5.1 Summary of Thesis
5.3 Future Work
5.1 SUMMARY OF THESIS
In this thesis, we designed, implemented and evaluated a multi-hop ad hoc messenger using Pocket PCs and Microsoft .Net Compact Framework. Each Pocket PC communicates wirelessly with another using the IEEE 802.11b technology without any aid of infrastructure. The main protocol implemented in this application was the DSR (Dynamic Source Routing) protocol, which consists of two important mechanisms, Route Discovery and Route Maintenance. Since the DSR protocol operates solely based on source routing and on-demand process, it has been selected as the routing protocol to be implemented and tested for our ad hoc messenger application characterized by a source on-demand chat conversation between nodes in a mobile ad hoc network. The mobility behavior of nodes in the application is modeled by the random waypoint model through which random locations to which a node move are generated, and the associated speed and pause time are specified to control the frequency at which the network topology is changed.
To test the performance of our wireless ad hoc messenger using DSR, five standard metrics are evaluated, namely, average latency to find a new route, average latency to deliver a data packet, delivery ratio of data packets, normalized control overhead, and throughput. These metrics tests if the application using DSR functions correctly and efficiently in an ad hoc network.
To provide the testing conditions, five parameters are manipulated reflecting changes in the network topology and design alternatives of the wireless ad hoc messenger application. In order to test the performance of the implemented ad hoc messenger under changing network topologies, the node speed and pause time parameters are manipulated. Three design alternatives, namely, Sliding Window Size (SWS), the use of receipt packets, and the use of cache, are examined. Specifically, SWS in our application is increased to know its effect of parallelism of data processing on the performance of our application. The effect of the use of receipt packets deigned for rapid detection of route errors by intermediate nodes is evaluated. Finally, the effect of using cache structure is tested.
The performance of our wireless ad hoc messenger using DSR has been evaluated against the five metrics selected based on scenarios generated through manipulating the speed and pause time of the mobility model and design variables (SWS, receipt packets, and cache structure). The results are summarized as follows:
The effect of speed on the performance of our wireless ad hoc messenger using DSR: As the node speed increases such that the network topology becomes more dynamic, the performance deteriorates.
The effect of pause time on the performance of our wireless ad hoc messenger using DSR: As the pause time increases so the network topology becomes more stable, the system performance improves.
The effect of increasing SWS on the performance of our wireless ad hoc messenger using DSR: There exists an optimal SWS value under which the average latency to find a new route would be minimized and the throughput would be maximized. When SWS goes beyond the optimal size, which is 2 in our experiment, the system performance degrades because the delay experienced per data packet in the presence of multiple data packets being transmitted simultaneously in IEEE 802.11 outweighs the benefit of data parallelism as SWS increases.
The effect of the use of receipt packets on the performance of our wireless ad hoc messenger using DSR: Receipt packets are designed to more rapidly detect broken routes by intermediate nodes. We discovered that there is a “break-even” point in term of the frequency of topology changes beyond which the use of receipt packets can improve the performance of the system because the use of receipt packets allows route errors to be detected quickly and reduces the delay caused by redundant retransmissions of data packets through an out-of-data route.
The effect of the use of cache structure on the performance of our wireless ad hoc messenger using DSR: We discovered that the use of cache is not beneficial or even harmful when the network topology is relatively dynamic. This is because route entries store in the cache in a highly dynamic network can become invalid quickly so using them to search for a route during the Route Discovery process can generate more route errors and cause more delays. On the other hand, when the network topology is relatively stable, cache entries are likely to be valid so the use of cache structure can improve the system performance in terms of the average latency to find a new route and the system throughput.
Based on these experimental results, it is clear that the three design alternatives studied in the thesis suggest that the ad hoc messenger application should be implemented in an adaptive manner such that the SWS value, whether to use receipt packets, and whether to use cache can be determined dynamically based on the current network condition detected through feedback from the system. A possible method is to dynamically measure the performance of system in terms of the five performance metrics introduced in the thesis and fine-tune these three parameters to maximize the system performance.
With respect to the purposes of the thesis, the followings goals have been accomplished:
A real chatting prototype using Pocket PCs has been designed and implemented using IEEE 802.11b wireless cards, Microsoft .Net Compact Framework (C#), and XML technology.
A real chatting prototype has been implemented based on the DSR protocol, fully implementing its basic two mechanisms, Route Discovery and Route Maintenance.
XML was utilized to describe data packets and result files to facility data exchanges and analysis.
A GEN application has been developed to generate a network configuration file modeling changing network topologies based on the random waypoint mobility model.
An Automatic Data Collection application has been developed to ease the testing and evaluation of our ad hoc chatting application. In particular, the Automatic Data Collection application implemented automatic data transmission, generation of result files using XML technology, and collection of result files using wireless transmission (using the Send and Receive Files application) into a centralized Statistical Analyzer application for generating appropriate result graphs according to a set of selected x-parameter and y-parameter.
The performance of our wireless ad hoc messenger under various scenarios has been tested and analyzed appropriately against five standard metrics and the results indicate conditions under which design alternatives (SWS, the use of receipt packets, and the use of cache) should be employed in order to improve the performance of the application. Physical interpretations of the results along with suggestions for dynamically adjusting the values of these design parameters so as to optimize the system performance are given.
5.3 FUTURE WORK
This work designed, implemented and evaluated a wireless multi-hop ad hoc instant messenger with only 7 Pocket PCs fully implementing the basic features of the DSR protocol along with a few design alternatives. It paves the way for more future work incorporating possibly more nodes and more optimized features of the DSR protocol. For example, an optimized cache structure [Lou2002, Wu2002, Hu2000], as mentioned in the related work of Chapter 2, can be considered and its effect may be tested with more nodes in the system. In addition, other optimized features [Johnson2001, Seet2003] can be implemented and their performances can be evaluated, such as reducing route discovery reply packets for route discoveries performed, packet salvaging, replying route request using cached routes, and so on, as mentioned in the additional Route Discovery and Maintenance features in Chapter 2.
The use of receipt packets at the network layer may not be necessary if the underlying link layer subsystem can return an ACK to indicate if a packet can be delivered successfully over a link between two hops. Exploration of a link-layer API is called for in future work.
The design of the wireless ad hoc messenger can also be improved by separating application-layer from network-layer implementations. The current implementation uses the same port to send/receive application-layer data packets and network layer route discovery and maintenance packets. Future work should separate network-layer protocol implementation from application-layer ad hoc messenger application implementation to follow the design concept of layering and separation of concern. This would allow other applications to be easily built on network-layer implementation.
A more sophisticated design for the GEN program for creating dynamically changing network topologies to test the ad hoc messenger application is also called for. When the ACM option is turned on, the topology generated can reflect various degrees of connectivity of nodes by design, e.g., generating a mesh vs. a tree structure, and to test the effect on system performance of the application.
Furthermore, future work can test performance of other routing protocols, such as AODV, DSDV, OLSR, geographical forwarding [Li2001], etc. to compare against the DSR protocol. In order to optimize the use of constrained resources in an ad hoc network, mobility prediction [Su2001] and battery power conservation techniques [Xu2001] can be developed and experimented to test the effect of these ad hoc routing protocols on a real application, such as the real ad hoc messenger developed in the thesis. Moreover, our small-scale network consisting of only 7 Pocket PCs can be extended to a large-scale deployment with more mobile devices included using Bluetooth or IEEE 802.11 technology in the future [Chen2002].