Mobile Movement Patterns and Applications in Wireless Networks

Abstract

In a real-life wireless network, the logical movements of mobile nodes are not purely random. Movements of individual mobile nodes have intrinsic patterns determined by regular activities of individual persons. Regularities in people's group activities also introduce patterns in co-location behavior of multiple mobile nodes. Mobile nodes are able to predict their future behavior using history information, and prediction results can be used to expedite network management processes and reduce the required overhead. Our research focus on characterization and applications of movement and co-location patterns. echanism with movement prediction for wireless IP networks. Each mobile node records movement history information, and predicts its next subnet before the actual movement. It explicitly notifies the current foreign agent to duplicate and forward packets to the predicted subnet. Simulation with real-life wireless network trace shows that the latency of network-layer handoff and the amount of packet loss are greatly reduced, only with a limited overhead in packet duplication and forwarding.

The topology matching issue for mobile peer-to-peer networks is also investigated, and a Local Topology Cache mechanism is designed to expedite topology matching for overlay topology optimization and reduce the associated overhead. As mobile nodes have patterns in their movement and interaction, the physical network topology nearby might be similar for a mobile node's two consecutive visits to a subnet. The mobile node caches the information of topologically matched P2P neighbors and reuses them when returning to the subnet, without probing the network again. We simulate this scheme with a real-life wireless network trace, and found the caching mechanism can greatly reduce network probing overhead, while achieving similar efficiency of P2P overlay topology.

We further investigate the co-location behavior of multiple mobile nodes. People's regular interactions determine that co-location of mobile nodes has regularities. Using real-life wireless network traces, we measure the characteristics of mobile nodes' co-location, and show that co-location has patterns and is repetitive, which provides the basis of co-location prediction. A Markov-family model is used to dynamically model the co-location behavior, and a fully distributed co-location prediction method only using a mobile node's own movement trace and co-location history is proposed. The effectiveness of this co-location prediction method is demonstrated with simulations based on real-life wireless network activity traces. We also utilize the co-location prediction method in the construction of the peer-to-peer overlay in a wireless network, and show that it can construct a peer-to-peer overlay as efficient as topology matching techniques, without probing the physical network. This demonstrates that co-location prediction can indeed expedite network management and reduce the associated overhead.