Fast Decentralized Averaging Via Multi-Scale Gossip

Date Added: Apr 2010
Format: PDF

Applications in sensor networks often demand that nodes cooperatively accomplish a task without centralized coordination. Autonomy is often equated with robustness and scalability in large-scale networked systems. This is especially true in wireless networks, where fundamental limits on spatial and temporal channel reuse limit the amount of communication possible at any instant in time. Moreover, when nodes are battery-powered - a typical design element of wireless sensor networks - each transmission consumes valuable energy resources. This has stimulated research into resource-efficient algorithms for distributed computing and coordination. Gossip algorithms are an attractive paradigm for decentralized, autonomous computation.