노상훈@pllab.kut 2009. 04. 27 PLLAB@KUT Introduction CGS Algorithm Fault Tolerance of The CGS Algorithm Result Summary ◦ ◦ ◦ ◦ Background Assumption QoS Metrics Algorithm PLLAB@KUT 1/16 Purpose Random approach Coordinated approach ◦ Energy conservation -> Prolong network lifetime ◦ Assign probability ◦ Awake/sleep randomly on probability ◦ Cannot ensure k-coverage ◦ ◦ ◦ ◦ ◦ Use “drowsiness factor” for scheduling Centralized approach Distributed approach Ensure k-coverage if physically possible Cost: network overhead and energy consumption PLLAB@KUT 2/16 CGS:Controlled greedy sleep Graph: G(R∪S,E) k-Coverage problem Minimal k-Coverage problem ◦ R:Range ◦ S:Sensors ◦ E:Edges ◦ subgraph problem ◦ G’(R∪S’,E’) where S’ ⊆ S ◦ Nonredundant with G’’(R∪S’’,E’’) where S’’ ⊆ S PLLAB@KUT 3/16 Centralized solution ◦ A Coordinated node knows whole graph G Distributed solution ◦ Each node q knows itself and it’s neighbors and covers Rq ◦ Gq(Rq∪Sq,Eq) PLLAB@KUT 4 /16 Assumption 1 ◦ communication radius ≥ 2*sensing radius Assumption 2 ◦ Coverage of each sensor modeled by a sensing disk and the radius is B ◦ Approximation of the sensing disk by a set of squares PLLAB@KUT 5 /16 Assumption 3 ◦ The sensors know their own coordinates and the observed area Σ ◦ Reflect Maximum location error Δl PLLAB@KUT 6 /16 ◦ k-coverage ratio ◦ Ak: area of the k-coverage regions ◦ A: Area of the target space Σ ◦ Approximation of ◦ Nk: number of the k-covered regions ◦ N: total number of regions in the target space k-lifetime of a network ◦ Maximum operation time of the network with PLLAB@KUT 7 /16 Drowsiness factor ◦ Energy status ◦ importance in the network Coverage ratio of region r ◦ cr : degree of region r in Gs ◦ positive if overcovered PLLAB@KUT 8 /16 Whole sensors awaken Drowsiness Factor Decision Time Delay Awake Message List of awake neighbors Delay List PLLAB@KUT 9 /16 Kind of Messages ◦ Hello messages ◦ DTD messages ◦ AMs Messages can be lost ◦ Collision ◦ Fading ◦ External disturbances Assume ◦ Any messages can be lost (worst case whole message) ◦ but received message is correct (error detecting coding) PLLAB@KUT 10 /16 Theorem: If physically possible, the algorithm will provide Lost Hello Messages Lost DTD messages Lost AMs Lost messages cause overcoverage ◦ Prevent inclusion of neighbors to in the alive neighbor set Ss → drowsiness factor will be unnecessarily high ◦ The DTD is higher → harmful ◦ Sender is not considered as a potential participant in the coverage ◦ Potential overcoverage ◦ Cannot rely on the presence of the senders of them ◦ Unnecessarily high number of sensors staying awake ◦ Shortening of the network lifetime PLLAB@KUT 11 /16 Effect of node failure can be divide by steps During an awake period During a sleep period: No effect During the Hello phase During the DTD phase (problem!) Node failure generally cause ◦ Some regions will undercovered until the end of the period ◦ Next election will provide sufficient coverage ◦ CGS behavior is equivalent to the loss of Hello messages ◦ Node with higher drowsiness factor may incorrectly rely on the presence of the failed node ◦ Shortening of the lifetime of the network ◦ Possible coverage in certain regions and total amount of energy in the network decrease PLLAB@KUT 12 /16 A out of synchronized node ◦ Wakes up while the synchronization phase is running Synchronize again ◦ Cannot synchronize Provide extra coverage in its neighborhood To ensure proper operation, Small timing errors have no effect In general, ◦ The synchronization period must be long enough ◦ To enable all nodes to wake up and join the synchronization ◦ On Hello/DTD messages ◦ But AM’s timing error might change priority ◦ Large timing errors cause large message delays(ultimately loss) ◦ But the algorithm is very tolerant in this respect ◦ Do not affect the provided k-coverage, still shorten the network lifetime PLLAB@KUT 13 /16 PLLAB@KUT 14 /16 Distributed algorithm proposed ◦ Solve k-coverage problems ◦ Provide prolong network time Proposed algorithm broadcast few messages CGS algorithm is robust and fault tolerant CGS algorithm is better than random k-coverage algorithm ◦ Guarantees required coverage if possible ◦ Degrading curve is much gentler PLLAB@KUT 15 /16 G. Simon, M. Molnar, L. Gonczy, B. Cousin, Robust k-Coverag e Algorithms for Sensor Networks, IEEE Transactions on Instru mentation and Measurement, Vol. 57, No. 8, pp. 1741-1748, Aug. 2008. PLLAB@KUT 16 /16 See Theorem 1 PLLAB@KUT 17 Back PLLAB@KUT 18 Back PLLAB@KUT 19
© Copyright 2026 Paperzz