Efficient Power Management based on Application Timing Overview of - PDF document

Efficient Power Management based on Application Timing Overview of presentation Semantics for Wireless Sensor � Motivation for power management Networks � Power management techniques � ESSAT Octav Chipara, Chenyang Lu, � Workload model � Safe Sleep Gruia-Catalin Roman � Traffic shapers � Protocol maintenance Presented by Obi Orjih � Experiments/results For CSE 520S Obi Orjih CSE 520S 2 Why is power management Power management important? techniques � Energy is the most critical resource in � Three main categories of power management remotely deployed WSNs approaches [2] � Without power management, a Mica2 mote lasts a � Topology control few days before the batteries die � Control network layout to reduce transmission power while maintaining connectivity � Three basic requirements for a WSN power � Power-aware routing management protocol: � Control routing to reduce transmission power and duty � Maintain acceptable QoS (reliability, latency) cycle � Simplicity and minimal overhead due to platform � Sleep management constraints � Turns off nodes (radio/sensor/processor) when they are not needed � Adjust to variations in workload Obi Orjih CSE 520S 3 Obi Orjih CSE 520S 4 Related work Related work � SYNC/S-MAC [3] � PSM � MAC protocol with distributed synchronized � Power-saving mode option in 802.11 sleep schedule (SYNC packets) and protocol which adapts duty cycle based on contention management (RTS-CTS) perceived workload � Disadvantage: significant latency � Disadvantage: may interfere with application timing resulting in wasted � Note: schemes with centralized power and/or increased latency synchronization do not scale well Obi Orjih CSE 520S 5 Obi Orjih CSE 520S 6 1

Related work ESSAT overview � Span [1] � E fficient S leep S cheduling based on � Nodes make local A pplication T iming (ESSAT) decisions on whether to � Exploits application timing to meet power sleep or join/form management requirements communication backbone � Combination of Safe Sleep algorithm and � Disadvantage: nodes on backbone have short traffic shaper lifetime � Layered above the MAC Obi Orjih CSE 520S 7 Obi Orjih CSE 520S 8 Workload model Critiques of workload model � ESSAT assumes a generalized workload � ESSAT works only for the assumed workload model where sources produce data model, unlike more general MAC protocols periodically for queries � To be fair, this model is used in a very large class � Query flooded through the network at setup time of WSN applications � At time Φ the leaf nodes begin generating and � In practical applications non-leaf nodes may sending data reports with period P also generate data reports � Non-leaf nodes aggregate and route the data reports to the sink � Dynamic query generation is not supported – � NOTE - Network traffic is not periodic due to queries only generated at setup time multi-hop delay jitter and multiple queries with different timing properties Obi Orjih CSE 520S 9 Obi Orjih CSE 520S 10 Safe Sleep (SS) Safe Sleep � Local sleep scheduling algorithm for the � Terminology � r(q,k,c) – expected reception time of k th data radio report for query q from child c � s(q,k) – expected send time of k th data report for � 2 states: query q � Busy – expects to receive or send data � q.r next (c) – next reception time for query q from � Free – no data reports pending child c � q.s next – next send time for query q � Guarantees no energy or delay � T wakeup – the minimum of the expected reception penalties incurred by turning radio off and send times of all queries Obi Orjih CSE 520S 11 Obi Orjih CSE 520S 12 2

Safe Sleep Safe Sleep � Algorithm � Terminology updateNextRecei updateNex tReceive ve(q,c, r(q, k + 1, c))) { Update the next expected receive time � Run after sending or q.r next (c) with r(q, k + 1, c); � T BE – break-even time – the minimum time checkState(); } receiving data report updateNex updateNextSend tSend(q, s(q, k + 1)) { the node must be free to compensate the � If T sleep > T BE Update the next expected send time Go to sleep q.s next with s(q, k + 1); cost of entering an inactive state Else stay awake checkState(); } � Handles multiple checkState() { checkStat � When P TR ≤ P ON ∀ t wakeup = min({t | t = q.s next q} queries without TDMA ∀ U {t | t = q.r next (c) q,c}) T BE = T TR = T ON → OFF + T OFF → ON schedule t sleep = t wakeup - now; if (t sleep > t BE ) � When P TR > P ON � Storage cost sleep and set time to wake up at (t sleep - t OFF → ON );} T BE = T TR + T TR (P TR - P ON )/(P ON - P OFF ) proportional to degree of routing tree Obi Orjih CSE 520S 13 Obi Orjih CSE 520S 14 Safe Sleep Safe Sleep Example � SS achieves maximum sleep time when 2 queries, 1 child per query t = 1.99s t = 2s t = 0s t = 3s t = 3.01s t = 3.02s t = 3.03s t = 3.99s t = 4s t = 2.01s t = 2.02s t = 6.03s t = 6.05s t = 6s t = 2.99s t = 6.02s t = 6.01s t = 6.04s t = 4.01s t = 2.03s t = 5.99s t = 4.03s t = 4.02s expected and actual reception times coincide r(1,1,1) = 2s � Traffic shaper should ensure child expected s(1,1) = 2.01s r(1,k+1,1) = r(1,k,1) + 2s send time and parent expected receive time s(1,k+1) = s(1,k) + 2s are the same A A A r(2,1,2) = 3s Radio off � Data reports that are ready before expected s(2,1) = 3.01s Radio turning on/off r(2,k+1,2) = r(2,k,2) + 3s are buffered until the send time s(2,k+1) = s(2,k) + 3s Radio on � If data report is late, sleep time of receiver is T BE = T ON → OFF + T OFF → ON reduced because radio is kept on longer = 10ms + 10ms = 20ms 10 ms to send/receive Obi Orjih CSE 520S 15 Obi Orjih CSE 520S 16 No Traffic Shaper (NTS) NTS-SS � Nodes send aggregated data report to � Advantages T collect : time to receive data reports from all children � No delay penalty parent immediately after they have � Disadvantages received and aggregated children’s data T comp : time to aggregate data � Energy efficiency sub- reports optimal T agg = T collect + T comp � Power consumption of a � For all nodes, s(k) = r(k) = Φ + k * P d : rank in routing tree (0 for leaf node) node is dependent on � All nodes turn on their radios when the its rank in the routing T recv (d) = tree data report is generated for that period 0 , if d = 0 (d-1)*T agg + T collect , if d ≠ 0 � Nodes of higher rank � A node turns off its radio after sending its leave their radios on data report for that period longer, and therefore consume more power Obi Orjih CSE 520S 17 Obi Orjih CSE 520S 18 3

NTS-SS Example Static Traffic Shaper (STS) � Assign global deadline D and local deadline l 1 query 50ms t = 0s t = 50ms t = 40ms t = 30ms t = 20ms t = 10ms = D/M , where M is the maximum rank of the P = 1s A A Φ = 0s tree T BE = 0s 10 ms to send � For each query 10 ms to aggregate 40ms 50ms B B C C � r(k) = Φ + k*P + l*(d-1) Radio off � s(k) = Φ + k*P + l*d Radio on � As before, early data reports are buffered and 10ms 20ms 10ms 20ms D D E E F F G G late data reports are sent immediately Obi Orjih CSE 520S 19 Obi Orjih CSE 520S 20 STS-SS STS-SS Example � Tuning l , the critical T recv (l,d) = 1 query 0 , if d = 0 20ms t = 0s t = 40ms t = 50ms t = 20ms t = 30ms t = 10ms parameter, involves a P = 1s A A A (T agg -l) *(d-1) + T collect , if l ≤ T agg Φ = 0s trade-off between & d ≠ 0 T BE = 0s energy efficiency and T collect , if l > T agg & d ≠ 0 l = 30ms = T agg 10 ms to send latency 40ms 50ms 10 ms to aggregate B B C C � When l ≤ T agg the radios Radio off L q = M*max(l,T agg ) are turned on before Radio on children are ready � When l > T agg the 10ms 20ms 10ms 20ms children are ready to D D E E F F G G transmit on time Obi Orjih CSE 520S 21 Obi Orjih CSE 520S 22 Dynamic Traffic Shaper (DTS) DTS-SS � Initially, for all nodes s(0) = r(0) = Φ � Behaves similarly to Release Guard protocol � After a node receives data reports from all � Added synchronization improves energy efficiency children and sends its aggregated data � DTS adapts to the network workload by report: adjusting send and reception times based on longest multi-hop delay of received data � If report ready on time – t ≤ s(k) reports � Child sends at s(k) and sets s(k+1) = s(k) + P � Parent sets r(k+1) = r(k) + P � On average, overhead of piggybacked phase � If report is late – t > s(k) updates was shown in experiments to be less � Child sends immediately and sets s(k+1) = t + P than one bit per data report at all tested � Child indicates phase shift by piggybacking s(k+1) in query rates data report � Parent sets r(k+1) = s(k+1) Obi Orjih CSE 520S 23 Obi Orjih CSE 520S 24 4