Throughput Maximizing Routing in a MANET: Protocol and Analysis
| Xiaojing Tao, Thomas Kunz, David Falconer | |
| Systems and Computer Engineering | |
| Carleton University | |
| tkunz@sce.carleton.ca |
| Routing in wireless Ad Hoc network | ||
| Proactive (OLSR, TBRPF) | ||
| Reactive (DSR, AODV) | ||
| Problems in Reactive Routing Protocols | ||
| Finding the shortest path between sender and receiver by flooding the requests | ||
| Shortest path leads to network congestion in central areas | ||
| Caching mechanisms for storing the routes causes the traffic load to concentrate on certain nodes | ||
| Traffic load is unevenly distributed | ||
| Our Solution: route discovery based on perceived network load | ||
The Rationale for Traffic Balancing
| NS 2 Implementation (based on DSR) | |||
| Two parameters and one counter: | |||
| the medium usage threshold | |||
| the measurement time period | |||
| overThresholdCounter | |||
| Functions: | |||
| For the router: Upon receiving a route request, check medium usage in past measurement period. If over medium usage threshold, increase the overThresholdCounter by one. | |||
| For the sender: Upon receiving the route replies, the sender chooses the route with the smallest overThresholdCounter. If more than one path has the same smallest overThresholdCounter, the one with fewer hops is chosen. If more than one path has the same overThresholdCounter and the same number of hops, the sender randomly chooses one. | |||
| Traffic Balancing determines the medium state by the percentage of time that the measured signal strength is over the threshold. What is a good threshold? | |
| If constant threshold, other important issues such as node mobility and congestion state are not considered even though these issues also influence the medium state |
| The number of collisions that can be seen by a node over a certain period can indicate the node mobility or medium usage condition more precisely than other parameters | ||
| The faster nodes move, the larger number of collisions can be observed by nodes | ||
| The larger number of collision happens, the more backoff periods are and less medium usage is detected | ||
| Method | ||
| Nodes measure the medium usage and number of collisions over a certain period | ||
| Based on the number of collisions, a suitable threshold is set | ||
| Other traffic balancing protocols: collect traffic information from node and maybe neighbours: | ||
| DLAR: look at local queue length | ||
| LBAR: local queue length plus 1-hop neighbours | ||
| Common problems: interference range exceeds measured area | ||
| Traffic Balancing: find routes based on perceived network load | ||
| Work done in NS2, but we also have driver modifications for Atheros MADWiFi driver to collect information in live networks | ||
| Simulation results: increases End-to-End Capacity (higher packet delivery ratio, lower packet latency) | ||
| Compared to minimum-hop-count pro-active routing protocols (AODV, DSR) | ||
| Compared to other load balancing routing protocols | ||
| Compared to theoretical bounds: still room for improvement: | ||
| Loss due to control message overhead | ||
| Loss due to dropped packets | ||
| Suboptimal routes | ||