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

SLIDE 1

1

1

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

3

Understanding Radio Power Cost

• Sleeping consumes much less power than idle listening

– Motivate sleep scheduling [Polastre et al. 04, Ye et al. 04]

• 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

0.001 32 32 21.2~106.8

Power consumption (mw) Idle (Pidle) Sleeping (Psleep) Reception (Prx) Transmission (Ptx) Radio States

Power consumption of CC1000 Radio in different states

4

Example: Minimizing Total Radio Energy

• a sends to c at normalized rate of

r = Data Rate / Band Width

• Source and relay nodes remain active
• Configuration 1: a → b → c
• Configuration 2: a →c, b sleeps

a c b

5

idle rx sleep idle tx

P r rP P P r c a rP c a P ) 1 ( ) 1 ( ) , ( ) ( − + + + − + = →

Average Power Consumption

idle rx idle rx tx idle tx

P r rP P r rP c b rP P r b a rP c b a P ) 1 ( ) 2 1 ( ) , ( ) 1 ( ) , ( ) ( − + + − + + + − + = → →

a b c

a’s avg. power c’s avg. power b’s avg. power

b’s activity

tx rx idle

• Configuration 1: a → b → c
• Configuration 2: a → c, b sleeps

time

6

Power Control vs. Sleep Scheduling

Transmission power dominates: use low transmission power Idle power dominates: use high transmission power since more nodes can sleep

) ( c a P → ) ( c b a P → →

3Pidle 2Pidle+Psleep

Power Consumption width band rate data

r0 1

SLIDE 2

2

7

Minimum Power Configuration (MPC)

sleep idle rx tx

P P r rP c a rP c a P + − + + = → ) 1 ( 2 ) , ( ) ( a b c ) 2 ) , ( ( 2

idle rx tx idle

P P c a P r P − + + =

• a transmits to c at rate r, b sleeps

edge (a,c) has a cost

• f Ca,c per unit of data

each active node has a cost of Pidle r · Ca,c Pidle Pidle

c a idle

C r P

,

2 ⋅ + =

ignore sleeping power group rate related terms Ca,c=Ptx(a,c)+Prx-2Pidle

rate r

• Assumed total workload

< bandwidth

8

Minimum Power Configuration (MPC)

• Given traffic demands I={( si , ti , ri )} and

G(V,E), find a sub-graph G´(V´, E´) minimizing

• Sleep scheduling

+

∑

∈I r t s i i i

i i i

t s P r

) , , (

) , (

idle

P V | ' |

idle

P V | ' |

∑

∈I r t s i i i

i i i

t s P r

) , , (

) , (

sum of edge cost from si to ti in G´ independent of data rate!

• Sleep scheduling
• Power control
• Sleep scheduling
• Power control
• MPC is NP-Hard

node cost

9

Solutions

• Existing centralized algorithm

– Matching Based Algorithm (MBA) can solve the one-sink case of MPC with approx. ratio O(lgk) [Meyerson et al. 00]

• Distributed implementation is expensive
• Cannot handle dynamic data flows
• Developed two new distributed and online algorithms

– Incremental Shortest-path Tree Heuristic

• Known approx. ratio is O(k), similar average-case

performance as MBA

• Adapt to dynamic network workloads and different radio

characteristics – Minimum Steiner Tree Heuristic

• Approx. ratio is 1.5(Prx+Ptx-Pidle)/Pidle (≈ 5 on Mica2 motes)

10

Incremental Shortest-path Tree Heuristic (ISTH)

• Initially, all nodes are labeled as asleep
• For each traffic demand (si, ti, ri)

– Find the shortest path from si to ti under cost functions He(u,v, ri) and Hn(u) – Label all nodes on the found path as active

⎩ ⎨ ⎧ = active is u asleep is u ) (

idle n

P u H

v u i i e

C r r v u H

,

) , , ( ⋅ =

Edge cost: Node cost:

11

Illustration of ISTH

• Cu,v=2, Pidle=1
• Find a new path in each iteration

1 1 1 1 1 1 Source 1 r1 = 0.2 Source 2 r2 = 0.2 1 1 1 1 × 3 + 0.2 × 2 × 2=3.8 1 + 0.2 × 2 × 2=1.8 New cost : 2 2 2 2 2 2 2 2 cost reduction! sink

12

Properties of ISTH

• Online, distributed implementation is easy
• Known approx. ratio is k, num of sources
• Performance for special cases

– Approx. ratio is 2 when r = 0

• Good performance when data rates are low

– Optimal when Pidle=0

SLIDE 3

3

13

Minimum Power Configuration Protocol (MPCP)

• Routing

– Extends DSDV with routing metrics He and Hn

• Sleep scheduling

– Turns on radio if on a route, runs a duty cycle otherwise

• Power control

– Determines transmission power to different neighbors

• Cross-layer optimization

– Data rate, transmission power, and state of the node jointly determines the routing cost

14

Simulation Environment

• Prowler simulator extended by Rmase project

– Prowler: http://www.isis.vanderbilt.edu/projects/nest/prowler/ – Rmase: http://www2.parc.com/spl/projects/era/nest/Rmase/

• Implemented USC model [Zuniga et al. 04] to simulate

lossy links of Mica2 motes

• Baseline protocols:

– MT: Extended DSDV that minimizes num of Txs – MTP: Extended DSDV that minimizes Tx Power

• Data rate per flow: 0.3 Kbps, 100 nodes

15

Network Energy Consumption

Energy cost of all nodes MPCP saves up to 30% energy Energy cost of non-source nodes MPCP saves up to 80% energy

Energy Cost of All Nodes (J) Energy Cost of Non-Source Nodes (J) 16

Delivery Rate and Delay

Delivery rate Packet delay

MPCP causes slightly higher network contention due to path reuse

17

Summary

• Minimum Power Configuration: minimize total power of

wireless sensor networks

– Sleep scheduling + minimum power routing – Adaptive to workload

• MPCP: Efficient online protocols
• Simulations based on MICA2: MPCP saves more energy

than min-power routing and shortest path routing.

18

Reference

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

Configuration for Wireless Communication in Sensor Networks, ACM Transactions on Sensor Networks, 3(2), June 2007. Extended version of MobiHoc'05 paper.