they've seen or overheard to prevent too many route requests from flooding the network. Ad Hoc On-Demand Distance Vector (AODV) is a cross between DSR and DSDV. It sends out route requests which return with number of hops and sequence numbers, and state is maintained on a hop-by-hop basis, with each node remembering next hop info only.

Simulations were run for different amounts of movement of the 50 nodes (from continuous movement to no movement at all) with 10 different patterns of movement for each of 7 different "pause time" values. The authors also varied the movement speed, sending rates, packet size, and number of CBR sources. Also of note is that TCP was not used in the simulation. Three metrics of comparison were used for the different scenarios: packet delivery ratio, routing overhead (total transmitted packets), and path optimality (difference of length for path taken and known shortest path).

In dealing with mobility, DSDV does poorly with short (< 300 s) of pause time, but DSR and AODV provide high packet delivery ratios. DSR generally has the lowest amount of overhead, with TORA having the most, and DSDV has a constant amount regardless of mobility because it is periodic in its broadcasts. The number of CBR sources only appears to affect TORA in terms of packet delivery, though there is a large amount of variability that is not captured by the graphs.
DSR has the least overhead from increased sources in terms of packets, though since each of its packets contains more bytes to deliver route information, it loses to AODV with regards to least sized overhead. I liked this paper a lot because it really took into account a lot of different scenarios for comparing these different protocols; unfortunately, since it was all in simulation, it's hard to say if these tests are really representative of these protocols in the wild.