Experience Based Energy Efficient Reactive

Routing Protocol (EXERP) for Mobile Ad-

Hoc Networks

Anuradha Banerjee, Paramartha Dutta

& Subhankar Ghosh

Arabian Journal for Science and

Engineering

ISSN 1319-8025
Volume 39
Number 2

Arab J Sci Eng (2014) 39:891-901

DOI 10.1007/s13369-013-0813-9

1 23
Your article is protected by copyright and
all rights are held exclusively by King Fahd
University of Petroleum and Minerals. This e-
offprint is for personal use only and shall not
be self-archived in electronic repositories. If
you wish to self-archive your article, please
use the accepted manuscript version for
posting on your own website. You may
further deposit the accepted manuscript
version in any repository, provided it is only
made publicly available 12 months after
official publication or later and provided
acknowledgement is given to the original
source of publication and a link is inserted
to the published article on Springer's
website. The link must be accompanied by
the following text: "The final publication is
available at link.springer.com”.

1 23
 Author's personal copy
Arab J Sci Eng (2014) 39:891–901
DOI 10.1007/s13369-013-0813-9

RESEARCH ARTICLE - COMPUTER ENGINEERING AND COMPUTER SCIENCE

Experience Based Energy Efficient Reactive Routing Protocol

(EXERP) for Mobile Ad-Hoc Networks
Anuradha Banerjee · Paramartha Dutta ·
Subhankar Ghosh

Received: 21 January 2012 / Accepted: 10 August 2012 / Published online: 20 September 2013
© King Fahd University of Petroleum and Minerals 2013

Abstract A mobile ad hoc network is an infrastructureless battery exhaustion that saves cost of flooding route_request
network where the nodes are free to move independently packets which would have been otherwise required to repair
in any directions. As the nodes have limited battery power the broken links. Less message cost generates less signal col-
so we require energy efficient route discovery scheme. This lision and also saves battery power of nodes preserving the
method reduces the cost of messages and energy consump- network connectivity as much as possible. Hence data packet
tion in nodes, thereby increasing the network throughput. delivery ratio increases up to a great extent. All these empha-
In mobile communication, an ad hoc network refers to a size the efficiency of EXERP over other energy-efficient
network connection recognized for a single session. During protocols.
route discovery phase, generally the route request packets
generated by the source arrive at the destination through mul- Keywords Battery power · Optimal path · Multiple path ·
tiple paths. EXERP assigns weights to those paths depend- Flooding
ing on the expected duration of their links, number of links
present in the path and energy efficiency of the routers. One
of the routes with highest weight is elected as optimal path
for communication between the corresponding pair of source
and destination nodes. Simulation results clearly establish
the fact that communication through energy-efficient link-
stability conscious routes significantly reduces the message
cost in the network. The main reason behind this improve-
ment is reduction in link breakage due to node movement and

A. Banerjee
Department of Computer Applications,
Kalyani Government Engineering College,
Kalyani, WB, India
e-mail: [email protected]

P. Dutta
Department of Computer and System Sciences,
Visva-Bharati University, Santiniketan, WB, India
e-mail: [email protected]

S. Ghosh (B)
Department of Computer Science and Engineering,
Regent Education and Research Foundation,
Barrackpore, WB, India
e-mail: [email protected]

123
 Author's personal copy
892 Arab J Sci Eng (2014) 39:891–901

1 Introduction The better the experience, the higher will be the weight of
the link in the present communication session. Among the
An ad hoc network is a group of wireless mobile devices various paths through which the route request arrives at the
or nodes that communicate with each other in a collabora- destination, any one with the highest path weight is elected
tive way over multihop wireless links without any station- as optimal and chosen for communication between the pair
ary infrastructure or centralized management. These net- of source and destination nodes. The caches are used in each
works are deployed mainly in the battle fields and disas- node for storing communication experience with frequently
ter situations, such as earthquake, floods, etc. Many routing communicated nodes. EXERP uses the concept that behav-
protocols have been proposed for ad hoc networks. They iour of ad hoc network nodes is modelled based on human
can be categorized as proactive and reactive routing proto- connectivity network [8] where each node regularly meets
cols. In proactive routing protocols, the nodes proactively some familiar stranger nodes (travelling to/from workplace
store route information to every other node in the network. while at work, living in same neighbourhood, etc.). The non-
Hence, they suffer from extremely huge storage overhead, random component of node mobility that prevails in many
because they store information about both active and inactive scenarios has been modelled as a list of stable positions and
routes. Reactive or on demand routing protocols are designed their associated time stamps [‘q’] in the node caches.
to reduce this overhead [1–7]. Among the reactive routing
protocols, some are energy constrained, for example, Mini-
mum Total Transmission Power Routing (MTPR) [1], Max- 2 Related Work
imum Battery Capacity Routing (MBCR) [2], Conditional
Min–Max Battery Cost Routing (CMMBCR) [3], Location Several energy constrained routing protocols have been pro-
Based Power Conservation [4], Hint-Based Power Efficient posed in the literature. Among them, we are discussing a few
Routing (HPER) [5], QoS Enabled Power Aware Routing state of the art protocols. In minimum transmission power
(QEPAR) [6], Energy-Efficient Ad Hoc on-Demand Routing routing (MTPR) [1], energy consumed per hop is used as
(EEAODR) [7], etc., are mentionworthy. Design of a Power the link metric. MTPR selects the route with minimum total
aware routing protocol for ad hoc networks is a state of the transmission power, as the optimal one for communication.
art problem as nodes run on batteries that have limited power. Maximum battery capacity routing (MBCR) [2] aims for a
High energy consumption in nodes results in high death rate route with maximum remaining battery capacity. Min–max
of nodes and frequent link breakages. To repair those links, battery cost routing (MMBCR) [3] assigns the battery power
more route request packets have to be injected into the net- of a route to the minimum residual battery power of a node
work that consume more battery power driving nodes speed- (bottleneck node) along the route. The designed route is then
ily to complete exhaustion. The present article proposes an the route with maximum battery power. If there is a tie,
Experience-based Energy-efficient Reactive Routing Proto- MMBCR chooses the route with shortest hop count. When
col (EXERP) that significantly reduces energy consumption all nodes in the network have almost identical residual bat-
during route discovery by electing a stable as well as agile tery power, MMBCR would result in frequent route changes.
path as the optimal one for communication between a given This is because the algorithm is very sensitive to even slight
pair of source and destination nodes. During route discovery changes in the residual battery power of nodes, and path
phase, the source node injects route request packets into the selection offer has to be done using the secondary criteria of
network. The neighbours of the source receive this packet hop count. The Conditional Min–Max Battery Cost Routing
and broadcast it again within their respective radio-ranges. (CMMBCR) [3] is a hybrid protocol that arbitrates between
The process of forwarding route request goes or till the des- MTPR and MMBCR. A threshold is defined in this context
tination is reached or the route request packet traverses the that ranges between 0 and 100. If the maximum battery capac-
maximum possible number of hops (popularly termed as ity is more than the threshold, then CMMBCR behaves like
hop count) allowed for the network. Generally, the route MTPR otherwise behaves like MMBCR. A Location-Based
request packet reaches the destination through multiple paths. Power Conservation scheme (LBPC) [4] is proposed in which
EXERP assigns weights to these paths depending on weight a transmitting node utilizes location information of its first
of the links. Weight of a link from a router on source n p hop neighbors to adjust its transmission range. However, it
to another router on destination n q depends on the data or does not take care of remaining energy of nodes. A QoS
control packets’ send/receive experiences between n p and Enabled Power Aware Routing Protocol (QEPAR) is pro-
n q in the present as well as earlier communication sessions. posed in [6]. Its main objective is to provide QoS in terms of
The communication experiences consist of duration of links power and bandwidth and to increase the throughput by find-
in those sessions, power of received packets compared to the ing out the optimal path from source to destination. It is based
threshold receive power of n q , residual energy, rate of energy on a table driven approach in which each node maintains a
depletion and pending message forwarding load of n p and n q . neighboring node table (NNT) containing the information

123
 Author's personal copy
Arab J Sci Eng (2014) 39:891–901 893

regarding the nodes falling in its velocity. Each node in the it is expected that the link will again survive for a similarly
network broadcasts a Beacon request message to retrieve the long time in subsequent communication session between the
information regarding the available bandwidth and battery same pair of nodes) and congestion load already present in
power. Message packets are not forwarded to the nodes that message queues of nodes. These factors also contribute to
fall short of bandwidth and battery power. An efficient cost balance message forwarding load among nodes in the net-
function is proposed by Lotfi et al. [9], which prevents mes- work. EXERP also does not require any assistance from any
sage traffics to be sent through the nodes with low energy and backbone infrastructure and it is completely distributed in
more buffered packets. A signal strength-based energy effi- nature. It is intelligent in the sense that it acquires knowl-
cient routing technique is introduced in [10] while proposing edge from earlier communication experience and utilizes it
a route discovery scheme with transmission power control. in selection of optimum routes in future. The function that
When a node receives a route request (RREQ) packet, it com- is used to elect the optimum route considers the number of
putes the routing level back off time as being inversely pro- links in the route compared to the maximum allowable hop
portional to the received power of the RREQ. After the route count in the network and also performance of the current
discovery, source and intermediate nodes transmit packets route with respect to all other routes available for communi-
by the power controlled MAC protocol. An Energy Efficient cation between same pair of nodes. Overall, we can say that
On Demand Routing Protocol (EEAODR) is proposed in [7] EXERP is energy-efficient load balancing, previous experi-
that tries to balance energy load among nodes so that a min- ence conscious and rigorous.
imum energy level is maintained and network longevity is
increased. A HPER protocol was proposed in [3], where the
hint value of a link from a node n p to another node n q is com- 3 Problem Definition and Proposed Solution
puted based on the time of first and last bit of n p as received
by n q and time of breakage of last time. On receiving a packet We consider an ad-hoc network composed of z number of
destined to node n d , a node n i forwards it to the neighbour nodes, in which a node n q may be considered as neighbour
with the best hint and best power value. However, all the of a node n p at time t provided n q lies within the radio range
above protocols completely ignore the frequency of break- of n p at time t. Each node broadcasts HELLO message at
age of link from n p to n q in either data/control interactions regular interval c within its own radio range. The nodes that
or the rate of energy depletion of n p and n q . reside within radio range of n p at time t receive the HELLO
The energy efficient reliable routing protocol proposed in message of n p and reply with acknowledgement (ACK) mes-
[13] is based on backbone nodes and is incorporated in the sage. If the link from n p to n q is established at time t1 and
AODV protocol. Inclusions of the backbone nodes incur extra broken at t2 , then the start time and end time of that link ses-
overhead. In [15] an Adaptive Energy-Efficient and Reliable sion will be at t1 and t2 , respectively. The link may again be
Gossip Routing Protocol (AEERGRP) is proposed in which formed sometime later, which will be a new session of the
a counter is set to represent the current number of neighbours link. The communication primitives of different messages
at each node which are in active (or alive) state. The value of are as follows:-
the counter is adaptively adjusted based on packet delivery
ratio to reduce the energy consumption in ad hoc network (a) HELLO ( p, t, E p , e p (t), L p , l p (t), time−star t p): The
communication. In Jin and Jo [14] proposed a power effi- HELLO message is transmitted by node n p at time t. E p
cient reliable routing protocol that includes the advantages of and e p (t) denote the maximum battery power and con-
on-demand protocols and claims to provide reliable packet sumed battery power till time t, of node n p respectively.
transmission. It applies a cost function to select the opti- L p and l p (t) denote the maximum number of available
mum path considering minimum residual energy of the nodes locations and number of filled locations at time t in the
in a path, total energy consumed by a path to transmit and message queue of node n p . Message queue is the queue
process a packet and the paths stability in accordance with where pending message forwarding requests are main-
node mobility. Our proposed scheme EXERP is based on the tained.
well-established fact that like human social networks, nodes (b) ACK (q, p, t, t1 , E q , eq (t1 ), L q , lq (t1 ), time −star t q):
in ad hoc network generally communicate with a fixed set of The acknowledgement message is transmitted by n q for
nodes [12]. This is extremely important from the perspective n p at time t1 acknowledging the HELLO message trans-
of behavioral aspect of source nodes in choosing a destina- mitted by n p at time t. Significance of all other symbols
tion. EXERP assigns weight to individual routes depending is already explained.
on the residual energy of nodes, their rate of energy deple- (c) Route-request (s, d, t, p, pr (t1 ), wt(t1 )): The route
tion, history of duration of the ingredients links (if a link sur- request message is emitted by source n s for discover-
vived for a long duration in the earlier communication session ing a route to destination n d . The message is being for-
between the same pair of source and destination nodes, then warded by router n p at time t. n q was the predecessor

123
 Author's personal copy
894 Arab J Sci Eng (2014) 39:891–901

of the router n p and wt(t1 ) is the set of weights of the exists much similarity between the behaviour of ad hoc net-
links from n s to n p at time t1 . If p = s the wt (t1 ) = −1, work nodes and human connectivity network [8], where each
because in that case there cannot be any predecessor n q node is expected to meet regularly some familiar stranger
to n p . pr (t1 ) is the set of predecessors of node p in the node (travelling to/from work, while at work, living in the
current route at time t1 . If p = s, then pr (t1 ) = −1. same neighbourhood, etc.). In EXERP, each node stores the
(d) Route-reply (d, s, t, R): The route reply is sent by des- communication history of these familiar stranger nodes dis-
tination n d to source n s at time t1 . It also informs n s tributed in two caches, namely H-CACHE (history cache)
that the optimal path for communication from n s to n d and P-CACHE (packet cache). Details of their structures are
is R. R contains the identification number; x-coordinate presented in the section III.A. Components of the communi-
and y-coordinate position of all routers in the optimal cation history are duration of links between involved nodes in
path, in the order in which they appear consecutively in present and earlier link sessions, ratio of actual receive power
the optimal route from n s to n d . If the optimal route is and receive threshold power of the corresponding receiver,
n s → n i → n j → n k → n d , then R = i, xi , yi , j, n j , residual energy, rate of energy depletion and pending mes-
y j , k, xk , yk . i is the identification number of node n i . sage forwarding requests in the message queues of involved
(xi , yi ) is the ordered pair describing the last known geo- nodes.
graphical position of n i in terms of x and y coordinates, The present problem is divided into three steps:-
respectively. Similar kind of information of subsequent
routers n j and n k also appears in R. i. Assigning weights to individual links in the routes
(e) Send-msg (s, d, i, t, m, j, R): The message m has been through which route request packets arrive at the des-
generated by source node n s for destination n d . The mes- tination from source.
sage is being forwarded by node n i at time t to its next ii. Computation of weights of the paths of traversal of route
router n j in route R. requests.
(f) Ack-msg (i, j, s, d, t): Node n i (router/destination) iii. Election of optimal path for communication between a
acknowledges receipt of the messages from its uplink given pair of source and destination nodes.
neighbour n j (source/router). The message was gener-
ated by source n s for destination n d at time t. 3.1 Description of H-CACHE and P-CACHE
Radio-range of any node n p is denoted as R p and its set
of neighbours as N p (t). The link between n p and one These two cache memories H-CACHE and P-CACHE are
of its neighbours n q will be termed as active at time t embedded in each node. Attributes of H-CACHE are node-
provided the distance D pq (t) between them is less than id, session-no, start time, end time and node-last access time,
or equal to R p . P-CACHE-loc start and P-CACHE-loc end. Significance of
Mathematically, this is expressed as, an entry (i, 1, 25, 44, 71, 3, 8) in the H-CACHE of node n p
indicates that first session of the link from n i to n p started at
D pq (t) ≤ R p or time-stamp 25, ended at time-stamp 44; the record was last
 read or modified at time-stamp 71 and information about the
(x p (t) − xq (t))2 + (y p (t) − yq (t))2 ≤ R p (t) (1) packets transmitted from n i to n p in that session is stored in
P-CACHE of n p from location 3 to location 8. P-CACHE
If the above condition is false, then we say that the wire- is a single attribute cache and name of the attribute is act-
less link between n p and n q is broken at time t. In order recv-power. If the value in the location 5 of P-CACHE of
that a node receives a packet properly the node needs to n p is 55, then it denotes that the third packet (information
receive the packet with a minimum predefined power, termed about the first packet is in location 3 of P-CACHE of n p )
as receive threshold power. The power with which a node transmitted from n i to n p in first link session was received by
actually receives a packet is called actual receive power. n p with actual receive power 55 units. Entries in H-CACHE
The better the ratio of actual receiving power and receive are stored in increasing order of (node id, session no.). The
threshold power, the better it is for the receiver. If a node n j fields end time and node last access time are set to −1 for
stays within the radio-range of another node n i at a time t, those sessions who have not ended yet. After completion of
then n i is termed as the n p link neighbour of n j and n j is those link sessions, their end time and node last access time
termed as the downlink neighbour or simply neighbour of n i both are set to current time-stamp. If more than one tuple exist
at that time. If n j receives a packet from n i with high actual for node n i in H-CACHE of n p and one of these corresponds
receive power at time t then it indicates that n i and n j are to an ongoing session, then none of the tuples associated with
in close proximity at time t. Please note that close proxim- n i can be deleted from H-CACHE of n p . If more than one
ity between consecutive routers in a path also reduces the tuple exist for n i in H-CACHE of n p , then node last access
delay along the path. EXERP utilizes the concept that there time attribute of those tuples is set to the maximum end time

123
 Author's personal copy
Arab J Sci Eng (2014) 39:891–901 895

among the tuples associated with n i , excluding those with Table 3 H-CACHE and P-CACHE configuration for 2nd link session
end time −1. from n i to n j
Idle time idli ( p, t) of node n i in the caches of n p at time
H-CACHE i 1 25 39 64 0 3
t is defined below:
⎧ i 2 56 64 64 6 7
⎨Z if there is no active link from n j to k 1 51 55 55 4 5
idli ( p, t) = n p at time t (2)
⎩ P-CACHE 10 25 15 19 28 20 22 17
−1 otherwise 0 1 2 3 4 5 6 7
where Z = t-node_last access timei ( p, t) and.
Node last access timei ( p, t) denotes the value of the
Table 4 H-CACHE and P-CACHE configuration for 3rd link session
attribute node last access time of any record corresponding from n i to n j
to n i in the cache of n p at time t. EXERP sets a threshold
ϒ(ϒ = −1) on the idle time of node records in the caches. H-CACHE i 1 25 39 70 0 3
If idli ( p, t)>= ϒ for some node n i , then all session entries i 2 56 64 70 6 7
of n i are deleted from cache memories (both H-CACHE and i 3 66 70 70 8 8
P-CACHE) of n p . This eliminates the obsolete importance k 1 51 55 55 4 5
of links. P-CACHE 0 25 15 19 28 20 22 17 11
0 1 2 3 4 5 6 7 8
3.1.1 Example

The record storage in H-CACHE and P-CACHE of node n j Table 5 H-CACHE and P-CACHE configuration for 4th link session
from n i to n j
from the following scenario. Initially, both H-CACHE and
P-CACHE are empty. n i and n j begin operation with full
H-CACHE i 1 25 39 70 0 3
battery capacity at timestamps 10 and 20, respectively.
i 2 56 64 70 6 7
1st link session from n i to n j :
i 3 66 70 70 8 8
The session begins at time-stamp 25, ends at time-stamp
i 4 73 −1 −1 9 11
39. Four data packets are transmitted by n i in this session with
k 1 51 55 55 4 5
actual receive power 10, 25, 15 and 19 units, respectively.
P-CACHE 10 25 15 19 28 20 22 17 11 8 12 15
Corresponding H-CACHE and P-CACHE configuration is
shown in Table 1. 0 1 2 3 4 5 6 7 8 9 10 11
1st link session from n k to n j :
The session begins at time-stamp 51, ends at time-stamp
55. Two data packets were sent from n k to n j with actual 2nd session of link from n i to n j :
receive power 28 and 20, respectively. Corresponding H- The session begins at time-stamp 56 and ends at time-
CACHE and P-CACHE configuration are shown in Table 2. stamp 64. Two data packets were sent from n i to n j with
The record of n k appears below the record of n i since actual receive power 22 and 17, respectively. Correspond-
i < k. ing H-CACHE and P-CACHE configurations are shown in
Table 3. Since session 2 has already ended, so node last access
time of all records of n i in caches of n j will be set to end
Table 1 H-CACHE and P-CACHE configuration for 1st link session
from n i to n j time of session 2.
3rd session of link from n i to n j :
H-CACHE i 1 25 39 39 0 3 The session begins at time-stamp 66 and ends at time-
P-CACHE 10 25 15 19 stamp 70. Only one data packet was sent with actual receive
0 1 2 3 power 11. Corresponding H-CACHE and P-CACHE config-
urations are shown in Table 4.
4th session of the link from n i to n j :
Table 2 H-CACHE and P-CACHE configuration for 1st link session The session begins at time-stamp 73 and has not ended
from n k to n j
yet. Three packets have been sent by n i so far with their
H-CACHE i 1 25 34 0 3
actual receive power 8, 12 and 15, respectively. Correspond-
k 1 51 55 4 5
ing H-CACHE and P-CACHE configurations are shown in
Table 5.
P-CACHE 10 25 15 19 28 20
End time of H-CACHE entry on 4th session of time from
0 1 2 3 4 5
n i to n j is set to −1 since the session has not ended yet.

123
 Author's personal copy
896 Arab J Sci Eng (2014) 39:891–901

Similarly the attribute node last access time of this session is = act_recvi j ( p1) − thres_recv j , f 3i j (t) = (αi (t) x
also set to −1. α j (t))/(| Si j (t) | x βi (t) x β j (t) x γi (t) x γ j (t))

3.2 Computation of Link Weight Significance of the used symbols is as follows:-

Si j (t): Set of all link sessions that took place from n i to
Computation of weight of a link through n i to one of its n j till time t
neighbours n j is based on the following heuristics:- end_timei j (s1 ): time of breakage of the link from n i to n j
in link session s1
I. If in the earlier link sessions and in the present session so start_timei j (s1 ): time of establishment of the link from n i
far, the link from n i to n j has remained alive continuously to n j in communication session s1
for a long time, then it is expected that in near future Pi j (s1 ): Set of all data and control packets sent from n i to
also, the link between those two nodes will survive for n j during communication session s1
a significantly long time. This concept is based on the act_recvi j ( p1 ): actual receive power of n j for packet p1
non-random component of mobility behaviour of nodes. sent from n i
II. The lesser the frequency of breakage of link from n i to n j thre_recv j : threshold receive power of n j
in earlier sessions, the better is for durability of the link in αi (t): residual energy quotient of n i at time t
near future. The difference between the two options is as It is mathematically expressed as:
500 sessions of a link each surviving for 2 min and 50 ses-
sions of that link each surviving for 20 min. The second ei (t)
αi (t) = 1 − (4)
option is definitely better. Frequent formation and break- Ei
age of links are uncalled for with respect to reliability of
the link. Values close to 1 increase stability of the link from n i to
III. High remaining energy and low depletion rate of both n i n j . Similarly follows the meaning of α j (t).
and n j are pre requisite for survival of the link from n i βi (t): rate of energy depletion of n i at time t. Its formu-
to n j . lation is as:
IV. If n i sends a message to its neighbour n j and n j does not li (t) + 1
acknowledge its receipt within a predefined time interval βi (t) = (5)
t_time_start + 1
due to its full message queue, then n i resends the message
to n j presuming that the message could not reach n j in time start will be n i started operating in the network with full
the first attempt for some reason. EXERP allows a maxi- buttery capacity. In Eq. (5), 1 is added in numerator as well
mum of 5 such attempts. The phenomenon of resending as denominator to avoid indeterminacy due to the division
the message unnecessarily consumes the energy of n i . by 0.
Moreover, the lesser the message forwarding load present Similarly follows the meaning of βi (t).
in the queues of n i and n j , the lesser will be the energy γi (t): Pending forward request quotient of n i at time t.
depletion rate of n i and n j increasing life of the link Its formulation as follows:
between them as well as agility of the communication.
li (t)
V. If the ratio of actual receive power and receive threshold γi (t) = 1 − (6)
power for n j w.r.t. the n p is higher than time neighbour Li + 1
n i , the clarity of communication due to the close prox- In (6), 1 is added in the denominator to avoid possible division
imity of n i and n j will be improved. Also this proximity by 0, γi (t) also lies between 0 and 1. Values close to 0 indicate
reduces the delay with which a packet arrives at n j from that n i does not have much pending forwarding load which
n i . Weight Wi j (t) of the link from n i to n j at time t is is good for reducing its rate of energy depletion.
formulated as:
Example of Link Weight Computation
Wi j (t) We illustrate below the steps of computation of weight of the
⎛⎛ ⎞⎞ link from n i to n j at time-stamp 80 based on the assumption,
  that E i = 50 units E j = 70 units, ei (80) = 20, e j (80) = 10,
= ( f 1i j (s1) ⎝⎝ ( f 2i j ( p1))/ | Pi j (s1) |⎠⎠ ∗ Q
s1 M p1 N
time_starti = 10, time_start j = 20 and thr e_r ecv j = 5.
 
(3) L i = 20, li (80) = 11, L j = 15, l j (80) = 7. The communi-
cation experience scenario is as per Table 5.
Given that current time is t = 80.
where M = Si j (t), N = Pi j (s1), Q = f 3i j (t), 20 3
f 1i j (s1) = end_timei j (s1)−start_timei j (s1), f 2i j ( p1) Then αi (t) = 1 − = = 0.6
50 5

123
 Author's personal copy
Arab J Sci Eng (2014) 39:891–901 897

Similarly (ii) The lesser the difference between Wmax (R, t),
20 + 1 21 Wmin (R, t), Wavg (R, t) the better is the consistency of
α j (t) = = = 0.296 the route because it is desirable that weights of the links
80 − 10 + 1 71
10 + 1 11 lesser than Wmax (R, t) be very close to Wmax (R, t).
βi (t) = = = 0.18 (iii) To reduce the delay in communication from source to
80 − 20 + 1 61
10 6 destination, the number of hops on links in the route
β j (t) = 1− = = 0.850 should be small. This also contributes to energy effi-
70 7
11 11 ciency of the network since improvement of lesser num-
γi (t) = 1− =1− = 0.476 ber of nodes in a route decreases consumption of energy
20 + 1 21
in the route as well as the network.
7 9
γ j (t) = 1− = = 0.563
15 + 1 16 Example of Route Weight Computation
So, 
Consider a route R from source n s to destination n d , such
(14 × 12.25 + 8 × 14.5 + 4 × 6 + 7 × 6.67) × D that n s → n a → n b → n c → n d . Assume that weights of
wi j (t) =
4 the lines (n s → n a ), (n a → n b ), (n b → n c ) and (n c → n d )
(171.5 + 116 + 24 + 46.69) × D are 23.9, 78.3, 92.5 and 68.9, respectively. Then the maxi-
=
4 mum, minimum and average of these weights are 92.5, 23.9
= 89.55 × 35.97 = 3221.11 [where D = 35.97] and 65.9. If the maximum allowable hop count in the network
 
is 9, then weight wt(R , t) of route R at time t is
Next H-CACHE computation of this weight in the node
lost access time of all i-associated entities will be changed 92.53 (1 − 4/9)
to −1 till the present communication session is over. After wt (R, t) =
(92.5 − 23.9)(92.5 − 68.9)(68.9 − 23.9)
the present communication session is over changes that the (791453.125) × (0.556)
different locations end time of the session are considerable. = = 6.04
(68.6) × (23.6) × (45)

3.3 Computation of Route Weight

3.4 Algorithm of EXERP
We explain below the procedure to compute weight of a route
R: source n s = n i → n i+1 → n i+2 → → n i+K = n d
Procedure r oute_discover y (s, d, t)
destination.
/∗ source n s is trying to find a route to destination n d at time
Let Wmax (R, t), Wmin (R, t), Wavg (R, t) denote the maxi-
t ∗/
mum, minimum and average link weight among all links in
begin
route R, respectively. Then,
f lood_r oute_r equest (s, d, t, s, −1, −1)
Wmax (R, t) = MAX0≤ j≤(K −1) w(i+ j)(i+ j+1) (t) (7) end procedure
Procedure f lood_r oute_r equest (s, d, t, p, pr(t), wt(t))
Wmin (R, t) = MIN0≤ j≤(K −1) w(i+ j)(i+ j+1) (t) (8)
/∗ source n s is trying to find a route to destination n d at time

K −1 t; n p is the current router that will flood the r oute_r equest
Wavg (R, t) = 1/K w(i+ j)(i+ j+1) (t) (9) packet; pr(t) is the set of predecessors of n p in that route at
j=0 time t; Both pr(t) and wt(t) are initialized to −1; wt(t) is the
where K is the number of links in route R. collection of weights of links so far at time t ∗/
Hence weight wt(R, t) of route R at time t is mathemati- begin
cally formulated in Eq. (10). r t_wt = −1
if p = d then
Wmax (R, t)(1 − K /H / ) /∗ destination is not yet reached ∗/
wt (R, t) = (10)
M×N ×T begin
where M = (Wmax (R, t)−Wmin (R, t)), N = (Wmax (R, t)− for each nr  DN p (t)
Wavg (R, t)), T = (Wavg (R, t) − Wmin (R, t)). /∗ DN p (t) is the set of downlink neighbours of n p at time t
H / is the maximum allowable hop count in the network, ∗/
The computation of wt(R, t) in Eq. (10) is based on the begin
following heuristic:- wegt = compute_linkweight(p, r, t)
if pr(t) = {−1} then
(i) The higher the value of Wmax (R, t), the better it is for begin
durability of route R. /* n p is the first router */

123
 Author's personal copy
898 Arab J Sci Eng (2014) 39:891–901

pr(t) = {p} sum = sum + ω

wt(t) = {wegt} Wavg = sum / | wt(t)|
end r t_wt = Wmax ( 1 - hp / H / ) / ((Wmax Wmin ) * (Wavg Wmin )
else * (Wmax Wavg ))
/* n p is not the first router */ return(r t_wt)
begin end function
pr(t) = pr(t) ∪ {p}
wt(t) = wt(t) ∪ {wegt}
end 4 Complexity Calculation
f lood_r oute_r equest(s, d, t, r, pr(t), wt(t))
end 4.1 Space Complexity
end
else The space complexity is mainly due to the storage spaces
begin required for the cache memories H-CACHE and P-CACHE.
/* destination is now reached, so weight of the entire route It is computed based on the following assumptions:
needs to be computed and elect the optimum route */
if r t_wt = −1 then time = 0 (i) On an average, the number of familiar/stranger nodes
if time < tmax do of a node is  and is the average number of packets
begin transmitted per session by any node.
/* tmax is the maximum time for which a destination waits (ii) Information about a maximum of last V communication
to receive some more r oute_r equests from the same source sessions of any node can be stored in H-CACHE. Hence
*/ required size M of H-CACHE is given as:
if r t_wt < compute_r outeweight(s, d, pr(t), wt(t))
r t_wt = compute_r outeweight(s, d, pr(t), wt(t))
M = ×V (11)
time = time + 1
end
if time ≥ tmax then Size Q of P-CACHE is as follows:
send_r oute_r eply(d, s, time, pr(t))
end Q =ψ ×M =ψ × ×V (12)
function compute_linkweight (i, j, t)
/* n i is the predecessor of n j ; t is the current time */ Total size of these two caches is:
begin
sum1 = 0 Z = V (1 + ψ) (13)
for each s1 ∈ Si j (t)
begin So, the size complexity of E 2 R 2 P is:
sum1 = sum1 + (end_timei j (s1) - star t_timei j (s1))
avg1 = sum1 / | Si j (t) |
E 2 R 2 P = O(V (1 + ψ)) = O(V ψ) (14)
sum2 = 0
for each p1 ∈ Pi j (s1)
sum2 = sum2 + (act_r ecvi j (p1) - thr es_r ecv j ) 4.2 Time Complexity
avg2 = sum2 / | Pi j (s1)|
itm = αi (t) * α j (t) / (βi (t) * β j (t) * γi (t) * γ j (t)) Time complexity of EXERP is due to
wegt = avg1 * avg2 * itm
return (wegt) (i) Search for communication history of a node
end function (ii) Computation of link weight
function compute_r outeweight (s, d, pr(t), wt(t)) (iii) Computation of route weight.
begin
hp =| pr(t)| 4.3 Complexity for Searching the Communication History
Wmax = max(wt(t)) of a Node
Wmin = min(wt(t))
/* Wmax and Wmin indicate the maximum and minimum ele- Since communication history of nodes is stored in ascending
ments among weights of the links in the current route */ order of node-id in H-CACHE, EXERP applies binary search
for all ω  wt(t) technique to search the communication history of one par-

123
 Author's personal copy
Arab J Sci Eng (2014) 39:891–901 899

ticular node. On an average, information about number of Table 6 Simulation environment

familiar-stranger nodes can be stored in an H-CACHE. So,
Network area 1,000 m × 1,000 m
the complexity of binary search is O(log2 ). After the first
No. of nodes 50, 100, 200, 500, 1000
second of one particular record is reached, at most V-1 more
consecutive cache records may need to be traversed time V Radio range 20–70 m
is the maximum possible number of communication session Node speed 0–30 m/s
for which information can be stored in cache. Complexity of Mobility model Random walk in various simulation
traversing these consecutive records is O (V). Hence the total Traffic pattern Constant bit rate
complexity for searching communication history of a node Mac protocol IEEE 802.11G
is O(log2 + V ). Simulation time 500 s for each run
Total no. of simulation runs 10
4.4 Complexity of Computing Link Weight

In order to compute weight of a link from n i to n j , n j needs to [12]. Performance of EXERP is compared with other state
access all existing records (at most V number of records) of of the art power efficient routing protocols, such as HPER
n i in its H-CACHE. For each of those records in H-CACHE, (Hint based power efficient routing protocol) [5], QEPAR
on an average number of records is to be read from P- (QoS enabled power aware routing protocol) [6], (S 2 E 2 R)
CACHE. Hence the complexity of computing max weights [10] (Signal strength based energy efficient routing proto-
is O( × V ). col) and CMMBCR (Conditional min-max battery cost rout-
ing) [3], EERRP(Energy Efficient Reliable Routing Protocol
4.5 Complexity of Computation of Route Weight for Mobile) [13], PERRP (Power-Efficient Reliable Rout-
ing Protocol) [14] and AEERGRP(Adaptive Energy Efficient
In order to compute weight of a route, the following steps are and Reliable Gossip Routing Protocol) [15].
executed. The simulation metrics are data-packet delivery ratio,
message cost, energy consumption and end to end delay of
(i) Weights of all links of the route are computed. nodes. Data packet delivery is the ratio of the number of data
(ii) Minimum, maximum and average of those weights are packets successfully delivered to the destination and the total
found out. number of data packets generated. Message cost denotes the
(iii) Route-weight is computed as per Eq. (10) based on the total number of data and control messages injected into the
above information. network. The metric energy consumption specifies the total
consumed energy of all nodes during simulation. End to end
Complexity of computing weight of a link is O( V ). At delay is the time delay required for arriving at the destination
 
most H many links can be present in a route time. H that from source. The simulation results are graphically shown in
is the maximum possible number of hops in a route. So, the Figs. 1, 2, 3 and 4. As the number of nodes in the network

complexity of computing weight of all links is O(H  V ). increases, nodes are blessed with more downlink neighbours
Complexity of finding both the minimum and maximum who improve network connectivity. This increases the data
  
among H number of weights is O(H ). Similarly, O(h ) is
the complexity of finding average of those weights. Provided
all this information is available, route-weight can be com-
puted within O(1) complexity. Hence the overall complexity
 
of computing weight of a route is O(H  V ) + 3O(H ) i.e.

O(H  V ).

5 Simulation Results

Simulation of the mobile network is carried out using Mary-

land [11] Simulator on 800 MHz Pentium IV processor, 40
GB hard disk capacity. The simulation environment appears
in Table 6 in detail. In our simulation environment, choice
of the destination node by each source is based on the
fact that like human social network, nodes in ad hoc net-
works generally communicate with a fixed set of nodes Fig. 1 Percentage of data packet delivery ratio versus no. of nodes

123
 Author's personal copy
900 Arab J Sci Eng (2014) 39:891–901

packet delivery ratio. This is evident from Fig. 1. Compared

to other state of the art routing protocols, EXERP is more
efficient since it considers rate of energy depletion of nodes,
pending load of message forwarding requests and frequency
of link breakages estimated from the earlier communication
experiences. The process of election of optimal path for com-
munication is much more rigorous and energy-efficient in
EXERP compared to other mentioned protocols. This incul-
cates stability and agility in the elected routes. As a result, the
need for route-repair becomes much less and also the injec-
tion of route-request packets to repair established routes can
be avoided. Hence the cost of messages decreases up to a
great extent reducing consumption of energy in nodes. Less
energy consumption increases node life and the phenom-
Fig. 2 Message cost versus no. of nodes enon of link breakage occurring due to node exhaustion also
becomes less frequent. Once again this contributes to the
reduction of route requests flooded to repair broken routes.
The improvement generated by EXERP over the other energy
efficient protocols from the perspective of message cost and
energy consumption is evident from Figs. 2 and 3. Less mes-
sage cost indicates less signal contention and collision. Hence
the message by senders and phenomenon of retransmission
and reforwarding of messages by routers reduce up to a great
extent, reducing the end to end delay of communication. This
can be noticed from Fig. 4.

6 Conclusion

This article produces an experience-based energy-efficient

Fig. 3 Energy consumption versus no. of nodes routing protocol EXERP that utilizes the similarity between
the mobility behavior of ad hoc network nodes and human
connectivity network. Weight of a route is computed from
weights of individual links. Weight of a link depends on the
communication experiences of involved nodes in earlier ses-
sions of the link. Among the various paths through which
route-request arrives at destination from source, any one with
the highest path weight is created as optional. EXERP incul-
cates stability and agility in communication; greatly reduc-
ing the energy consumption, message cost and end-to-end
delay, thereby increasing the packet delivery ratio, compared
to other well-known energy-efficient routing competitors.

References

1. Banerjee, A.; Dutta, P.: Survey of unicast routing protocols for

Fig. 4 End-to-end delay versus no. of nodes (in seconds) mobile ad hoc networks. Int. J. Adv. Sci. Technol. 1 (2010)
2. Meghanathan, N.: Survey and taxonomy of unicast routing proto-
cols for mobile ad hoc networks. Int. J. Appl. Graph Theory Wirel.
ad hoc Sens. Netw. 1 (2009)
packet delivery ratio for all the mentioned protocols till the
3. Toh, C.K.: Maximum battery life routing to support ubiquitous
network is overcrowded with nodes. When the network is mobile computing in wireless ad hoc networks. IEEE Commun.
overcrowded, message contention and collision reduce the Mag. (2001)

123
 Author's personal copy
Arab J Sci Eng (2014) 39:891–901 901

4. Kaabneh Halasa, K.; Al-Bahadili, A.: An effective location-based 11. University of Maryland. www.cs.umd.edu/projects/netcalliper/
power conservation scheme for mobile ad hoc networks. Am. J. software.html. MaRS (Maryland Routing Simulator)
Appl. Sci. B, 1708–1713 (2008) 12. Mashhadi, A.J.; Mokhtar, S.B.; Capra, L.: Habit: leveraging human
5. Manasa, B.; Vijayalakshmi, P.: Hint based power efficient routing mobility and social network for efficient content dissemination in
protocol for manets. Int. J. Comput. Appl. 10 (2010) manets. In: Proceedings of the 10th IEEE International Sympo-
6. Suri, P.; Soni, M.K.; Tomar, P.: Qos enabled power aware routing sium on a World of Wireless, Mobile and Multimedia Networks
protocol (qepar). Int. J. Eng. Sci. Technol. 2, 4880–4885 (2010) (WoWMoM09), vol. 3 (2009)
7. Dhurandher, S.; Misra, S.; Obaidat, M.S.; Bansal, V.; Singh, P.R.; 13. Rao, G.M.: Energy efficient reliable routing protocol for mobile
Punia, V.: Eeaodr: an energy ficient ad hoc on-demand routing ad hoc networks. In: Electronics Computer Technology (ICECT),
protocol for mobile ad hoc networks. Int. J. Commun. Syst. 22(7) 2011, 3rd International Conference, vol. 5, pp. 296–299 (2011)
(2009) 14. Kim, K.-J.; Yoo, S.-J.: Power- efficient reliable routing protocol
8. Selvi, V.V.; Parthasarathi, R.: Trace based mobility model for ad for mobile ad-hoc networks. In: IEICE Transaction on Communi-
hoc networks. In: 3rd IEEE International Conference on Wireless cations, vol. E-88B, pp. 4588–4597 (2005)
and Mobile Computing, Networking and Communications. IEEE 15. Rajeswari, S.; Venkataramani, Y.: An adaptive energy efficient and
(2007) reliable gossip routing protocol for mobile adhoc networks. In:
9. Lotfi, M.; Jabbehdari, S.; Shahmirzadi, M.A.: A new energy effi- IJCTE, vol. 2 (2010)
cient routing algorithm based on a new cost function in wireless ad
hoc networks. J. Comput. 2 (2010)
10. Bandai, M.; Nakayama, S.; Watanabe, T.: Signal strength based
energy efficient routing for ad hoc networks. IEICE Trans. Com-
mun. 91.B, 1006–1014 (2008)

123