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

SLIDE 1

1

Efficient Power Management based on Application Timing Semantics for Wireless Sensor Networks

Octav Chipara, Chenyang Lu, Gruia-Catalin Roman

Presented by Obi Orjih For CSE 520S

Obi Orjih CSE 520S 2

Overview of presentation

Motivation for power management Power management techniques ESSAT

Workload model Safe Sleep Traffic shapers Protocol maintenance Experiments/results Obi Orjih CSE 520S 3

Why is power management important?

Energy is the most critical resource in

remotely deployed WSNs

Without power management, a Mica2 mote lasts a

few days before the batteries die

Three basic requirements for a WSN power

management protocol:

Maintain acceptable QoS (reliability, latency) Simplicity and minimal overhead due to platform

constraints

Adjust to variations in workload Obi Orjih CSE 520S 4

Power management techniques

Three main categories of power management

approaches[2]

Topology control

Control network layout to reduce transmission power

while maintaining connectivity

Power-aware routing

Control routing to reduce transmission power and duty

cycle

Sleep management

Turns off nodes (radio/sensor/processor) when they are

not needed

Obi Orjih CSE 520S 5

Related work

SYNC/S-MAC[3]

MAC protocol with distributed synchronized

sleep schedule (SYNC packets) and contention management (RTS-CTS)

Disadvantage: significant latency

Note: schemes with centralized

synchronization do not scale well

Obi Orjih CSE 520S 6

Related work

PSM

Power-saving mode option in 802.11

protocol which adapts duty cycle based on perceived workload

Disadvantage: may interfere with

application timing resulting in wasted power and/or increased latency

SLIDE 2

2

Obi Orjih CSE 520S 7

Related work

Span[1]

Nodes make local

decisions on whether to sleep or join/form communication backbone

Disadvantage: nodes on

backbone have short lifetime

Obi Orjih CSE 520S 8

ESSAT overview

Efficient Sleep Scheduling based on

Application Timing (ESSAT)

Exploits application timing to meet power

management requirements

Combination of Safe Sleep algorithm and

traffic shaper

Layered above the MAC Obi Orjih CSE 520S 9

Workload model

ESSAT assumes a generalized workload

model where sources produce data periodically for queries

Query flooded through the network at setup time At time Φ the leaf nodes begin generating and

sending data reports with period P

Non-leaf nodes aggregate and route the data

reports to the sink

NOTE - Network traffic is not periodic due to

multi-hop delay jitter and multiple queries with different timing properties

Obi Orjih CSE 520S 10

Critiques of workload model

ESSAT works only for the assumed workload

model, unlike more general MAC protocols

To be fair, this model is used in a very large class

• f WSN applications

In practical applications non-leaf nodes may

also generate data reports

Dynamic query generation is not supported –

queries only generated at setup time

Obi Orjih CSE 520S 11

Safe Sleep (SS)

Local sleep scheduling algorithm for the

radio

2 states:

Busy – expects to receive or send data Free – no data reports pending

Guarantees no energy or delay

penalties incurred by turning radio off

Obi Orjih CSE 520S 12

Safe Sleep

Terminology

r(q,k,c) – expected reception time of kth data

report for query q from child c

s(q,k) – expected send time of kth data report for

query q

q.rnext(c) – next reception time for query q from

child c

q.snext – next send time for query q Twakeup – the minimum of the expected reception

and send times of all queries

SLIDE 3

3

Obi Orjih CSE 520S 13

Safe Sleep

Terminology

TBE – break-even time – the minimum time

the node must be free to compensate the cost of entering an inactive state

When PTR ≤ PON

TBE = TTR = TON→OFF + TOFF→ON

When PTR > PON

TBE = TTR + TTR(PTR - PON)/(PON - POFF)

Obi Orjih CSE 520S 14

Safe Sleep

Algorithm

Run after sending or

receiving data report

If Tsleep > TBE

Go to sleep Else stay awake

Handles multiple

queries without TDMA schedule

Storage cost

proportional to degree

• f routing tree

updateNex updateNextRecei tReceive ve(q,c, r(q, k + 1, c))) { Update the next expected receive time q.rnext(c) with r(q, k + 1, c); checkState(); } updateNex updateNextSend tSend(q, s(q, k + 1)) { Update the next expected send time q.snext with s(q, k + 1); checkState(); } checkStat checkState() { twakeup = min({t | t = q.snext q}

U {t | t = q.rnext(c) q,c})

tsleep = twakeup - now; if (tsleep > tBE) sleep and set time to wake up at (tsleep - tOFF→ON);}

∀ ∀

Obi Orjih CSE 520S 15

Safe Sleep

SS achieves maximum sleep time when

expected and actual reception times coincide

Traffic shaper should ensure child expected

send time and parent expected receive time are the same

Data reports that are ready before expected

are buffered until the send time

If data report is late, sleep time of receiver is

reduced because radio is kept on longer

Obi Orjih CSE 520S 16

Safe Sleep Example

2 queries, 1 child per query r(1,1,1) = 2s s(1,1) = 2.01s r(1,k+1,1) = r(1,k,1) + 2s s(1,k+1) = s(1,k) + 2s r(2,1,2) = 3s s(2,1) = 3.01s r(2,k+1,2) = r(2,k,2) + 3s s(2,k+1) = s(2,k) + 3s TBE = TON→OFF + TOFF→ON = 10ms + 10ms = 20ms 10 ms to send/receive

A t = 0s t = 1.99s t = 2s t = 2.01s t = 2.02s t = 2.03s t = 2.99s A t = 3s t = 3.01s t = 3.02s t = 3.03s t = 3.99s t = 4s

Radio off Radio turning on/off Radio on

t = 4.01s t = 4.02s t = 4.03s t = 5.99s A t = 6s t = 6.01s t = 6.02s t = 6.03s t = 6.04s t = 6.05s

Obi Orjih CSE 520S 17

No Traffic Shaper (NTS)

Nodes send aggregated data report to

parent immediately after they have received and aggregated children’s data reports

For all nodes, s(k) = r(k) = Φ + k * P

All nodes turn on their radios when the

data report is generated for that period

A node turns off its radio after sending its

data report for that period

Obi Orjih CSE 520S 18

NTS-SS

Advantages

No delay penalty

Disadvantages

Energy efficiency sub-

• ptimal

Power consumption of a

node is dependent on its rank in the routing tree

Nodes of higher rank

leave their radios on longer, and therefore consume more power

Tcollect : time to receive data reports from all children Tcomp : time to aggregate data Tagg = Tcollect + Tcomp d : rank in routing tree (0 for leaf node) Trecv(d) = 0, if d = 0 (d-1)*Tagg + Tcollect , if d ≠ 0

SLIDE 4

4

Obi Orjih CSE 520S 19

NTS-SS Example

1 query P = 1s Φ = 0s TBE = 0s 10 ms to send 10 ms to aggregate

A B D E G C F t = 0s t = 10ms t = 20ms t = 30ms t = 40ms t = 50ms

10ms 10ms 20ms 20ms 40ms 50ms 50ms

A B C D E F G

Radio off Radio on Obi Orjih CSE 520S 20

Static Traffic Shaper (STS)

Assign global deadline D and local deadline l

= D/M, where M is the maximum rank of the tree

For each query

r(k) = Φ + k*P + l*(d-1) s(k) = Φ + k*P + l*d

As before, early data reports are buffered and

late data reports are sent immediately

Obi Orjih CSE 520S 21

STS-SS

Tuning l, the critical

parameter, involves a trade-off between energy efficiency and latency

When l ≤ Tagg the radios

are turned on before children are ready

When l > Tagg the

children are ready to transmit on time

Trecv(l,d) = 0, if d = 0 (Tagg-l) *(d-1) + Tcollect , if l ≤ Tagg & d ≠ 0 Tcollect , if l > Tagg & d ≠ 0 Lq = M*max(l,Tagg)

Obi Orjih CSE 520S 22

STS-SS Example

1 query P = 1s Φ = 0s TBE = 0s l = 30ms = Tagg 10 ms to send 10 ms to aggregate

A B D E G C F t = 0s t = 10ms t = 20ms t = 30ms t = 40ms t = 50ms

10ms 10ms 20ms 20ms 40ms 50ms 20ms

A B C D E F G A

Radio off Radio on Obi Orjih CSE 520S 23

Dynamic Traffic Shaper (DTS)

Initially, for all nodes s(0) = r(0) = Φ After a node receives data reports from all

children and sends its aggregated data report:

If report ready on time – t ≤ s(k)

Child sends at s(k) and sets s(k+1) = s(k) + P Parent sets r(k+1) = r(k) + P

If report is late – t > s(k)

Child sends immediately and sets s(k+1) = t + P Child indicates phase shift by piggybacking s(k+1) in

data report

Parent sets r(k+1) = s(k+1)

Obi Orjih CSE 520S 24

DTS-SS

Behaves similarly to Release Guard protocol

Added synchronization improves energy efficiency

DTS adapts to the network workload by

adjusting send and reception times based on longest multi-hop delay of received data reports

On average, overhead of piggybacked phase

updates was shown in experiments to be less than one bit per data report at all tested query rates

SLIDE 5

5

Obi Orjih CSE 520S 25

DTS-SS Example

1 query P = 1s Φ = 0s TBE = 0s 10 ms to send 10 ms to aggregate

A B D E G C F t = 0s t = 10ms t = 20ms t = 30ms t = 40ms t = 50ms

10ms 10ms 20ms 20ms 40ms 50ms 50ms

A B C D E F G

Radio off Radio on

t = 1s t = 1.01s t = 1.02s t = 1.03s t = 1.04s t = 1.05s

10ms

D E F G B C A D E F G B C A

10ms 20ms 20ms 40ms 50ms 20ms Regular message Phase shift Obi Orjih CSE 520S 26

Protocol maintenance

Transient packet loss

NTS

No action – send and receive times don’t change

STS

No action – send and receive times don’t change

DTS

If no phase update after missed packet, receiver can

request phase update (through ACK or new packet)

Receiver can tolerate loss of multiple consecutive phase

updates, but at the cost of wasted energy

Obi Orjih CSE 520S 27

Protocol maintenance

Topology changes (node failure)

Parent removes dependencies on failed node For child,

NTS No action – send and receive times don’t change STS If rank changes, node and descendants must recompute

send and receive times according to new rank

DTS No action – protocol dynamically resynchronizes

Critique: STS-SS rank recalculation should be

performed for new parent and its ancestors

Obi Orjih CSE 520S 28

Protocol maintenance

Selecting timeout values

NTS

tTO(d) = (d + 1)*D/M

STS

tTO(d) = s(k) + l – tTO , tTO is constant

DTS

tTO(d) = maxc(s(k,c)) + tTO , tTO is tunable

Obi Orjih CSE 520S 29

Experimental setup

ns-2 simulation environment 80 nodes 500 x 500m2 area 125m communication range IEEE 802.11b MAC with 1 Mbps bandwidth 52 byte packets Query ratio rates 6:3:2 Queries start randomly between 0 and 10s Experiments last 200s

Obi Orjih CSE 520S 30

Experimental setup

First scenario

Single query per class, 1Hz to 5Hz base rate

Second scenario

Fixed 0.2 Hz base rate, # queries increased

Experimental assumption: all aggregated data

reports fit in a single data packet

Critique: this assumption may not be realistic,

especially in real WSN, and does not scale well

SLIDE 6

6

Obi Orjih CSE 520S 31

Energy efficiency

STS-SS and DTS SS have the best performance in terms of energy efficiency/duty cycle.

Obi Orjih CSE 520S 32

Query performance

NTS-SS, Span, and DTS-SS have lowest query latency Latency performance of STS-SS improves with

increased query rate

Local deadline configured to equal the period of query

Obi Orjih CSE 520S 33

Break-even time

DTS-SS has the highest percentage of short sleep

periods when TBE = 0

Therefore, DTS-SS is most affected by change in break-even

time

Energy efficiency of DTS-SS decreases as TBE

increases

Obi Orjih CSE 520S 34

Conclusions

ESSAT protocols take advantage of

application timing to conserve energy

Safe Sleep ensures sleep schedule of radio

does not waste energy or delay response

Traffic shapers have varying performance DTS-SS dynamically adjusts to the network

workload

Duty cycle 38-87% lower than Span Query latency 36-98% lower than PSM and SYNC Obi Orjih CSE 520S 35

Additional references

1.

• B. Chen, K. Jamieson, H. Balakrishnan, R. Morris. Span: An Energy-

Efficient Coordination Algorithm for Topology Maintenance in Ad Hoc Wireless Networks. MobiCom 2001.

2.

• G. Xing, C. Lu, Y. Zhang, Q. Huang, R. Pless. Minimum Power

Configuration in Wireless Sensor Networks. MobiHoc 2005.

3.

• W. Ye, J. Heidemann, and D. Estrin. An Energy-Efficient MAC

Protocol for Wireless Sensor Networks. INFOCOM 2002.