Energy-efficient distributed algorithms for wireless ad hoc networks Costas Busch Louisiana State University CCW’08
Energy in distributed algorithms • Becomes an issue when designing algorithms • The output of the algorithms may affect the energy efficiency
Typical assumptions • Computation power at each node is abundant • Unlimited energy for computations at each node • Computation time at each node does not affect total time complexity • Point to point communication • Messages in local neighborhood can be sent simultaneously • Message delivery dominates time complexity
Typical assumptions continued… • The network is reliable • The network topology does not change • The messages are delivered as expected • Global Synchronization • All nodes can synchronize • A special node initiates the algorithm • The algorithm runs only once • One shot problems
Wireless ad hoc network restrictions • Computation power is limited • Communication is not point-to-point • Requires more energy due to channel interference • The network is unreliable (ad hoc, mobility) • More energy to transfer messages • Global synchronization is not easy • More messages, energy to achieve synchronization • An algorithm may run forever • It continuously consumes energy
Rethink distributed algorithm design • Consider energy consumption when designing algorithms • Do not make strong assumptions • Design algorithms with: • Smaller computation at each node • Low message complexity • Self-stabilizing • Local • Online
Performance metrics Classic metrics: • Number of messages • Total time New metrics: • Max, Average utilization of the nodes • Combination of the above metrics • Number of Messages X Total Time? • What are realistic metrics of performance?
Algorithm’s output affects energy efficiency • Topology Control • Focuses on obtaining sparse connected spanners, • But what is the effect on load balancing? • Routing • Focuses on just obtaining routing paths, • But what is the effect on congestion?
Peer-to-peer • Focus on uniformly distributing and accessing the data • But what about the actual node utilization and actual network paths? • Data aggregation • Focus on minimizing the total aggregation cost, • But how does this affect the max cost at a node? • Facility location • Focus on path distances • But how about the load on each facility?
Simplifications are good, but how realistic are they? • How frequently do uniform disc graphs appear in practice? • Can we afford to ignore maximum node utilization? • Is the computation power at each node abundant?