rP b c rP r P rP r P tx idle tx rx idle rx idle a - PDF document

Announcement • Start at 9am tomorrow. • Short presentations on your research. Minimum Power Configuration Chenyang Lu Department of Computer Science and Engineering Washington University in St. Louis 1 Understanding Radio Power Cost Example: Minimizing Total Radio Energy c Transmission ( P tx ) Reception ( P rx ) Idle ( P idle ) Sleeping ( P sleep ) Radio States • a sends to c at normalized rate of Power consumption 21.2~106.8 32 32 0.001 (mw) r = Data Rate / Band Width Power consumption of CC1000 Radio in different states • Source and relay nodes remain active b • Configuration 1: a → b → c • Sleeping consumes much less power than idle listening • Configuration 2: a → c, b sleeps – Motivate sleep scheduling [Polastre et al. 04, Ye et al. 04] a • Transmission consumes most power – Motivate transmission power control [Singh et al. 98,Li et al. 01,Li and Hou 03] • None of existing schemes minimizes the total energy consumption in all radio states 3 4 Average Power Consumption Power Control vs. Sleep Scheduling c Transmission power • Configuration 1: a → b → c → Consumption P ( a c ) dominates: use low Power transmission power → → = + − + + + − + + − ( ) ( , ) ( 1 ) ( , ) ( 1 2 ) ( 1 ) P a b c rP � a b r P � rP b c rP r P rP r P � tx idle tx rx idle rx idle a ’s avg. power b ’s avg. power c ’s avg. power → → ( ) P a b c b 3P idle rx a 2P idle +P sleep idle b ’s activity time tx data rate r 0 1 band width • Configuration 2: a → c, b sleeps Idle power dominates: → = + − + + + − ( ) ( , ) ( 1 ) ( 1 ) P a c rP a c r P P rP r P use high transmission power tx idle sleep rx idle 5 since more nodes can sleep 6 1

Minimum Power Configuration (MPC) Minimum Power Configuration (MPC) • a transmits to c at rate r , b sleeps • Given traffic demands I={( s i , t i , r i )} and → = + + − + ignore sleeping power P ( a c ) rP ( a , c ) rP 2 ( 1 r ) P P G(V,E) , find a sub-graph G ´ (V ´ , E ´ ) minimizing tx rx idle sleep = + + − 2 P r ( P ( a , c ) P 2 P ) group rate related terms idle tx rx idle = + ⋅ ∑ ∑ 2 P r C C a,c =P tx (a,c)+P rx -2P idle + | | V | V | ' ' idle a , c P P r r ( ( s s t t ) ) P P , , idle idle i i i i i i ∈ I ∈ I edge (a,c) has a cost ( ( s s , , t t , , r r ) ) i i i i i i each active node of C a,c per unit of data sum of edge cost independent of has a cost of P idle P idle node cost from s i to t i in G ´ data rate! r · C a,c c • Sleep scheduling • Sleep scheduling • Sleep scheduling P idle • Assumed total w orkload • Power control • Power control < bandwidth a rate r b • MPC is NP-Hard 7 8 Solutions Incremental Shortest-path Tree Heuristic (ISTH) • Existing centralized algorithm • Initially, all nodes are labeled as asleep – Matching Based Algorithm (MBA) can solve the one-sink case of MPC with a pprox. ratio O(lgk) [ Meyerson et al. 00] • For each traffic demand ( s i , t i , r i ) • Distributed implementation is expensive – Find the shortest path from s i to t i under cost • Cannot handle dynamic data flows functions H e (u,v, r i ) and H n (u) • Developed two new distributed and online algorithms – Incremental Shortest-path Tree Heuristic – Label all nodes on the found path as active • Known approx. ratio is O(k), similar average-case performance as MBA = ⋅ H ( u , v , r ) r C • Adapt to dynamic network workloads and different radio Edge cost: e i i u , v characteristics ⎧ – Minimum Steiner Tree Heuristic P u is asleep = Node cost: idle ( ) ⎨ H u Approx. ratio is 1.5(P rx +P tx -P idle )/P idle ( ≈ 5 on Mica2 motes) • n ⎩ 0 u is active 9 10 Illustration of ISTH Properties of ISTH • Online, distributed implementation is easy • C u,v =2 , P idle =1 sink • Find a new path in each iteration • Known approx. ratio is k , num of sources 1 0 2 cost reduction! 2 • Performance for special cases 1 – Approx. ratio is 2 when r = 0 1 1 0 • Good performance when data rates are low 2 2 New cost : 2 – Optimal when P idle =0 1 × 3 + 0.2 × 2 × 2=3.8 2 1 1 1 2 2 Source 2 1 + 0.2 × 2 × 2=1.8 1 0 1 r 2 = 0.2 Source 1 r 1 = 0.2 11 12 2

Minimum Power Configuration Simulation Environment Protocol (MPCP) • Prowler simulator extended by Rmase project • Routing – Prowler: http://www.isis.vanderbilt.edu/projects/nest/prowler/ – Extends DSDV with routing metrics H e and H n – Rmase: http://www2.parc.com/spl/projects/era/nest/Rmase/ • Sleep scheduling • Implemented USC model [Zuniga et al. 04] to simulate – Turns on radio if on a route, runs a duty cycle otherwise lossy links of Mica2 motes • Power control • Baseline protocols: – Determines transmission power to different neighbors – MT: Extended DSDV that minimizes num of Txs • Cross-layer optimization – MTP: Extended DSDV that minimizes Tx Power – Data rate, transmission power, and state of the node • Data rate per flow: 0.3 Kbps, 100 nodes jointly determines the routing cost 13 14 Network Energy Consumption Delivery Rate and Delay Energy Cost of Non-Source Nodes (J) Energy Cost of All Nodes (J) Delivery rate Packet delay Energy cost of all nodes Energy cost of non-source nodes MPCP causes slightly higher network MPCP saves up to 30% energy MPCP saves up to 80% energy contention due to path reuse 15 16 Summary Reference • Minimum Power Configuration: minimize total power of • G. Xing, C. Lu, Y. Zhang, Q. Huang and R. Pless, Minimum Power wireless sensor networks Configuration for Wireless Communication in Sensor Networks, – Sleep scheduling + minimum power routing ACM Transactions on Sensor Networks, 3(2), June 2007. Extended – Adaptive to workload version of MobiHoc'05 paper. • MPCP: Efficient online protocols • Simulations based on MICA2: MPCP saves more energy than min-power routing and shortest path routing. 17 18 3