Idle Communication PowerIdle Communication Power

Lei Guo, Xiaoning Ding, Haining Wang, Qun Li,

Songqing Chen, and Xiaodong Zhang

ExploitingExploiting

to Improve Wireless Network to Improve Wireless Network Performance and Energy EfficiencyPerformance and Energy Efficiency

Challenges in Wireless System Design

• Energy saving is not easy– Limited battery capacity in wireless devices – High power consumption in wireless communication

• High performance costs energy and fairness – Wireless users demand high throughput, but …– A high throughput device needs less sleep. – A channel allocation mechanism can favor some but

degrade performance of others. • Can we win both instead of addressing the trade-off?

Power Consumption for Wireless Communication

• Energy consumption %

• A standard way to save energy

– Put the WNI into sleep when idle (for a 5 V device)

> 50% total energy

up to 10%total energy

high power mode450 mA

low power mode15 mA

802.11 Power Saving Plan in its Basic Infrastructure Mode

• Access point– Buffer data for sleeping

stations– Broadcast beacon with

TIM periodically (100 ms)

• Sleeping station– Wake up periodically to

receive beacon– Poll access point to receive

data– Sleep again

Access Point

Internet

Traffic Indication Map (TIM)

sleeping station

wake uppoll

receive data

Observations of IEEE 802.11 Protocol • A client/server model

– Each station independently communicates with AP– AP serves a station one at a time via the channel.

• The saving mode affects TCP traffic– Increasing RTT and decreasing throughput.

• Performance anomaly (Infocom’03)– Non-uniform transfer rates between different stations to

AP due to distance and obstacle condition differences. – A low speed station has low channel utilization rate.

• Waste energy while a station is waiting for its turn.– Idle communication power due to strong dependency

Existing Solutions to address the Limits• Reducing idle communication power by

– Traffic prediction: bounded slowdown (MOBICOM’02)– Self-tuning with application hints (MOBICOM’03)– Limits: case by case, and accuracy can vary.

• Address the performance anomaly– Time-based fairness scheduling: a constant time unit is

given to each station (USENIX 04) – Limits: poorly conditioned stations suffer: fast is faster,

and slow is slower. (energy: 1 bit = CPU 10,000 cycles)Our work: to win both performance and energy

subject to fair scheduling.

Source of Idle Communication PowerWhile the channel is used by one station, idle communication power is wasted in many other stations

AP

Wireless performance anomaly makes this power waste worse, but also with an opportunity.

Outline

• Motivation and rationale

• System model and algorithms

• System design and implementation

• Performance evaluation

• Conclusion

Restructure a Wireless Network to P2P model to Enable Multi-hop Relays

To help low channel rate stations to Increase throughput and extend network coverage

AP

X

Effectiveness of Relays is from Strong Dependency • Slow stations become faster

– Completing the data transfer ahead of the unit time.– Equivalent to move the station closer to AP or improve

the station’s communication condition.

• Fast stations serve as proxies for slow stations– Performance improvement of slow stations reduced the

waste of idle communication powers of fast stations --- shortening the waiting time.

• Effective P2P coordination among stations is the key.

Incentive and Fairness to Fast Stations

• Why not sleep or wait, but proxy/relay for others?– Sleep lowers throughput, and wait wastes energy. – Idle communication energy can be used – The saved time in slow stations should be contributed.

• How do we handle fairness?– A proxy should be given incentive for its service– For either proxy or client, the throughput and energy

efficiency should be improved after relays.

Rationale

• Energy efficiency:– effective number of bits delivered per energy unit

• Self-incentive multi-hop relay– Use channel time to pay the relay service

A win-win solution

Throughput Energy efficiencyProxy Increase IncreaseClient Increase Increase

System Model• Time based fairness in the shared radio channel

• Consequence of multi-hop relays

– Proxy: throughput idle time energy efficiency– Client: channel rate throughput

S1 S2 … Si … Sn

ti = t = 1/n

1 roundidle idle

Sq

Client

Sp

Proxy

S0

AP

Token-based Channel Scheduling• A token is a ticket for a data transfer (RX/TX) in

one time unit• AP initially distributes an equal amount of tokens to

each station (fairness).• A pair of RX & TX consumes one token.• A token bucket is used in channel scheduling.• Multi-hop relays are operated under token

exchanges. • The token mechanism provides incentive to fast

stations: receive more time units than relays needed.

packets

A Token Bucket Associated with Each Station

tokens from AP

Overflow!Re-allocate to other

stations by AP

Token Bucket

Packet Queue

Transmitter 1 token per packet

Maintaining a Routing Table in AP

APS1

STA Proxy RouteS1 --- R(0,1)S2 --- R(0,2)S3 S2 R(0,3)S4 S2 R(0,4)

S2

S3

S4

Hop Station Route1 Self R(0,2)2 S3 R(2,3)

S4 R(2,4)

Hop Station Route1 S2 R(0,2)2 Self R(2,3)

Put Them Together: Selfish Forwarding - SFW

• Proxy discovery and selection– A poorly conditioned client broadcasts a relay request – One or more stations bet to relay with a price (tokens).

AP assigns a relaying station for clients based on the second price auction in game theory.

• Channel scheduling– AP distributes tokens without any enforcement. – The relaying actions are determined by token exchanges

among stations. • Multi-hop routing is done by a routing table in AP

SFW System Implementation

• Access Point– NetGear MA311 802.11b PCI wireless adaptor– HostAP linux driver version 0.1.3

• Wireless Stations– NetGear MA401 802.11b PCMCIA wireless

adaptor– ORiNOCO linux driver version 0.15rc2

Protocols Compared in Experiments

• DCF– Most widely used protocol in 802.11b network– Distributed Coordination Function

• TBF– Time-based Fairness (USENIX 2004)

• SFW– Selfish Forwarding (our own)

Single Client Experiment

AP

11Mbps

11Mbps

1Mbps slow link!

Channel allocation scheme

Channel allocation scheme

Performance Evaluation1 proxy (P), 1 client (Q)

ener

gy e

ffici

ency

(Mb/

J)th

roug

hput

(M

bps)

16%

170%

266%

Multi-clients Experiment

AP

11Mbps

1Mbps

Performance Evaluation

Proxy throughput gain

1 proxy, multiple clients

Conclusion

• Exclusive communications between AP and stations create idle communication power.

• Wireless performance anomaly is an opportunity.• P2P based Selfish Forwarding Protocol

– Improve performance and energy efficiency for everyone– Make the channel sharing and scheduling more fair.

• It is easy to implement for practical usage.

Thank you!Thank you!