adaptive traffic-based control method for energy conservation in wireless devices

www.elsevier.com/locate/simpat

Simulation Modelling Practice and Theory 13 (2005) 213–232

Adaptive traffic-based control methodfor energy conservation in wireless devices

Constandinos X. Mavromoustakis *, Helen D. Karatza

Faculty of Sciences, Department of Informatics, Aristotle University of Thessaloniki,

54124 Thessaloniki, Greece

Received 4 June 2004; accepted 4 October 2004

Available online 18 November 2004

Abstract

Mobile units have many constraints for reliable communication in today�s mobile environ-

ments. Unlike wired networks in mobile networks, mobility induces frequent route changes.

Energy conservation is an important issue that has to be taken into account for mobile devices.

Every infrastructureless network must be adaptively self-configured particularly in terms of

energy, connectivity, resource allocation and memory. Efficient utilization of battery power

is important for wireless users because due to their movements their energy is fluctuating at

different levels during operation mode. Traffic plays a major role for energy consumption

because of the unpredictable incoming flow nature. Generally, the minimum transmission

power required to keep the network connected achieves the optimal throughput performance

in a dynamically changing topology network. In this paper an adaptive traffic-based control

method for energy conservation is described and examined, which bounds an asynchronous

operation where each node evaluates dissimilar sleep-wake schedules/states based on each

node�s incoming sleep-history traffic. Simulation study is carried out for throughput, traffic

characterization against performance and energy conservation evaluation of the proposed

model taking into account a number of metrics and estimation of the effects of incrementing

the sleep time duration to conserve energy. The proposed method could be applied to infra-

structureless networks with any underlying routing protocol to provide independency,

1569-190X/$ - see front matter � 2004 Elsevier B.V. All rights reserved.

doi:10.1016/j.simpat.2004.10.003

* Corresponding author. Tel.: +30 2310 997974; fax: +30 2310 998310.

E-mail addresses: [email protected] (C.X. Mavromoustakis), [email protected] (H.D.

Karatza).

mailto:[email protected]

mailto:[email protected]

214 C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232

portability, as well as ‘‘fair’’ collaboration among energy conservation mechanism and routing

protocol.

� 2004 Elsevier B.V. All rights reserved.

Keywords: Wireless network; Adaptive traffic scheme; Routing; Energy conservation; Dynamic caching

1. Introduction

Wireless devices are a collection of wireless nodes that communicate over radio.

This set of mobile nodes does not need any infrastructure. These kinds of networks

are very flexible and suitable for several situations and applications allowing an

infrastructureless network without any preinstalled components. Every node acts

as a router and has limited transmission range while the communication traffic

has to be relayed over several intermediate nodes (multi-hop) to enable the commu-nication from a source to a destination.

Nodes in wireless networks typically rely on their battery energy. Wireless inter-

face is one of the major impacts on energy reservation of the nodes. If nodes have

low or moderate traffic load, network interfaces are in the idle state most of the time.

Thus turning off idle network interfaces is a way to conserve energy. Roughly speak-

ing by turning off idle network interfaces we simply reduce the number of nodes that

are available for packet routing, resulting in a significant reduction in network per-

formance. Furthermore by turning off idle network interfaces the network forward-ing capacity is dramatically reduced. Effective energy conservation techniques should

maintain a balance between the amount of energy a network conserves and the avail-

able resources (spare forwarding capacity) to sustain acceptable network perfor-

mance. In opposite situation the disastrous effects would be the partition of the

network in order to consume energy by turning off network interfaces for a proper

time. Hence, the energy conservation mechanism has to be closely collaborative with

the routing protocol used to enable the proper behavior of the nodes in idle state

while maintaining the forwarding mechanism.In this paper, an Adaptive Energy Conservation (ADEC) protocol is examined

which is further research extension of our previous work [21]. In [21] the Adaptive

Energy Conservation (ADEC) method is introduced where selectively turns off nodes

interfaces to save power of each node and maximize the network lifetime. As known

in a terminal, the latency, connectivity, energy, and memory are the essential ele-

ments of today�s mobile environments whose performance may be significantly im-

proved by caching techniques. In this work ADEC protocol uses traffic patterns

for the thorough examination of network behavior. Adaptively ADEC scheme as-signs a variable sleeping time based on traffic that each node ‘‘accepts’’ throughout

the sleep time duration based on a self-similarity nature of traffic. Therefore ADEC

by using dynamic caching manipulate network state in a distributed form and ac-

tively control each node�s sleeping duration, to minimize the energy cost of peer-

to-peer communication among mobile terminals. In addition this method provides

C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232 215

a way for guarantee the buffering of data and a way to overcome the loss of packet

information. Finally this model could be applied to ad-hoc networks with any under-

lining routing protocol support to provide independency as well as ‘‘fair’’ collabora-

tion among energy conservation mechanism and routing protocol.

The organization of the paper is as follows: Section 2 discusses the related workthat has been done on energy conservation (EC) featuring out the basic energy con-

servation principles and conducted solutions by different schemes. Section 3 then

introduces the proposed Adaptive Dynamic Caching Energy Conservation (AD-

CEC), followed by Section 4 which provides the evaluation and simulation results

of the proposed scheme in contrast to the energy consumption associated with Adap-

tive Dynamic Caching. Finally, Section 5 concludes with a summary of our contri-

bution and suggestion for further research.

2. Related work

Minimizing energy consumption and maximizing the system lifetime has been one

of the major design goals for wireless networks. In the last few years, many research-

ers around the world actively explored advanced power consumption approaches for

wireless devices. On one hand, many wireless device manufacturers have been striv-

ing for low power consumption in their products [2], exploring and discussing theirlow power transceiver architectures and low power signal processing systems. On the

other hand quite a lot of protocols have been designed and use different mechanisms

to reduce energy consumption, that can be classified into two categories: Active and

passive protocols. Active techniques conserve energy by performing energy con-

scious operations, such as transmission scheduling using a directional antenna [6],

and energy-aware routing [4,5]. On the other hand passive techniques conserve en-

ergy by scheduling network interface devices to the sleep mode when a node is not

currently taking part in communication activity.Many different protocols were designed taking into account separately aspects

dealing with MAC layer [7] network layer [3–5], topological and geographical infor-

mation-based techniques (GAF) [8]. In [8] the entire network is divided into small

virtual pieces (grids) and this area as recognizable by geographical information al-

lows only one node to be active in the grid while the other nodes turn off their inter-

faces to conserve energy. In [9] the goal is to turn off nodes without significantly

diminishing the capacity or connectivity of the network. The connectivity and for-

warding capability as stated in [9], is maintained by keeping the nodes that constitutea backbone infrastructure (BNs [20]) in active mode, and switching off the other.

However nodes� participation in network forwarding activity [1] is adaptively ad-

justed, based on their battery remaining energy. In Table 1 the rates of consumption

are presented for some commercial transceivers [1–5] and their Digital Radio Power

States (DRPS). It is easily recognizable from different radio card samples that the

rate of consumption in the ‘‘receiving’’ state could be less 50% of that consumed

in ‘‘transmitting’’ state. There is importance to appropriately determine when and

Table 1

The degree of power consumption and Digital Radio Power States (DRPS)

Radio interface IEEE 802.11

Interfaces (2.4GHz)

Transmit Receive Idle Sleep mode Mbps

Lucent Silver 1.3W 0.90W 0.74W .048W 11

WaveLAN Turbo Card 285mA 185mA 185mA 9mA 11

Aironet PC4800 1.4–1.9W 1.3–1.4W 1.34W 0.75W 11

Cabletron Roamabout 1.4W 1.0W 0.83W 0.13W 2

Lucent Bronze 1.3W 0.97W 0.84W 0.066W 2


at what power level a mobile host should attempt transmission or retransmission of

packets. 1

Generally, energy consumption protocols may use the following sets of mecha-

nisms to reduce energy consumption: first, nodes (end user devices) that play a‘‘key’’ role in energy consumption must be encouraged more frequently to enter in

low energy states. Then routing protocol must be optimal (able to choose best routes

that consume the least energy) and provide a significant overhead reduction. And

nodes could be selectively used depending on their current energy status. On the

other hand the routing table entries must be adaptively updated by routing protocol

in order to ensure each link�s connectivity status and the connection capability of

each node in the network with the rest. Researches have shown that reactive routing

(route is established on demand) schemes generate less overhead packets, thus con-sume less power than proactive ones (nodes maintain and periodically refresh routes,

even if no traffic is carried on a maintained route).

A traffic-load history determination in association with battery lifetime has not

been yet investigated. In this work the overall research is focused on load history char-

acterization for each node targeting the energy conservation concept. The self-simi-

larity of packet traffic characterization [14,15] enables us to examine such scenario.

The proposed adaptive method allows nodes to change their state depending entirely

on their traffic history under various conditions as discussed in the following section.

3. Adaptive Dynamic Caching Energy Conservation (ADCEC)

This section describes the Adaptive Dynamic Caching Energy Conservation (AD-

CEC) method, which bounds a partially asynchronous operation where each node

evaluates dissimilar sleep-wake schedules/states based on each node�s incoming

sleep-history traffic.

3.1. A scheme for dynamic caching

In a fully distributed decision system, efficient message passing plays a major role

for issues concerning various QoS parameters as well as power consumption meth-

1 Node�s transceiver should be powered off when not in use.


ods. As known energy conservation mechanism has to be closely collaborative with

the routing protocol 2 for the proper/acceptable behavior of nodes being in idle state

and maintain the packet forwarding mechanism. But routing packets in wired/wire-

less communication network needs essentially a dynamic approach due to the

stochastic nature of data traffic. An on demand cost-effective, adaptive solution-find-ing algorithm is in principle more attractive because it can keep abreast of any pos-

sible changes within network load and failures due to energy deficiency. Particularly

for wireless devices/networks where nodes cooperatively form a network indepen-

dently of any fixed base station infrastructure (usually infrastructureless). On the

other hand, in dynamic caching-oriented methods there are some trade-offs that have

to be taken into account, such as generated overhead in messaging passing for delay

sensitive services, network utilization, simplicity for the implementation on the wire-

less environment etc.In Adaptive Dynamic Caching, messages are manipulated using a specific routing

protocol, as will be described in the next section. Packets for routing and control

purposes are contributing any time in the network for informing the neighboring

nodes or all nodes in hops (route) for residual energy. Roughly speaking one of

the most important task in dynamic caching is whether a node will decide to sent

an acknowledgement message (ACK) to neighboring nodes or in a zone (based on

the routing protocol used) for informing about the current state and the state of

the next time step. If a packet needs to be sent from a source to a destination viasome intermediate nodes or even the destination might be in a sleep state. For inter-

mediate nodes packets are routed via another optimal path depending on the routing

protocol used. A question arises for the decision of which node will ‘‘cache’’ the

information for the destination node in the route (hop-by-hop basis) when the des-

tination node or any intermediate node ‘‘lie’’ in the sleep mode. As known in the

sleep state, node�s interface can neither transmit nor receive, so it consumes energy.

As a result the destined information for a proper node in sleep mode after a ‘‘time-

out’’ will be lost.In order to achieve an adaptive solution for this issue, acknowledgement packets

ACKs are sent to each neighbor in the zone (as ZRP [6,12,16,17] will be explained in

the next section). ACK packets are sent for informing each node about the state of

the candidate node for sleep state. The mechanism used selects the previously fol-

lowed node in the route (neighbor to node in the sleep mode) to cache the packets

destined for the node with turned off interfaces (sleep). This is shown in Fig. 1 where

packet is destined for wireless device (node 10) from node 1. Suddenly (for triggering

or energy consumption reasons) node enters into the sleep mode in order to achieveenergy conservation. The time instant is not suitable since all packets destined for

node 10 will be lost (partially or in total depending on sleep time). ADCEC method

enables the packets to be ‘‘cached’’ in the previous hop node (node 7) from node 10

and forwards the packets destined for node 10 when node 10 enters the active state.

Therefore no information will be lost since it will be held in node�s 7 buffer (storage

2 Network layer power save protocols.

2

6

3

6

1

4

8

5

9

7

N in sleep state

N N in active state

N

Wireless interface

cached packets for Nodes 10,5

10

Fig. 1. Wireless devices connectivity and different activity states.


unit). Additionally the main difference from [3–9] is the adaptivity. The adaptivity

occurs for the incoming traffic of the destined node [13–15]. In Fig. 1 it is considered

that for node N = 10 the traffic is TC(t),N where T is the incoming traffic (capacity in

the certain time t). Node 7 caches the packets for node 10 thus increasing the storage

unit for node 10 in node 7�s unit. This corresponds to Cs,i(t) where i is the destinationnode and s is the buffering node (a hop before destination).

It must be also pointed out that in Fig. 1, node�s 5 destined information could be

cached also in node 3. Routing decision blocks this option since there are no memory

limitations for node 7. Procedural packets PROC continuously inform for route dis-

covery the ‘‘route neighboring’’ nodes, as well as for TTL (applicable for delay sen-

sitive services where if necessary other optimal route should be activated), residual

energy, and history of all visited nodes. Furthermore PROC packets pursue one

important operation: they determine whether a node can enter with safety to sleepstate i.e. if there is forwarding activity in the proper time which in turn depends

on the routing protocol, 3 idle time and residual battery energy.

The main issue for conserving energy is to identify and associate the cached pack-

ets with the next sleeping time of the intended node. This requires the adaptivity of

the cached capacity (stored packets- traffic during sleep time for node 10) with the

next-following duration of sleep time of node 10. This sleeping time will be propor-

tionally influenced with the cached capacity in the time distance whish is ti(t�tD),

where tD represents the equally spaced/triggered time slots. A thorough explanationof this principle is conducted in the following section.

3.2. Operations for energy conservation

When heavy load exists in wireless devices, the commonly used and known

periodic sleep/listen scheme does not benefit for efficient energy conservation. The

3 The routing protocol used in the implementation is ZRP [12,16].

Active Active ActiveSleepSleep Sleep

Active SleepSleep SleepActive

Active

ActiveSleep

Periodic sleep and listen

Traffic-based sleep and listen

Fig. 2. Different duration of sleep/wake schedules, based on node�s incoming sleep-history traffic.


evaluation and design of energy efficient communication protocols therefore requires

practical understanding of the energy consumption behavior of the underlying net-

work interface. The main aspect for interface to consume energy is basically the

mode it does operation. As known there are two basic states: The idle state and sleepstate. In the idle state, an interface can transmit or receive data at any time, but it

consumes more energy due to the number of circuit elements that must be powered.

On the other hand in the sleep mode, an interface can neither transmit nor receive, so

it consumes significantly less energy. For transmitting or receiving, an interface must

explicitly transition to the idle state, which requires both time and energy. The idle

energy consumption is quite high, comparable to that of receiving and in an order of

magnitude more than that of sleeping [7–9] (see Table 1).

As known the periodic sleep/listen scheme does not benefit for heavy loads andconversely. In the previous section a simple scheme is described for caching the data

packets destined for a proper node. The capacity of the cached data corresponds to

time duration ti(t � tD) for the next sleep duration of the destination node. Sleeping

time for destination node is high enough when the node in the previous time sleeping

slot, that is ti(t � tD(s�1)) did not receive any packets. Hence it stands that:

T CðsÞ;N < T Cðs�1Þ;N then tDðsþ1Þ P tDðsÞ for node N ; and s ¼ 1; 2; 3; . . . ;m:

ð1ÞThus if tD is high then the next sleep duration of node N will be in turn higher

than the previous one (due to inactivity of the node in the tD(s�1)). In this way each

node evaluates dissimilar sleep and active states based entirely on each node�s incom-

ing sleep-history traffic (in-sleep history). This principle is shown in Fig. 2.

Sleep time duration can be measured using the following expression:

SðtnewÞD ¼ SðtÞDðs�1Þ � SðtÞDðs�1Þ �Packs cached

Total packets destined for D

� �

where S(tnew)D is the new time duration 4 depending on the traffic history of the

route destined for D, and S(t)D(s�1) is the previous time step (slot) duration.

4 New time duration cannot exceed twice the corresponding value.


Capacities and durations are normalized in order to overcome latencies problems.

Only latency (usually for delay sensitive services) can ‘‘disturb’’ sleep period by plac-

ing a limit on the sleep time duration of the nodes. PROC packets are purely respon-

sible for informing the node for any limitations. If a node does not receive any

packets for a long period due to the stochastic nature of incoming traffic, the sleeptime will increase at an undesirable grade. Thus PROC packets by using a specified

field PROC_tLIM place a limit on the sleep time of the nodes to avoid increased la-

tency thus resulting in network partitioning. Also in order to prevent a large number

of nodes to enter into the sleep state and at the same time avoid network partition,

we set a maximum number of nodes that are allowed to be in the sleep state (entirely

based on the total number of nodes of a non-partitioned network).

For each node, there are three states: the ‘‘active’’, the ‘‘sleep’’ and the ‘‘wait to

sleep’’ or ‘‘listening’’ state. The state transition diagram is shown if Fig. 3.Fig. 3 shows the state transition diagram of each node. In the active state node�s

interface can transmit or receive data at any time but consuming more energy. If

incoming traffic is continuously coming/destined for the node, it keeps remain in

the active state. This ‘‘loop’’ is not energy efficient and it breaks if traffic is reduced.

In this case node changes into listening mode for a while. Then if traffic is further

reducing, after tD(s) with no incoming traffic for the time duration tD(s�1) node enters

into the sleep mode. Sleep mode is only corrupted if there are serious latency limita-

tions or sleep period expires. For these situations node ‘‘violently’’ enters the activestate to overcome these problems.

Roughly speaking while traffic is not uniformly distributed, it does not allow the

equally spaced sleeping duration to be efficient for conserving energy. During initial-

ization period node from active state enters the sleep state by measuring the traffic

and setting a time T after activation. After network is biased [18–20] with packets,

it almost rarely changes from active to sleep state by this measure. Traffic transits

Sleep

Active

Listening

After trafficmeasurementfor time T

Communicationprocess,Forwarding etc

sleep time expiresor latency limitations

incoming traffic for (τ–1)Dt

(τ–1)Dtreduced trafficfor

Dtafterwith no incoming traffic

Fig. 3. Node operation cycle as a state transition diagram.


each node�s state through the traffic history of each node during tD(s�K), where K is

the time duration for the previous slot of that node for which data is cached.

This cooperative caching method among nodes for energy consumption bounds

many aspects for mobile environments. One major aspect is the node popularity

problem where multiple accesses (for forwarding purposes or being the destination)throughout time is marking the proper node resulting in node lock (which in turn

results in path-route lock for time t [18,20]). This principle is examined in the next

section where an evaluation through simulation is done.

4. Simulation experiments and discussions

The design and evaluation of energy efficient communication protocols requirespractical understanding of the energy consumption behavior of the underlying net-

work interface. To demonstrate the methodology discussed in this paper, we per-

formed exhaustive discrete time simulations of the proposed scenario under

several different conditions.

4.1. Routing protocol used

One basic issue is the selection of the routing protocol that should be used in orderto cooperate with the described scenario. Considering the need of bandwidth and the

limited battery power for wireless devices, it is necessary to apply efficient routing

algorithms to create, maintain and repair paths with least possible overhead produc-

tion [2]. There are two classes of routing protocol: proactive and reactive. In proac-

tive or table-driven protocols the routes are maintained for all possible destinations

continuously periodically, even if routes will not be actually used. The generated

overhead from route maintenance cause significant reduction of network perfor-

mance, increase in end-to-end delays and delay variations. Reactive or on-demandprotocols on the other hand, create and maintain routes only when they are needed.

In the implementation of the proposed scenario the Zone Routing Protocol (ZRP)

[6,10–12,16,17] is used. ZRP is a hybrid protocol that combines the reactive and pro-

active modes. The ZRP is considered advantageous because it allows a certain node

to accurately know the neighbors of any mobile terminal within a zone. These de-

vices should be in zone that could be accessible in a fixed number of hops. Since

ZRP allow the absolute communication with neighbors, is considered less expensive,

while neighbors contribute in the routing process [12]. Particularly ZRP divides thenetwork into several routing zones specifying a determined number of hops. This al-

lows the routing protocol to be adjustable for different operational network condi-

tions such as heavy traffic [13–15].

Cooperating caching is applied while ZRP is maintaining the path from a source

to a destination. A proper mobile device enters the zone and thus belonging to the

zone, form a cooperative cache terminal for its neighbors. This is because the cost

for communicating with them is low, both in terms of energy consumption and mes-

sage passing/exchanging. Fig. 4 shows graphically this principle for four different

Wireless interface

2

6

3

61

4

8

5

10

9

7

Zone A

ZoneC

Zone B

Zone D

Fig. 4. Wireless network divided into zones A, B, C and D using ZRP.


zones for 10-node wireless network. Thus every absolute neighbor of any mobile ter-minal in the routing path could become a cache storage unit while a ‘‘candidate’’

node enters the sleep state.

4.2. Simulation results of the proposed scenario

To emulate the scenario described earlier, the need of a possible realistic environ-

ment must be achieved. In this section, we present some experimental and simulation

results for performance evaluation and energy conservation offered by our scheme.The power could be measured by monitoring the three basic metrics-energy compo-

nents: (i) transmission power required to send a packet, (ii) reception power required

to receive or listen to a packet, and (iii) idle power required to stay in at the active

state (awake) in contrast with sleep time duration that follows. Transmission power

includes both the power required to drive the circuit and the transmission energy

from the antenna [1,2]. Therefore, the energy consumed by any mobile terminal

for sending, receiving or discarding a message is given by the linear equation 5 [1]

Energy = m Æ size + b; where size is the message size, and m denotes the incrementalenergy cost associated with the message and b a fixed cost of each operation.

The energy consumption model used in the simulation, for the calculation of the

amount of energy consumed, is based theoretically on the WaveLAN PC/Card en-

ergy consumption characteristics found in study by Feeney and Nilsson [1]. In [1]

it was shown that the average currents drawn by the card at 4.74V power supply

are measured at 280, 204, 178, and 14mA for transmit, receive, idle, and sleep

modes, respectively.

In a proper zone as shown in Fig. 5, all mobile terminals consume the same en-ergy. Thus, the overall energy consumption within a zone is the sum of the energy

5 Linear regression is used to test the model and find values for m and b.

(a) (b)

Fig. 5. (a) Sample communicational grid in which mobile hosts can access each other. In (b) a mobile

device is shown and the area for transmitting and receiving data.


consumed by every mobile terminal in this zone. Routing properties are observed asdescribed earlier, according to ZRP.

Two sets of experiments were performed. One set deals with the caching concept

and the grade of contribution in conserving energy, and the second deals with the

energy conserved under significant traffic and network partition limitations and

the latency issues that arise. As mentioned the caching capacity of each node could

be unlimited as memory becomes cheaper and larger. An issue then arises from

node�s unlimited capacity because if the traffic will reach a high capacity destined

for a node at the sleep state, then latency and delay problems will occur while theforwarding mechanism takes place. Thus taking into account this issue, in this paper

two different types of caching capacity have been evaluated. Investigation has been

performed for the following:

(i) Unlimited capacity.

(ii) Limited capacity for each node as 64KB, 128 KB, 256KB, 512KB, 1MB,

2MB.

An issue that has to be taken into account is whether the cached information des-

tined for a proper node could be stored in a node with higher residual energy. As

shown in simulation process if nodes with higher level of residual energy are chosen

in the path then the network partitioning probability is further reduced. For this rea-

son cached information is chosen on a recursive path basis, where source node while

having the path tries to find the node with the higher residual energy in order to as-

sign the caching process.

Another issue that arises for wireless devices is the dynamically changing topologyof the network. If a caching information process takes place and a node or some

nodes in the path will change their state into sleep mode then the path simulta-

neously changes having at least the ‘‘caching’’ node within the new path. If this

enterprise is impossible due to sudden network partitioning then before network

splits, the node which caches the information forwards the cached information to

destination which violently enters the active state.


In the implementation-simulation of this work, we have modeled and simulated

(based in C/Objective C programming language) the previously discussed scenario.

Topologically network was modeled on a ‘‘grid’’ basis as shown in Fig. 5. Each node

can communicate with other node if the area situated is in the same (3 · 3 center)

rectangular area of the node (Fig. 5(b)). Generally each node can directly communi-cate within a block.

The simulation used a two-dimensional network, consisting of 25 nodes dynami-

cally changing the topology on a non-periodic basis (asynchronously as real time

mobile users do). For each node it stands that after random time each node moves

at a random walk to one of the possible destinations (north, east, west, south) shown

in Fig. 5(b). Each link (frequency channel) has max speed reaching 2Mb per sec. The

propagation path loss is the two-ray model without fading. The network traffic is

modeled by generating constant bit rate (CBR) flows. Each source node transmitsone 512-bytes ( � 4Kbits) packet. Packets generated at every time step by following

Pareto distribution (2) as depicted in [13–15], destined for a random destination uni-

formly selected.

P ðX ¼ xÞ ¼ a � ba

xaþ1; with EðxÞ ¼ a � b

a� 1: ð2Þ

Eq. (2) generates the probability density function of the Pareto distribution andthe corresponding mean value. Where a is a shape parameter and b is the minimum

value of x. When a 6 2 the variance of the distribution is infinite. For the generation

of self-similar traffic [14], a should be between the values of 1 and 2. Roughly speak-

ing the load generated by one source is mean size of a packet train divided over mean

size of packet train and mean size of inter-train gap or it is the mean size of ON per-

iod over mean size of ON and OFF periods as follows (3):

Li ¼ONi

ONi þOFFi

: ð3Þ

Therefore for generating Pareto distribution a pseudo-Pareto generator was used

with the following formula:

X PARETO ¼ b

Uniform1=a; ð4Þ

where Uniform is a uniformly distributed value in the range (0,1].

Additionally we have modeled in each node an agent which evaluates the informa-

tion destined for a proper destination. In this way we have at any time measures of

the information destined for each node (for a given time interval) by any node. Thishas been implemented as a [N � 1] row, [N � 1] column for each node being a pos-

sible destination.

Fig. 6 illustrates the periodic sleep time in comparison with the non-periodic sleep

time. As seen the periodic sleep time is almost the half time of that of the device

‘‘live’’. In contrary the non-periodic time, where each node evaluates dissimilar sleep

and active states based on each node�s incoming sleep-history traffic, shows a signif-

icant reduction in active period and an increase in the mean sleep time. During sim-

00.10.20.30.40 50.60.70.8

0 45 150

500

790

820

1000

3000

5000

7000

9000

1100

0

1230

0

1290

0

1300

0

1500

0

1700

0

1950

0

2050

0

Simulation time

% p

erio

dic

vs n

on-p

erio

dic

slee

p

non-periodic sleep time periodic sleep time

Fig. 6. Comparison of periodic sleep time with non-periodic (unlimited capacity).


ulation time it was shown that sleep time should not increase more than 71% at atime from previous measure (as peak, and not simultaneously but in different slots)

because this will cause network partitioning and routing lock [18]. According to Fig.

6 sleep time increases by almost 10% mean (non-periodic time) in comparison with

the sleep time occurring on a periodic basis.

The impact of caching on load at any time during simulation plays a major role

for the performance of any wireless network. Fig. 7 shows the average hop number

at any time during simulation, with the cached information. It is clearly shown that

for small number of hops the cached information is negligible (in terms of few Kbs).On the other hand for relatively higher number of hops the average cached informa-

tion in KB are increasing, but not exceeding the 3.87MB (extremely high traffic-net-

work splits). Of course these measures have been taken for unlimited capacity used

for caching in each node�s buffer.Fig. 8 clearly shows the frequency of caching technique necessary for saving pack-

ets destined for a proper node, which prevents the loss of packets. It must be stated

that this particular measure was extracted several times to prove the fact that for

nodes with high connectivity the caching frequency becomes higher than for nodeswith low number of neighbors. Having 0.3 as a mean value of the sleep time, the

02468

10121416

2 5 7 9 14 16 50 100

160

220

250

270

300

512

640

1152

1664

2176

2688

3200

3712

Cached information (KB)

Hop

num

ber

Fig. 7. Illustration of the average hop number with respect to cached information.

0

0.1

0.2

0.3

0.4

0.5

0.6

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

Number of nodes

% c

achi

ng f

requ

ency

Fig. 8. How frequently each node caches information for any other node (depending on network topology

and node�s mobility in the presence of excessive traffic).


node which caches packets destined for a node laying in the sleep state, saves the

packets by using the caching technique.

Figs. 9 and 10 show the mean number of hop count at any time in the network

and the mean number of nodes ‘‘laying’’ in the sleep state respectively. It is remark-

able to point out that nodes in order to prevent network partitioning in the sleep

state must not exceed the limit of 10.

From Fig. 11(a) it is indicated that unlimited capacity in each device could enable

vulnerabilities for network connectivity. Roughly speaking when memory is unlim-ited, after some simulation steps (by reaching steady state [18,19]) nodes will enter

in a higher duration sleep time period. As a result network could split in parts (par-

titioning) causing significant reduction in performance, even if a limited number of

nodes are allowed to enter in the sleep state. Hence a limited capacity for caching

information destined for other nodes as shown in Fig. 11 might offer better connec-

tivity and in turn network partitioning prevention.

For 64, 128, 256, 512KB cached capacity, it can be seen that for non-periodic

sleep time duration ADCEC offers higher connectivity even in the presence of extre-

0

1

2

3

4

5

6

7

0 45 150

500

790

820

1000

3000

5000

7000

9000

1100

0

1230

0

1290

0

1300

0

1500

0

1700

0

1950

0

2050

0

Simulation Time

Mea

n nu

mbe

r of

hop

cou

nt

Fig. 9. Mean number of hop count at any time in the network.

2

4

6

8

10

12

14

0 45 150

500

790

820

1000

3000

5000

7000

9000

1100

0

1230

0

1290

0

1300

0

1500

0

1700

0

1950

0

2050

0

Simulation Time

Num

ber

of n

odes

at

slee

p st

ate

Fig. 10. Mean number of nodes ‘‘laying’’ in the sleep state (excessive traffic).

0102030405060708090

100

Unlimited 64KB 128KB 256KB 512KB 1MB 2MB

Cached capacity

Net

wor

k C

onne

ctiv

ity

Lik

elih

ood

0102030405060708090

100

Unlimited 64KB 128KB 256KB 512KB 1MB 2MB

Cached capacity

Mea

n ut

iliza

tion

of

WD

%(a

ctiv

e m

ode)

Non-periodic sleep Periodic sleep

(a)

(b)

Fig. 11. (a) Network connectivity maintenance with respect to different capacity measures. (b) Mean

utilization of nodes while in active state with respect to different capacity measures.


mely high traffic. Thus in the implementation of the described scenario when latency

for transmitting the corresponding packets to destination increases, then a limitation

in sleep time duration must be placed for all nodes. Fig. 11(a) basically ensures the

fact that unlimited memory for caching is not useful particularly for dynamically

changing network. Thus placing a limit on caching capacity would enable higher

0102030405060708090

100

0 45 150

500

790

820

1000

3000

5000

7000

9000

1100

0

1230

0

1290

0

1300

0

1500

0

1700

0

1950

0

2050

0

Simulation time

Fra

ctio

n of

rem

aini

ng E

nerg

y%

periodic sleep-wake

non-periodic sleep-wake

Fig. 12. Fraction of % remaining energy.

0102030405060708090

100

0 45 150

500

790

820

1000

3000

5000

7000

9000

1100

0

1230

0

1290

0

1300

0

1500

0

1700

0

1950

0

2050

0

Simulation time

Dat

a de

liver

y ra

tio

%

periodic sleep-wake

non-periodic sleep-wake

Fig. 13. Data delivery ratio during simulation time.


network connectivity and better battery utilization by increasing the average sleep

time duration. In Fig. 11(b) it becomes evident that by limiting the cached capacity

in every node the utilization becomes better in active periods for forwarding activity

etc.

After exhaustive simulations it is beneficial to examine the fluctuations of remain-

ing energy at every time during simulation. Fig. 12 shows the fraction of the percent-

age of the remaining energy. It is obvious that when system enters the steady state,

remaining energy reduces progressively during simulation. The network traffic ischaracterized as constant bit rate (CBR) moderate flows where each source node

transmits one 512-bytes (�4Kbits) packet. It is remarkable to figure out that remain-

ing energy for non-periodic sleep time, offers potentially more battery ‘‘life’’ for mo-

bile users.

Fig. 13 shows the data delivery ratio during simulation time. At any time during

simulation network�s nodes show a resistance in packet loss which offers remarkable

data delivery ratio. It is also obvious that for long time after transient state the non-

periodic sleep method marks a data delivery ratio over 71% which is satisfactory


enough particularly for infrastructureless networks where mobility induces frequent

route changes, resulting network partitioning.

Mean utilization of nodes while in active state with respect to the number of flows

is shown in Fig. 14. When packet flow is in low levels network�s nodes are not uti-

lized at all. On the other hand when packet flow is high enough (approaching a 50packet flows at each time step after overcoming the transient state) utilization is

obviously reaching a mean of 53% at each node. Similarly for large scale networks

our further experiments done showed that nodes are utilized during caching and

routing process at 49% even in the presence of excessive traffic.

The average throughput of the system is shown in Fig. 15 for moderate, light and

heavy load. Moderate load seems to reach maximum system�s throughput response.Overall throughput of the system is estimated as the total linkutilization Æ •(1 � Ploss)

and is shown in Fig. 16. Finally in Fig. 17 the mean energy consumption is presentedat any time in the network during simulation.

010

2030

405060

7080

90100

0 10 15 20 22 30 42 45 50 55

Number of flows

Uti

lizat

ion

%

Fig. 14. Mean utilization of nodes while in active state with respect to the number of flows.

0100200300400500600700800900

2 3 4 5 6 7 8 9 10 11 12

Number of Hops

Thr

ough

put

(kbp

s)

Moderate load light load Heavy load

Fig. 15. Average throughput measure for moderate, light and heavy load in links.

00.20.40.60.8

11.21.41.61.8

2

2 3 4 5 6 7 8 9 10 11 12

Number of hops

Thr

ough

put

(M b

ps)

Fig. 16. Overall throughput for moderate traffic during simulation time with respect to the number of

hops.

0200400600800

100012001400160018002000

0 45 150

500

790

820

1000

1200

1400

1600

2120

2600

3000

3800

5000

7000

9000

1100

0

1230

0

1290

0

1300

0

1500

0

1700

0

1950

0

2050

0

Simulation time

Ove

rall

EC

(µW

)-m

ean

valu

es

Fig. 17. Mean energy consumption in the network at any time during simulation.


5. Conclusions and further research

In this paper, we have implemented an Adaptive Dynamic Caching Energy Con-

servation (ADCEC) method based on each node�s incoming sleep-history traffic,

which bounds a partially asynchronous operation where each node evaluates dissim-

ilar sleep-wake schedules. One basic issue outlined for conserving energy was the

association of cached packets with the next sleep time duration of the intended node.

Results have shown that unlimited memory for caching in wireless networks is not

useful compared with certain caching capacities and specified mobility frequency.Additionally it was shown that ADCEC method would enable a significant increase

in the total average sleep time duration, the same time keeping network connectivity

at high levels (reliability) which results in remarkable energy conservation. We have

shown that the proposed method could be applied to infrastructureless networks

with any underlying routing protocol support to provide independency and portabil-

ity (routing protocol independency) where needed in order to overcome network par-

titioning problems, the same time conserving energy.

If a wireless device needs to be connected with internet via other wireless networksor directly, a scope of interest could be the capabilities of such an enterprise to


service mobile users on demand. This research could be extended by using a web

caching mechanism, in order to temporary store in wireless devices the intended

web pages. The main challenge in wireless multi-hop ad-hoc networks is the efficient

routing problem, which is aggravated by the node mobility. Therefore a routing tech-

nique that would enable an efficient routing scheme for web information retrieval bywireless devices for infrastructure-based mode or even by using the ad-hoc-based

mode becomes necessity with the tremendous growth of mobile users intending to

access the web. With the proposed scheme extended, the web caching mechanism will

enable portability to access any time on-demand the internet. Furthermore a scope

of interest would be the exploration of an agent-based approach [18] for web-caching

mechanism where an asynchronous sleep state will also exists (using epochs [22] for

activation period).

Acknowledgment

Constandinos X. Mavromoustakis is partially funded by the A.G. Leventis

Foundation.

References

[1] L. Feeney, M. Nilsson, Investigating the energy consumption of a wireless network interface in an

ad hoc networking environment, Proceedings of IEEE InfoCom 5 (8) (2001) 1548–1557.

[2] A. Porret, T. Melly, C.C. Enz, E.A. Vittoz, A low-power low-voltage transceiver architecture suitable

for wireless distributed sensors network, Proceedings of IEEE International Symposium on Circuits

and Systems�00 Geneva 1 (2000) 56–59.

[3] B. Liang, Z.J. Haas, Optimizing Route-cache Lifetime in Ad Hoc Networks, InfoCom 2003, pp. 233–

244.

[4] S. Banerjee, A. Misra, Minimum energy paths for reliable communication in multi-hop wireless

networks, in: ACM Symposium on Mobile Adhoc Networking and Computing (MOBIHOC 2002),

Lausanne, Switzerland, June 9–11 2002.

[5] S. Doshi, T.X. Brown, Minimum energy routing schemes for a wireless ad hoc network, Proceedings

IEEE InfoCom 2002, pp. 212–220.

[6] Z.J. Haas, M. Pearlman, The performance of query control schemes for the zone routing protocol,

ACM/IEEE Transactions on Networking 9 (4) (2001) 427–438.

[7] S. Singh, C. Raghavendra, PAMAS: power aware multi-access protocol with signalling for ad hoc

networks, SIGCOMM Computer Communication Review 28 (3) (1998) 5–26.

[8] Y. Xu, J. Heidemann, D. Estrin, Geography-informed energy conservation for ad hoc routing, in:

Proceedings of the ACM/IEEE International Conference on Mobile Computing and Networking

(2001), pp. 70–84.

[9] B. Chen, K. Jamieson, H. Balakrishnan, R. Morris, Span: an energy-efficient coordination algorithm

for topology maintenance in ad hoc wireless networks, in: Proceedings of MobiCom�2001, Rome,

Italy, July 2001, pp. 85–96.

[10] J. Kulik et al., Negotiation-based protocols for disseminating information in wireless sensor

networks, Wireless Networks 8 (2–3) (2002) 169–185.

[11] S. McCanne and S. Floyd, NS network simulator version 2.1b8, June 2001, Available from: <http://

www-mash.cs.berkeley.edu/ns/>.

[12] Z.J. Haas, A new routing protocol for the reconfigurable wireless networks, in: Proceedings of

theIEEE ICUP�97, 1997, pp. 108–116.

http://www-mash.cs.berkeley.edu/ns/

http://www-mash.cs.berkeley.edu/ns/


[13] K. Park, G. Kim, M. Crovella, On the relationship between file sizes, transport protocols and self-

similar network traffic, in: ICNP �96, 1996.[14] S. Robert, J.-Y. Le Boudec, New models for pseudo self-similar traffic, Performance Evaluation 30

(1997) 57–68.

[15] P. Ulanovs, E. Petersons, Modeling methods of self-similar traffic for network performance

evaluation, in: Scientific Proceedings of RTU, Series 7. Telecommunications and Electronics, 2002,

pp. 40–49.

[16] Z.J. Haas, M.R. Pearlman, The zone routing protocol (ZRP) for ad hoc networks, Internet Draft,

draft-ietf-manet-zone-zrp-02.txt, June 1999.

[17] M.R. Pearlman, J. Deng, B. Liang, Z.J. Haas, Elective participation in ad hoc networks based on

energy consumption, in: Proceedings IEEE GLOBECOM 2002, Taipei, Taiwan 17–21 November

2002, pp. 119–127.

[18] C.X. Mavromoustakis, H.D. Karatza, Agent-based throughput response in presence of node and/or

link failure (on demand) for circuit switched telecommunication networks, Computer Communica-

tions 27 (3) (2004) 230–238.

[19] C.X. Mavromoustakis, H.D. Karatza, On the extensibility properties and performance measures of

circuit switched telecommunication networks, using agent-based distributed routing algorithm, in:

Proceedings of 2003 International Symposium on Performance Evaluation of Computer and

Telecommunication Systems—SPECTS, Montreal, Canada, July 20–24 2003, pp. 240–247.

[20] C.X. Mavromoustakis, H.D. Karatza, Split agent-based routing in interconnected networks,

International Journal of Communication Systems (IJCS) 17 (2004) 303–320.

[21] C.X. Mavromoustakis, H.D. Karatza, Adaptive energy conservation model using dynamic caching

for wireless devices, in: To appear in the Proceedings of 37th Annual Simulation Symposium (ANSS),

IEEE Computer Society, Arlington, VA, April 18–22, 2004, pp. 257–264.

[22] H.D. Karatza, R.H Hilzer, A simulation-based performance analysis of epoch load sharing in

distributed systems, Simulation: Transactions of the Society for Modeling and Simulation

International 78 (7) (2002) 461–471.

adaptive traffic-based control method for energy conservation in wireless devices

Documents