ns-2 Project Report

by Patrick Carpenter

April 26, 2011

Note: This is an abbreviated version of the report. The full report, all simulation and data-processing scripts, and summarized output data can be downloaded from this site by clicking on the appropriate link: report.pdf, scripts.tar.gz, data.ods.

Overview & Introduction

The purpose of this report is to recommend either a wired or wireless network configuration to be implemented on a school campus. To make an informed recommendation, the ns-2 network simulation tool was used to measure long-term, end-to-end throughput under a variety of scenarios and compare the relative performance of both options. Measures derived from throughput - such as efficiency, link-utilization and fairness - were also considered.

The wired configuration consists of a chain of six hosts and six routers, where each host is connected directly to one router and the routers are connected in a chain (See Figure 1). The bandwidth and propagation delay of each of the eleven links are 1.5 Mbps and 5 ms, respectively.

The Wired Configuration

Figure 1: The Wired configuration.

The wireless configuration consists of six hosts arranged in a straight line, where the distance between adjacent hosts is fixed at 150 m (see Figure 2).

The Wireless Configuration

Figure 2: The Wireless configuration.

Method & Analysis

Three sets of experiments were used to evaluate the performance of the wired and wireless configurations:

Experiment Set A: This set of experiments is designed to measure the throughput achieved between five pairs of nodes - {(1, 2), (1, 3), (1, 4), (1, 5), (1, 6)} - when no other connetions are competing for resources. The results of this set of experiments can be useful for putting an empirical upper bound on the throughput which can be achieved over a certain number of hops.
Experiment Set B: This set of experiments is designed to measure the throughput achieved between five pairs of nodes - {(1, 2), (1, 3), (1, 4), (1, 5), (1, 6)} - under varying degrees of competition for resources. Specifically, a connection is established between the pair of nodes (3, 4) that runs concurrently with the other connections. The results of this set of experiments are interesting in that they give some idea of how one connection can interfere with, or influence, the performance of other connections.
Experiment set C: This set of experiments is designed to measure the throughput achieved by three concurrent connections between the pairs of nodes {(1, 6), (2, 5), (3, 4)}. The results of this set of experiments are interesting since they show performance under a somewhat realistic scenario. Since there are meaningful connections active simultaneously, fairness - in terms of the evenness of the distribution of resources - can also be derived.

In each case, the steady-state, end-to-end throughput is taken as the preferred measure of performance. There are at least a few reasons for this choice:

Packet loss, efficiency, and other measures of performance are generally only important inasmuch as they affect the end-to-end throughput (i.e., other measures are not generally visible).
Other visible measures (such as delay) may unfairly penalize one configuration or the other. This is also why we consider only the steady-state throughput.
The metric is easy to measure, easy to understand and important to actual users. Moreover, comparing the throughput of multiple competing connections can be used to gauge other meaningful criteria, such as fairness.

Experiments are conducted via simulation using the ns-2 network simulation tool. Trace files are post-processed using scripts which count the number of packets successfully delivered each second, and these data are analyzed to determine time-varying connection throughput for each scenario.

Results & Conclusions

Figures 3, 5 and 7 show the time-varying throughput of the Wired configuration for experiment sets A, B and C, respectively. Figures 4, 6 and 8 show the same for the Wireless connection. In general, the Wired configuration leads to more stable and evenly-distributed throughput with better scalability and efficiency and lower latency, whereas the Wired configuration leads to an overall higher steady-state throughput in many cases.

Bandwidth vs Time

Figure 3: Results for Wired Connection, Set A

Bandwidth vs Time

Figure 4: Results for Wireless Connection, Set A

In the above figures, we see that the Wireless configuration leads to better steady-state throughput for connections involving fewer than three hops, and that the Wired configuration performs better for connections involving more hops. Note also the delay in the Wireless configuration, and the inversely-proportional relationship between steady-state throughput and the number of hops.

Bandwidth vs Time

Figure 5: Results for Wired Connection, Set B

Bandwidth vs Time

Figure 6: Results for Wireless Connection, Set B

In the above figures, we see that the Wireless configuration leads to distinct levels of throughput depending on the connection; the level of throughput is correlated to the degree of interference by the (3, 4) connection. The long-term, end-to-end throughput is reached quite quickly in any event. By comparison, the Wireless configuration leads to noticeable delays for connections requiring more than one hop, and the long-term throughput is reached over a much longer period of time. Also, the peak throughput of connections under the Wireless configuration is affected regardless of the connection (in the Wired configuration, connections which avoid interference are not affected).

Bandwidth vs Time

Figure 7: Results for Wired Connection, Set C

Bandwidth vs Time

Figure 8: Results for Wireless Connection, Set C

In the above figures, we notice the closeness of the achieved throughput by the three connections under the Wired configuration, and the comparatively more distant throughputs under the Wireless configuration. This is an indication of a higher degree of fairness achieved by the Wired configuration in evenly distributing resources across competing connections.

Recommendation

After collecting and analyzing simulation data, it is this author's recommendation that the wired configuration be selected for use in this case. Some of the major factors which led to this decision are:

The Wireless configuration demonstrates less scalability, and does not outperform the Wired configuration by as much as could be hoped for by a so much more modern technology.
The Wireless configuration leads to a much less equitable, or fair, distribution of communication bandwidth among competing connections than does the Wired configuration.
The fact that the hosts do not move negates one of the most compelling reasons for adopting wireless technology; without mobility, wireless technology is almost universally inferior (in terms of reliability and security, at least) to equivalent wired technology.

Given the inferred usage of this network by a broad and uncoordinated segment of the student body, and under the assumption that most users care more about consistency than peak achievable performance, and looking into future expansion of the system, the Wired configuration emerges as the winner. Note that if peak and average throughput were the only criteria, the Wireless configuration should be chosen. That being said, comparing 802.11b (with a peak rate of 11 Mbps) and a wired connection of 1.5 Mbps capacity isn't necessarily a fair comparison, in the sense that the disparity in channel capacity is almost an order of magnitude. If the Wired configuration's channel bandwidth increased even by a factor of two (2), given the results of this simulation, it seems clear that a Wired solution would win on all counts.

References & Acknowledgments

The simulations were developed using Mark Greis's tutorial as a guide and reference. Joshua Robinson's article was also used to determine some options for the wireless protocol to make the simulation as realistic as possible.