Maintaining High-bandwidth Communication Between Mobile Groups of Nodes Kamin Whitehouse, Ying Zhang UC Berkeley, Palo Alto Research Center Motivation Objects or events in a sensor network will often be identified by groups of nodes, whose locations and member nodes will change as the object/event moves. 13 10 15 14 12 11 9 5 7 6 8 As the groups move, the current route becomes invalid. How can we repair this route without consuming the bandwidth of the route we are trying to maintain? 1.Route Discovery 4. Route Capture Route discovery is expensive. The source broadcasts to the entire network and the destination replies, forming a reverse route along the broadcast tree. Because this algorithm maintains an explicit route, it is extremely fragile to node/link failure. To solve this problem, the route captures nearby nodes in the case of a single node failure. 2. Route Append 13 10 When the leader of a group changes, the old route is no longer valid. Assuming each successive leader is within communication radius of the old leader (ie. there is a clean handoff) the new leader’s parent is simply set to be the old leader. 13 2 10 4 15 14 For groups to communicate, a route must be maintained between the two group leaders. However, as the group leaders change this may require message passing and data propagation, which consumes valuable bandwidth. Our goal is to maintain a route between two mobile groups with a minimal number of maintenance messages. 12 11 9 5 7 6 8 3 Because the new leader is always within communication range of the old leader, it reuses the old route by appending a link to the old leader, node number 9. This creates a suboptimal route. 2 1 4 3. Route Pruning Overall Idea Assuming that the group leaders are elected such that each successive leader is within communication range of the previous leader, we can simply append a new leg onto the existing route. By adding more state to each node indicating where it is on the route, a node further down the route can usurp a message from the beginning of the route if it can hear it. This eliminates any extraneous nodes by shortcutting them. 13 13 10 15 14 12 11 9 5 ? 7 6 3 ? 8 ? ? 2 1 As the orange objects moves, 8 becomes the new leader of the orange group. In order to maintain connectivity with 1, the new leader can try to connect with the old route. 15 14 12 11 17 10 15 14 12 11 9 X 5 X 7 6 Node 8 can be pruned from the new route by node 7, who knows that it is downstream but can still overhear node 9’s messages. 8 16 3 2 1 4 Note that all candidates are almost guaranteed to be within radio range of each other and that the receiving nodes can break ties by multiple volunteers. 5. Route Optimization This routing algorithm is far from optimal in terms of finding shortest path routes. While pruning takes care of many sub-optimalities, it is still possible to use unnecessarily long paths. 13 10 15 14 12 11 9 5 7 2 1 8 4 3 1. Loops 1. Loops: prune the cycle if a leader moves in a circle 2. Long routes 2. Turns: prune redundant path if a leader back-tracks 3. Baby steps: prune redundant nodes if multiple successive nodes are within communication range, (perhaps caused by the sensing range being much smaller than the communication range). When node 20 dies, the link between 9 and 5 is broken. Those nodes that can hear both 5 and 9 are candidates to replace 20, and self-elect themselves when they see that 20 is not responding. Each components of this algorithm besides route discovery are done implicitly, not requiring extra messages but instead utilizing eavesdropping and maintaining state. While this makes the route fragile, this is ameliorated in part by having nodes that neighbor the route in the network also maintain state. In total, only a single extra message is required for route maintenance after the route is initially discovered. In practice, this allows an order of magnitude higher bandwidth between the groups than competitive methods; the bandwidth of the leaders is not consumed by messages from their neighbours. All nodes that can hear a blocked message and that can forward the message to other half of the route are candidates to be captured. Nodes randomly elect themselves to be in the route and forward the message, thereby suppressing all other candidates from joining the route. 3 This simple algorithm runs into several problems: We will try to solve these without using extra bandwidth 8 19 6 This technique automatically prunes: 4. Redundant Links 7 6 4 3. Broken Links 9 20 X 5 18 3 1 Analysis Conclusion This algorithm was developed for the NEST pursuer/evader game Mid-term demo. One main problem with the demo was the large amount of network traffic required for each tracking update. By using a higher-bandwidth routing algorithm, the network can provide more tracking updates faster. Without sacrificing bandwidth, nodes far from source and destination can optimize the route during quite periods, thereby finding non-local shortcuts. 2 1 4 Nodes not near the source or destination can propagate a cost-to-go estimate to other nodes in the network, similar to the distributed Bellman-ford algorithm. By restricting this to the center of the route while the route is inactive, we can optimize for path length while not reducing bandwidth. Contact Kamin Whitehouse [email protected] 13 10 15 14 12 11 9 5 7 6 8 3 2 1 4
© Copyright 2026 Paperzz