A survey of Energy Efficient Network Protocols for Wireless

Networks

Christine E. JonesKrishna M. Sivalingam

Prathima AgrawalJyh-Cheng Chen

Issue 1/2 Rapid expansion of wireless

services, mobile data and wireless LANs

Greatest limitation: finite power supplies

Issue 2/2 Typical example of power consumption

from a mobile computer (Toshiba 410 CDT):

o 36% Displayo 21% CPU/memoryo 18% Wireless interfaceo 18% Hard drive

Goalo Incorporate energy conservation at all layers of

protocol stack

Energy Efficiency Research in Protocol Stack

Physical Layer Two different perspectives Increase battery capacity

o Increase capacity while restricting weighto However battery technology hasn’t

experienced significant advancement in the past 30 years

Decrease of energy consumedo Variable clock speed CPUso Flash memoryo Disk spindown

Sources of Power Consumption

Two typeso Communication relatedo Computation related

Tradeoff between them

Communication related Three modes:

o Transmito Receiveo Standby

Example:o Proxim RangeLAN2 2.4 GHz 1.6 Mbps PCMCIA

card 1.5W transmit, 0.75W receive, 0.01W standby

Turnaround between transmit and receive typically takes 6 to 30 microseconds

Optimize the transceiver usage

Computation related Usage of CPU, main memory and

disk Data compression techniques for

reduction of packet length increase power consumption

General Guidelines and Mechanisms 1/5

Reduce collisions in MACo Retransmissions lead to power

consumption and delayso Cannot be completely eliminated due to

user mobility and varying set of mobiles Change typical broadcast

mechanismo 802.11: Receiver keeps track of channel

status through constant monitoring

General Guidelines and Mechanisms 2/5

Turnaround between transmit and receive mode spends time and powero Allocate contiguous slots for transmission

or receptiono Request multiple transmission slots with a

single reservation packet Computation of transmission schedule

should be relegated to base station

General Guidelines and Mechanisms 3/5

Scheduling algorithm may additionally consider battery power level

Allow mobile to re-arrange allocated slots under low-power conditions

At link layer:o Avoid transmissions when channel

conditions are pooro Combine ARQ and FEC mechanisms

General Guidelines and Mechanisms 4/5

Energy efficient routing protocolso Ensure all nodes equally deplete their power

levelo Avoid routing through nodes with lower

battery power Requires mechanism for dissemination of node

battery powero Periodicity of routing updates can be

reduced May result in inefficient routes

General Guidelines and Mechanisms 5/5

OS levelo Suspend of specific sub-unit (disk,

memory, display etc.) when detect prolonged inactivity

MAC Sublayer Three specific MAC protocols

o IEEE 802.11o EC-MACo PAMAS

IEEE 802.11 Standard 1/2 A mobile that wishes to conserve power

may switch to sleep mode and inform the base station

The base stationo Buffers packets that are destined for the

sleeping mobileo Periodically transmits a beacon that contains

information about such buffered packets When the mobile wakes up, it listens for

this beacon, and responds to the base station which then forwards the packets

IEEE 802.11 Standard 2/2 Conserves power but results in additional

delays and may affect the QoS Experimental measurements of per

packet energy consumptiono Same incremental costs for both unicast and

broadcast traffico Higher fixed costs for unicast transmission

because of MAC coordination and CTS and ACK messages

EC-MAC Protocol 1/7 Energy Conserving-Medium Access

Control Developed with the issue of energy

efficiency as a primary goal Defined for infrastructure network but

can be extended to ad-hoc by allowing mobiles to elect a coordinator

It is based on reservation and scheduling and supports QoS

EC-MAC Protocol 2/7

EC-MAC Protocol 3/7 FSM:

o transmitted at the start of each frame by the base station

o contains synchronization information and uplink transmission order for subsequent reservation phase

Request/Update Phase:o Each registered mobile transmits new

connection requests and status of established queues

o Collisions avoided

EC-MAC Protocol 4/7 New User Phase (Aloha):

o Registration of new userso Collisions occuro Provides time for BS to compute the data

phase transmission schedule Schedule Message:

o Broadcasted by the base stationo Contains the slot permissions for the

subsequent data phase

EC-MAC Protocol 5/7 Data phase (Downlink):

o Transmission from base station to mobiles

o Scheduled considering QoS requirements

Data phase (Uplink):o Slots allocated using a suitable

scheduling algorithm

EC-MAC Protocol 6/7 Collisions are avoided and this

reduces the number of retransmissions

Mobile receivers are not required to monitor the channel because of schedules

Centralized scheduler can optimize schedule so that mobiles transmit and receive within contiguous slots

EC-MAC Protocol 7/7 Scheduling algorithms may consider

also battery power level in addition to packet priority

Frames may be fixed or variable lengtho Fixed are desirable from energy efficient

perspective since a mobile will know when to wake up to receive FSM

o Variable are better for meeting the demands of bursty traffic

PAMAS Protocol 1/3 Designed for ad hoc network, with

energy efficiency as primary goal Provides separate channels for

RTS/CTS control packets and data packets

PAMAS Protocol 2/3 A mobile with a packet to transmit sends a

RTS over the control channel, and awaits the CTS

If no CTS arrives the mobile enters a backoff state

However, if CTS is received, then the mobile transmits the packet over the data channel

The receiving mobile transmits a “busy tone” over the control channel for the others to determine that the data channel is busy

PAMAS Protocol 3/3 The use of control channel allows mobiles

to determine when and for how long to power off

A mobile can power off when:o It has no packets to transmit and a neighbor

begins transmitting a packet not destined for ito It does have packets to transmit but at least

one neighbor-pair is communicating The length of power off time is determined

through the use of a probe protocol (Singh and Raghavendra, 1998)

LLC Sublayer Is responsible for the error control The two most common techniques

for the error control are Automatic Repeat Request (ARQ) and Forward Error Correction (FEC)

Both waste network bandwidth and power resources due to retransmissions and greater overhead

LLC Sublayer Recent research has addressed

low-power error control and several energy efficient link layer protocols have been proposed:o Adaptive Error Control with ARQo Adaptive Error Control with ARQ/FEC

Combinationo Adaptive Power Control and Coding

Scheme

Adaptive Error Control with ARQ 1/3

Three guidelines:o Avoid persistence in retransmitting

datao Trade off number of retransmission

attempts for probability of successful transmission

o Inhibit transmission when channel conditions are poor

Adaptive Error Control with ARQ 2/3

Works as normal until the transmitter detects an error due to the lack of a received ACK.

Then the protocol enters a probing mode in which a probing packet is transmitted every t slots. Probe packet contains only header

This mode continues until an ACK is received. Then the protocol returns to normal mode and continues transmission from where it was interrupted

Adaptive Error Control with ARQ 3/3

Analysis results show that under slow fading channel conditions it is superior to standard ARQ in terms of energy efficiency

There is an optimal transmission power in respect to energy efficiency

o Decreasing the transmission power results in an increased number of transmission attempts but may be more efficient than attempting to maximize the throughput

Adaptive Error Control withARQ/FEC Combination

Each packet streamo is associated with service quality parameters

(packet size, QoS requirements)o maintains its own time-adaptive customized

error control scheme Error control scheme

o is a combination of an ARQ scheme (Go-Back-N, CACK, SACK, etc.) and a FEC scheme

o modifies as channel conditions change over time

Adaptive Power Control andCoding Scheme

Each transmitter operates at a power-code pair

o Power level lies between a specified minimum and maximum

o The error code is chosen from a finite set At each iteration (timeframe):

o Receiver checks the word error rate (WER)o If the WER lies within an acceptable range,

power-code is retained, otherwise a new power-code pair is computed by the transmitter

Variations of algorithm include average WER

Network Layer Energy efficient routing algorithms for ad

hoc networks Does not apply to infrastructure networks

because all traffic is routed through BS Two different approaches:

o Frequent topology updates Improved routing Consumes bandwidth

o Infrequent topology updates Decreased update messages Inefficient routing and occasional missed packets

Network Layer Typical metrics for ad hoc routing

protocolso Shortest-hopo Shortest-delayo Locality-stability

However they may result in the overuse of energy resources of a small set of mobiles decreasing mobile and network life

Network Layer example Using shortest-hop

routing, traffic from A to D will always be routed through E

E’s energy reserves will be drained faster and then F will be disconnected from network

A to D traffic should also use the B-C path extending networks life

Network Layer: Unicast Traffic 1/6

Five different metricso Energy consumed per packeto Time to network partition

Given a network topology, a minimal set of mobiles exist such that their removal will cause the network to partition

The traffic in that mobiles should be divided in such a way that they drain their power at equal rates

Network Layer: Unicast Traffic 2/6

o Variance in power level across mobiles All mobiles are equal and remain powered-on

together for as long as possibleo Cost per packet

Routes should be created such that mobiles with depleted energy reserves do not lie on many routes

o Maximum mobile cost By minimizing the cost experienced by a mobile

when routing a packet through it significant reductions in the maximum mobile cost result

Network Layer: Unicast Traffic 3/6

The goal is to minimize all the metrics except for the second which should be maximized

Shortest-cost routing protocol is more appropriate instead of shortest-hop

So although packets may be routed through longer paths, the paths contain mobiles that have greater amounts of energy reserves

Also routing traffic through lightly loaded mobiles conserves energy because it minimizes contention and retransmission

Network Layer: Unicast Traffic 4/6

Simulation results showed no extra delay over the traditional shortest-hop metric

This is true because congested paths are often avoided

However this approach requires that every mobile have knowledge of every other mobile and the links between them

This creates significant communication overhead and increased delay

Network Layer: Unicast Traffic 5/6

Stojmenovic and Lin proposed localized routing algorithms

These algorithms depend only on information about the source location, the location of neighbors and location of the destination

This information is collected through GPS receivers which are included in every mobile

Network Layer: Unicast Traffic 6/6

They proposed a new power-cost metrico Incorporates both a mobile’s lifetime and

distance based power metrics Three power-aware localized routing

algorithms were developedo Power

Minimize total amount of power utilized when transmitting a packet

o Cost Avoid mobiles with low battery reserves

o Power-cost Combination of the other two

Network Layer: Broadcast Traffic 1/4

Each mobile needs to receive a packet only once

Intermediate mobiles are required to retransmit the packet

Key idea: allow each mobile’s radio to turn off after receiving a packet if its neighbors have already received a copy of the packet

Network Layer: Broadcast Traffic 2/4

In traditional networks broadcast technique is a simple flooding algorithm

o No global information topology gatheredo Requires little control overheado Completes with minimum number of hops

Not suitable for wireless networks because many intermediate nodes must retransmit packets needlessly

It is more beneficial to spend some energy in gathering topology information in order to determine the most efficient broadcast tree

Network Layer: Broadcast Traffic 3/4

A broadcast approach is presented in (Singh et al., 1999)

The tree is constructed starting from the source and expanding to the neighbor that has the lowest cost per outgoing degree

Mobile costs continuously change so broadcast transmissions may traverse different trees

Simulations showed very little difference in delay but 20% or better in energy consumption

Network Layer: Broadcast Traffic 4/4

In (Wieselthier et al., 2000) is presented an algorithm for determining the minimum-energy tree

There exists an optimal point in the trade-off between reaching greater number of mobiles in a single hop by using higher transmission power versus reaching fewer mobiles but using lower power levels

Transport Layer TCP was designed initially for wired networks

o Physical links are fairly reliableo Packet loss is random in nature

Over a wireless link it degrades significantlyo It resorts to a larger number of retransmissions

and frequently invoke congestion control measures because it confuses link errors and loss as channel congestion

The increased retransmissions consume battery energy and bandwidth

Transport Layer Various schemes have been

proposedo Split connection protocolso Link-layer protocolso End-to-end protocols

Split connection protocols 1/2

Split connection protocols 2/2

Completely hide the wireless link from the wired network by splitting each TCP connection into two separate connections at the BS

The second one may use modified versions of TCP that enhance performance over the wireless channel

Link-layer protocols 1/2

Link-layer protocols 2/2 Hides link related losses from the

TCP source Uses a combination of local

retransmissions and FEC over the wireless link

Local retransmissions use techniques that are tuned to the characteristics of the wireless channel

End-to-end protocols Include modified versions of TCP that

are more sensitive to the wireless environment

Uses mechanisms such as o SACK

allow the TCP source to recover from multiple packet losses

o ELN Aid the TCP source to distinguish between

congestion and other forms of loss

Energy Consumption Analysis of TCP 1/4

The energy consumption of Tahoe, Reno and New Reno is analyzed in (Zorzi and Rao, 2000)

Efficiency is defined as the average number of successful transmissions per energy unit

Results demonstrate thato error correlation affects the energy performanceo congestion control algorithms of TCP allow for

greater energy savings by backing off and wait during error bursts

o energy efficiency is sensitive to the version of TCP

Energy Consumption Analysis of TCP 2/4

The same versions of TCP were studied in (Tsaoussidis et al., 2000a) in terms of energy/throughput tradeoffs

Results showed thato no single version is most appropriate within

wired/wireless heterogeneous networkso the key to balancing energy and throughput is

through the error control mechanism They proposed a modified version of TCP,

referred to as TCP-Probing in (Tsaoussidis and Badr, 2000)

Energy Consumption Analysis of TCP 3/4

In TCP-Probing when a segment is delayed or lost, instead of invoking congestion control, transmission is suspended and a probe cycle is initiated

Probe cycle:o exchange of probe segments (TCP header

with no payload) between sender and receiver

o terminates when two consecutive RTT are successfully measured

Energy Consumption Analysis of TCP 4/4

The sender invokes standard TCP congestion control if persistent error conditions are detected

However, if conditions indicate transient random error, then the sender resume transmissions according to available network bandwidth

OS/Middleware The main functions of an operating system

is to manage access to physical resources like CPU, memory and disk space

CPUs can be operated at lower speeds by scaling down the supply voltage (quadratic relationship between power and supply voltage)

Predictive shutdown Different page placement algorithms exploit

the new power management features of memory technology

Application Layer 1/2 Load Partitioning

o Applications may be selectively partitioned between the mobile and base station

o Most of the power intensive computations of an application are executed at the BS

Proxieso Middleware that automatically adapt the

applications to changes in battery power and bandwidth

o Either on the mobile or BS side of wireless link

Application Layer 2/2 Databases

o Minimize power consumed per transaction through embedded indexing

Video Processingo Reduce effective bit rate of videoo Carefully discard video frames