adaptive traffic-based control method for energy conservation in wireless devices
TRANSCRIPT
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).
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
216 C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232
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.
C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232 217
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.
218 C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232
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.
C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232 219
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.
220 C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232
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.
C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232 221
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.
222 C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232
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.
C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232 223
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.
224 C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232
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).
C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232 225
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).
226 C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232
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.
C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232 227
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.
228 C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232
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
C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232 229
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.
230 C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232
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
C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232 231
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.
232 C.X. Mavromoustakis, H.D. Karatza / Simulation Modelling Practice and Theory 13 (2005) 213–232
[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.