Simulation Parameters

All the simulations are conducted using NS-2. The Distributed Coordination Func- tion (DCF) of IEEE 802.11 for wireless LANs is adopted as the MAC layer pro-

5.4 Simulations in NS-2 99

tocol. Therefore, the two nodes which make up a link will assign the link timeout and delete the expired link from their respective caches practically simultaneously, avoiding extra overhead traffic.

The simulation model used in this chapter is based upon that used in [41], for comparison purposes. Thus, the MANET simulated in this paper consists of 50 mobile nodes (MNs), moving according to a RWMM, in a bounded area of size 1500m × 1500m. The Two-Ray Ground Reflection Approximation [79] is used as the radio propagation model with a transmission range of 250m. The radio model is based on the Lucent Technologies WaveLAN 802.11 product, providing a 2Mbps transmission rate. The communication traffic simulated in all scenarios is generated by the NS-2 traffic generator script. Similarly to [41], 20 constant bit rate (CBR) connections (each with a maximum of 1200 packets for transmission) are generated, with a packet rate of 4 packets per second. The size of each data packet is 64 bytes, and each node has at most 2 CBR connections at the same time. Note that we choose small packets because this leads to network performance being dominated by changes of network topology rather than other issues, such as network congestion. The simulated time is set to 900 seconds per trial. For each node speed, 10 different mobility scenario files are generated, and the final network performance metric is the average of these trials. Note that the scenarios were generated in advance, and that identical scenarios were used to evaluate each of the caching schemes.

5.4.2

Simulation Results

In Fig. 5.2 network performance, using various measures, with the uniform path du- ration link caching scheme and the adaptive link residual time link caching scheme are compared with that for the static-5 link caching scheme. The performance mea- sures compared are packet delivery ratio, end-to-end delay, packet overhead and average path length. Note that the packet delivery ratio and the end-to-end delay are calculated with respect to transmitted and received data packets, whereas the routing overhead is calculated with respect to routing control packets.

For each of the performance measures: packet delivery ratio, end-to-end delay and overhead, as would be expected, the adaptive link residual time scheme per- forms the best overall, followed by the path duration scheme and then the static-5 scheme. This is because the static-5 scheme uses the same link cache timeout re- gardless of average node speed; the uniform path duration scheme, similarly to the static-5 scheme, utilizes the same value for every link in the network, but chooses a different timeout value for different average node speeds; and, finally, the link residual time scheme adapts to the properties of each link, taking into account av- erage node speed but also including other parameters in its choice of cache timeout value. That is, the link residual time scheme “fits” the given network more closely.

100 Choice of Timeout for Link Caching 0 5 10 15 20 25 70 75 80 85 90 95 100

Mean Node Speed

Packet Delivery Ratio (%)

Adaptive−LRT Uniform−PD Static−5

(a) Packet Delivery Ratio

0 5 10 15 20 25 0 500 1000 1500 2000 2500 3000 3500

Mean Node Speed

End−to−End Delay (ms) Adaptive−LRT Uniform−PD Static−5 (b) End-to-end Delay 0 5 10 15 20 25 0 50 100 150 200 250 300 350 400

Mean Node Speed

Overhead (thousand packets)

Adaptive−LRT Uniform−PD Static−5 (c) Overhead 0 5 10 15 20 25 3.3 3.4 3.5 3.6 3.7 3.8 3.9 4 4.1 4.2

Mean Node Speed

Average Path Length

Adaptive−LRT Uniform−PD Static−5

(d) Average Path Length

Figure 5.2: Performance Comparison between Static-5, Uniform-PD and Adaptive- LRT Link Caching Strategies using the Random Walk Mobility Model

5.4 Simulations in NS-2 101

We then consider the average path length, as shown in Fig. 5.2(d). It can be seen that the performances of the three caching schemes are practically indistin- guishable. This is because the average path length is largely dependent upon node density, relating to the transmission range, the simulation area size and the number of simulated nodes. The caching scheme is, then, largely inconsequential.

It is interesting to note that there is very little difference in network performance among the caching schemes when the average node speed is low, despite large differences in the actual cache timeout values. For example, for an average speed of 2m/s, the uniform path duration caching scheme has a link timeout of 27.7 seconds, which is more than 5 times greater than that for the static-5 caching scheme. The negligible performance differences are because at low speeds the network has a highly stable topology so that all of the links are highly reliable. Thus, even in the static-5 caching strategy, the link timeout is extended frequently and rarely expires, due to constant incoming communication traffic. Consequently, route discovery is required infrequently resulting in lower routing overhead.

However, it can be seen that the newly proposed caching schemes gradually outperform the static-5 scheme as the average node speed increases, since they can adaptively track the changes of the network topology and the link reliability. Com- pared with the static-5 strategy, the adaptive link residual scheme and uniform path duration scheme improve packet delivery ratio 7 percent and 4.3 percent, re- spectively, when the average node speed is 22m/s. The end-to-end delay is reduced by 990ms and 800ms, and the routing overhead is reduced by 95000 packets and 76000 packets, respectively.

5.4.3

Implementation Issues

We note that the path duration caching scheme is easy to implement. The original DSR simply requires the addition of a 1-byte field into the routing packet, carrying the current clock time plus path duration. The implementation of the link residual time scheme requires the aid of node location information, which is piggy-backed in the packet header. As mentioned in Section 5.3, one way to obtain the node location is to utilize GPS. Alternatively, the relative distance between a pair of nodes can be estimated by other means, such as TDOA.