You are on page 1of 8

Double-Ruling-Based Location-Free

Data Replication and Retrieval Scheme


in Mobile Ad Hoc Networks
Yao-Jen Tang, Jian-Jhih Kuo, and Ming-Jer Tsai
Department of Computer Science
National Tsing Hua University, Taiwan 30013, R.O.C.
Email: mjtsai@cs.nthu.edu.tw

AbstractUsing the double-ruling technique, many data replication and retrieval schemes achieve low data retrieval latency.
However, none of these schemes are location-free schemes in
mobile ad hoc networks (MANETs). In this paper, we propose a
zero-knowledge double-ruling-based location-free data replication
and retrieval scheme (MobiMark) in MANETs. Our primary idea
is to label the grid-like structured landmarks in the network using
the landmark-labeling, dynamically designate the node that is
nearest to a landmark as the landmark broker, and transmit
the consumers interests (or producers data) to all horizontal
(or vertical) landmark brokers using the double-ruling technique.
Simulations show that MobiMark achieves good performance in
terms of data retrieval rate and data retrieval latency.

I. I NTRODUCTION
In a data replication and retrieval scheme, which is also
known as an information brokerage system or a data publish
and subscribe mechanism, a consumer submits its interests
to brokers and a producer distributes its data to brokers.
Subsequently, the broker matches the consumers interests and
the producers data and delivers the matched data to the
corresponding consumer. Note that the consumer does not
need to know the identity or location of the producer. A wide
range of applications exist for data replication and retrieval
schemes, such as system monitoring and management [1],
news distribution [2], location service [3], and information
brokerage [4][6]. In this paper, we develop a data replication
and retrieval scheme in mobile ad hoc networks (MANETs).
In MANETs, several location-free data replication and retrieval schemes are studied.1 The schemes can be further
divided in two categories: knowledge-based schemes [10][12]
and zero-knowledge based schemes [13][16]. Knowledgebased schemes guide the message forwarding based on the
collected historical contact information, such as the contact
probability and social relationship, and can be employed only
under the circumstance that the contact probability among
nodes changes very slowly with time (otherwise, the contact
situation in the future cannot be predicted accurately based
1 Although a node could know the geographic location information by being
equipped with a Global Positioning System (GPS) receiver nowadays, there are
still many limitations on GPS [7]. For example, there are several applications
in indoor environments, where the GPS signals are unavailable [8]. In addition,
in the battleeld, the GPS signals cannot be normal amplied and detected due
to the military jamming signals [9].

on the historical contact information collected). Different from


knowledge based schemes, zero-knowledge based schemes do
not use the historical contact information for message forwarding. In [13], all brokers are organized by a multicast tree.
A consumer submits its interests, and a producer distributes
its data to the nearest broker. Subsequently, the consumers
interests are sent to a set of brokers and the matched data
are forwarded to the consumers broker through other brokers
in the multicast tree. In MANETs, this scheme demands a
considerable amount of message overhead for maintaining the
multicast. In [14], consumers interests and producers data are
circulated amongst a large set of brokers; thus, a consumer can
retrieve the data of interest from a producer if the consumers
interests and the producers data are delivered to the same
broker. This scheme incurs considerable message overhead to
replicate/retrieve the data. In 3DLS [15], all nodes are divided
into hills. The consumers interests and producers data are
both delivered to the hilltops of hills that are traversed one
by one using the depth-rst search strategy. The scheme stores
the message trail and visited hilltops in the packet header of
the replication/retrieval message. In DRIP [16], some nodes
are selected as brokers. Each broker periodically broadcasts
its information message into the network, and each node
periodically joins the nearest brokers region. A consumer
submits its interests and a producer distributes its data to
their nearest brokers, and these brokers share the obtained
consumers interests or producers data to a xed number
of neighboring brokers. When the consumers and producers
brokers are located in the same or neighboring broker regions,
the consumers broker obtains and delivers the matched data to
the consumer. To achieve moderate performance, DRIP needs
to adopt an appropriate number of brokers and data copies.2
Note that in DRIP, either the consumers interests or the
producers data may be stored by the brokers in a local area,
2 In DRIP, the lesser the number of brokers, the more the data stored in a
broker and the higher the broker coverage. Higher broker coverage indicates
that there is more transmission for a broker, which results in worse throughput
due to the bandwidth bottleneck. On the other hand, the greater the number
of brokers, the smaller the probability that the consumers interest is matched
to the producers data. In addition, to increase the data retrieval rate, it is
necessary to circulate a larger number of copies among brokers; however, this
incurs greater message overhead of data replication and retrieval.

978-1-4799-3572-7/14/$31.00 2014 IEEE

II. T HE N ETWORK M ODEL

(a)

(b)

Fig. 1. Snapshots of (a) DRIP and (b) MobiMark with 25 brokers, 5 copies
each of the producers data and consumers interests. Nodes in the same broker
region are shown in a color; nodes circumscribed by black squares denote
brokers; nodes with crosses (`) and diagonal crosses () denote the brokers
with consumers interests and producers data, respectively; and, nodes with
eight-point compass rose () denote the brokers with both consumers interests
and producers data.

as illustrated in Fig. 1a, resulting in a decrease in data retrieval


rate and an increase in data retrieval latency, especially in
the networks where most nodes only move locally. Therefore,
this provides the motivation for this paper. Similar to DRIP,
we use an appropriate number of brokers to divide the nodes
into several regions and store the consumers interests and
producers data in a subset of the brokers. However, different
from DRIP, we construct the grid-like structured brokers and
transmit the consumers interests (or producers data) to all horizontal (or vertical) brokers using the double-ruling technique,
as illustrated in Fig. 1b. As described before, the double-ruling
technique can achieve a high data retrieval rate and low data
retrieval latency in static ad hoc networks. Our idea is to port
the double-ruling technique in MANETs where nodes do not
possess the geographic location information. The challenges
are to construct structured brokers and orient the nodes for
data replication and retrieval. The contribution of this paper is
summarized as follows:

A distributed landmark-labeling algorithm is proposed for


constructing a grid-like structured landmark brokers in
MANETs, where each node does not possess geographic
information.
A zero-knowledge double-ruling-based location-free data
replication and retrieval scheme, termed MobiMark, is
proposed in MANETs. MobiMark is the rst zeroknowledge double-ruling-based location-free data replication and retrieval scheme in the literature.
Simulations show that MobiMark achieves good performance in terms of data replication and retrieval message
overhead, data retrieval rate, and data retrieval latency.

The remainder of this paper is organized as follows. The


network model and problem are described in Section II. The
basic ideas are illustrated in Section III. Structured landmark
brokers are constructed in Section IV. MobiMark is proposed
in Section V. MobiMarks performance is studied using simulations in Section VI. Related works are discussed in Section
VII. Finally, the conclusion is presented in Section VIII.

AND

P ROBLEM

The network model is described as follows:


1) There is no infrastructure.
2) Each node has a unique identier.
3) Each node does not possess the geographic location
information.
4) The future contact probabilities or social relationship
among nodes are unpredictable.
5) Each communication is bidirectional.
Note that the unique node identier assumption can be
justied by permitting each node to select a random number
in a large enough interval or by employing the universal
unique identier. Furthermore, the bidirectional communication
assumption can be eliminated by employing the technique for
the heterogeneous communication ranges in [17].
In MobiMark, nodes play three roles: consumer, provider,
and broker. A consumer can submit its interests. A producer can
distribute its data. A broker can collect the consumers interests
and the producers data, match the interest and the data, and
delivers the matched data to the corresponding consumer. The
following assumptions are made in MobiMark:
1) There are one leader node and several member nodes; the
leader node determines the networks moving trend and
the member node randomly moves under the networks
moving trend.
2) Each node broadcasts a Hello message to its neighbors
periodically.
3) Each node is aware of the architecture level l and the
broadcast expiration time.
4) The messages used to construct and maintain the
MobiMark system architecture, such as Leader Hop and
p0, 0q Hop messages, are aggregated into the Hello messages after the broadcast expiration time.
5) The hop counter contained in a message is initially set to
1 and advanced in increments by the forwarding nodes.
6) For any competition with the same value, the node ID is
used to break the tie.
The architecture level l and the broadcast expiration time
are decided by the leader node and broadcast to all nodes
in the network at the beginning of the construction of the
MobiMark system architecture. The architecture level l is used
to determine the number of landmarks in the MobiMark system
architecture, as described in Section IV. Once the leader node
decides the architecture level l, it does not change. The broadcast expiration time is used to limit the broadcast overhead
generated by the construction of the MobiMark system architecture.3 In addition, the construction and maintenance of the
MobiMark system architecture only requires the neighboring
nodes to exchange a little information periodically, as described
in Section IV; therefore, we aggregate the information used to
construct and maintain the MobiMark system architecture into
the periodic Hello messages. Furthermore, because the node
3 A large value of the broadcast expiration time can reduce the construction
time of the MobiMark system architecture.

does not possess the geographic location information, the hop


count is used as the distance metric in MobiMark. Our goal
is to propose a data replication and retrieval scheme such that
each node can retrieve the data collected by other nodes in a
timely manner.
III. T HE BASIC I DEAS
In a MANET, where the node does not possess the geographic location information, the designing of a double-rulingbased data replication and retrieval scheme poses the following
challenges: (1) How can brokers be identied in MANETs
in a distributed manner? (2) How can structured brokers be
established in MANETs in a distributed manner so that the
double-ruling technique can be employed in the data replication
and retrieval scheme? Ideas for overcoming challenges 1 and
2 are demonstrated in Sections III-A and III-B, respectively.
A. Identication of Brokers
To identify brokers, we use landmarks. A landmark is an
object with special features in a town. For example, Grand
Central Terminal is a landmark in New York City. Clearly,
it is convenient for us to rendezvous near Grand Central
Terminal. To simply elaborate the idea, we rst assume that
there are some landmarks already existing in the network.
For each landmark, there is a device periodically broadcasting
the beacon message to the network. We will remove this
assumption later in Section III-B. In addition, since there is
no geographical location information, the distance between a
node and a landmark is evaluated by the hop counts.
In a data replication and retrieval scheme, brokers provide the data retrieval and replication services. However, in
a MANET, brokers may move anywhere, so it is hard for
consumers to nd the brokers containing the data of interest
and retrieve it. Our idea is to label landmarks in a MANET,
and designate the node nearest a landmark the landmark broker.
Note that the landmark broker is played by the node nearest
the landmark at any time. Since the landmark does not move,
the landmark broker is quasi-steady.
B. Establishment of Structured Brokers
As discussed in Section III-A, brokers are identied by landmarks. However, real landmarks may not exist in a MANET.
Thus, to establish structured brokers, we need to label structured (virtual) landmarks inside the network. The idea is the
recursive centroid-nding procedure of a square. Consider a
1 1 square with four corners located at p0, 0q, p1, 1q, p1, 0q,
and p0, 1q. Observe that the centroid of the square located
at p 12 , 12 q has the minimum maximum distance from four
corners among all points within the square. We denote the
coordinates px, yq the px, yq-landmark. Suppose that the p0, 0q-,
p1, 1q-, p1, 0q-, and p0, 1q-landmark brokers are located near the
coordinates p0, 0q, p1, 1q, p1, 0q, and p0, 1q, respectively. Then,
the node with the minimum maximum hop distance from the
p0, 0q-, p1, 1q-, p1, 0q-, and p0, 1q-landmark brokers is located
near the coordinates p 12 , 21 q with a high probability, which is
suitable to be the p 12 , 12 q-landmark broker. Similarly, the nodes

(a)

(b)

(c)

Fig. 2. The landmarks for (a) l 0, (b) l 1, and (c) l 2, where each
landmark shows a name in the parentheses.

with the minimum maximum hop distances from the p0, 0qand p1, 0q-landmark brokers, the p1, 0q- and p1, 1q-landmark
brokers, the p1, 1q- and p0, 1q-landmark brokers, and the p0, 1qand p0, 0q-landmark brokers can be designated the p 12 , 0q-,
p1, 21 q-, p 12 , 1q-, and p0, 12 q-landmark brokers, respectively. In
n
summary, the p m
2 , 2 q-landmark brokers can be designated after
the p0, 0q-, p1, 1q-, p1, 0q-, and p0, 1q-landmark brokers are
identied, where 0 m, n 2, and m or n is equal to
n
1. Similarly, the p m
4 , 4 q-landmark brokers can be designated
m1 n1
after the p 2 , 2 q-landmark brokers are identied, where 0
m1 , n1 2, 0 m, n 4, and m or n is prime to 4. In
conclusion, given l P Z` Y t0u (the architecture level), for all
n
1 i l, the p m
2i , 2i q-landmark brokers can be designated
m1
n1
after the p 2i1 , 2i1 q-landmark brokers are identied, where
0 m1 , n1 2i1 , 0 m, n 2i , and m or n is prime
to 2i . See Fig. 2 for the illustration. Therefore, if we can
properly elect four nodes to become the p0, 0q-, p1, 1q-, p1, 0q-,
and p0, 1q-landmark brokers, then the structured landmarks can
be established by those four landmark brokers. Note that the
landmark brokers are not dened by geographical locations in
a rectangle place. The coordinates here are the names that are
used to dene the landmark brokers and their relationships. The
process of nding the rst four landmark brokers is described
in Section IV-A.
IV. M OBI M ARK S YSTEM A RCHITECTURE
The MobiMark system architecture consisting of structured
landmark brokers is constructed and maintained in a threen
phase process. Given the architecture level l, the p m
2l , 2l ql
landmark brokers, where 0 m, n 2 , are elected and
maintained in a distributed manner in rst two phases. See Fig.
2 for examples of l 0, 1, 2. In the rst phase, the brokers
of p0, 0q-, p1, 0q-, p0, 1q-, and p1, 1q-landmarks, called polar
landmarks, are elected and maintained. Each node evaluates
the hop distances from the brokers of all polar landmarks, as
described in Section IV-A. In the second phase, if necessary,
the brokers of other landmarks, called ordinary landmarks, are
elected and maintained. Each node evaluates the hop distances
from all ordinary landmark brokers, as described in Section
IV-B. After the polar/ordinary landmark brokers are elected,
the landmark regions are established. The data integrity in
landmark brokers is ensured in the third phase, as described
in Section IV-C.

Algorithm 1 Evaluation of Hop Distance from Leader Node

Algorithm 2 Generatation of p0, 0q Hop Message

Require: A node U and Leader Hop messages


1. M the rst Leader Hop message;
2. h hop counter in M ;
3. for each Leader Hop message M 1 do
4.
h1 hop counter in M 1 ;
5.
if h h1 then
6.
h h1 , M M 1 ;
7.
end if
8. end for
9. set the hop distance from the leader node to h;
10. forward M to the neighbors;

Require: A node U and the hop distance Dv from the leader node
of each node v P tU, U s 2-hop neighborsu
11. if U = arg minvPtU,U s 2-hop neighborsu tDv u then
12.
generate a p0, 0q Hop message M containing U s ID, a hop
distance from the leader node set to DU , and a hop counter
set to 0;
13. end if

A. Election of Polar Landmark Brokers


The p0, 0q-landmark (or p1, 1q-landmark) broker is the node
with the maximum hop distance from the leader node (or
the p0, 0q-landmark broker). The p1, 0q-landmark (or p0, 1qlandmark) broker is the node with the maximum hop distance
from the leader node (or the p1, 0q-landmark broker) among
the nodes whose hop distances from the p0, 0q- and p1, 1qlandmark brokers differ not greater than 1. The p0, 0q-, p1, 1q-,
p1, 0q-, and p0, 1q-landmark brokers are elected and maintained
in order in the following steps.
First, each node evaluate the hop distance from the leader
node with the Leader Hop message generated and broadcast by
the leader node, as described in Algorithm 1. The Leader Hop
message contains a hop counter. After a node receives a
Leader Hop message, the node sets the hop distance from the
leader node to the hop counter contained in the message, and
forwards the message to the neighbors (lines 12, 910). If a
node receives more than one Leader Hop message, the node
sets the hop distance from the leader node to the smallest hop
counter and forwards the message containing the smallest hop
counter to the neighbors (lines 38).
Second, the p0, 0q-landmark broker is elected. Algorithm 2
describes how to become the candidates of p0, 0q-landmark
broker and generate a p0, 0q Hop message. The nodes with
the maximum hop distance from the leader node in the 2hop neighborhood become the candidates of p0, 0q-landmark
broker (line 11). Each node evaluates the hop distance from
the p0, 0q-landmark broker by p0, 0q Hop messages generated
and broadcast by the candidates of p0, 0q-landmark broker, as
implemented below. The p0, 0q Hop message contains the ID
and the hop distance from the leader node, and a hop counter.
The node sets the hop distance from the p0, 0q-landmark
broker to the hop counter contained in the p0, 0q Hop message
received that was generated by the node with the maximum
hop distance from the leader node. Then the node forwards the
p0, 0q Hop message to the neighbors similar to Algorithm 1.
Third, the p0, 0q-landmark broker is maintained by designating a node, u, the new p0, 0q-landmark broker if u has the
maximum hop distance from the leader node among the p0, 0qlandmark broker and its neighbors. The polar landmark brokers
are elected and maintained the same way as the p0, 0q-landmark
broker. Each node evaluates the hop distance from the polar
landmark brokers in the same way as for the p0, 0q-landmark

broker.
B. Election of Ordinary Landmark Brokers
n
i
The p m
2i , 2i q-landmark broker, where m or n is prime to 2 ,
is the node with the minimum maximum hop distance from the
n1
m`1 n`1
m`1 n1
m1 n`1
p m1
2i , 2i q-, p 2i , 2i q-, p 2i , 2i q-, and p 2i , 2i qlandmark brokers, if both m and n are odd; otherwise, if
n
m is odd, the p m
2i , 2i q-landmark broker is the node with
n
the minimum maximum hop distance from the p m1
2i , 2i qm`1 n
and p 2i , 2i q-landmark brokers; otherwise, if n is odd, the
p 2mi , 2ni q-landmark broker is the node with the minimum maxin1
m n`1
mum hop distance from the p m
2i , 2i q- and p 2i , 2i q-landmark
m n
brokers. The p 2i , 2i q-landmark brokers, where m or n is prime
to 2i , for i 1 to l are elected in order. Each ordinary landmark
broker is elected and maintained the same way as the p0, 0qlandmark broker. Each node evaluates the hop distance from
each ordinary landmark broker the same way the hop distance
from the p0, 0q-landmark broker is evaluated.

C. Establishment of Landmark Regions and Preservation of


Data Integrity
Each landmark region consists of nodes. A node joins the
px, yq-landmark region if the px, yq-landmark broker has the
smallest hop distance from the node among all landmark
brokers. In addition, to preserve data integrity, once the px, yqlandmark broker is newly elected, the ex-broker generates
a Data Integrity message containing all consumers interests
and producers data the ex-broker collects. The Data Integrity
message is forwarded from one node to another node with a
smaller hop distance from the new px, yq-landmark broker by
1 until the new px, yq-landmark broker is reached.
V. M OBI M ARK
MobiMark is based on the system architecture constructed
in Section IV. Each consumer in the px, yq-landmark region
submits its interests to all pu, vq-landmark brokers with v y.
Each producer in the px, yq-landmark region distributes its
data to all pu, vq-landmark brokers with u x. Once a
landmark broker receives the producers data that matches the
collected consumers interests, the landmark broker delivers the
matched data to the corresponding consumers. The submission
of consumers interests and the distribution of producers data
are given in Sections V-A and V-B, respectively. The delivery
of the matched data is described in Section V-C. In addition,
we deal with the handoff of consumers interests, which is
necessary due to the mobility of consumers, in Section V-D.
For the convenience of description, the px, yq-landmark and
the pu, vq-landmark denote the same landmark if x u and

Algorithm 3 Submision of Consumers Interests


Require: A node U , an architecture level l, and a message M
14. if M is My Broker Appr message then
15.
px, yq the name of consumers landmark in M ;
16.
if U is the px, yq-landmark broker then
17.
store consumers subscription information;
18.
generate an East Appr message and a West Appr message,
where the East Appr (or West Appr) message contains
consumers subscription information, and the name of the
target landmark equal to px ` 21l , yq (or px 21l , yq);
19.
forward the East Appr (or West Appr) message to the
neighbor with a smaller hop distance from the px ` 21l , yqlandmark broker (or px 21l , yq-landmark broker) by 1;
20.
else
21.
forward M to the neighbor with a smaller hop distance from
the px, yq-landmark broker by 1;
22.
end if
23. else if M is East Appr (or West App) message then
24.
pu, vq the name of the target consumers landmark in M ;
25.
if U is the pu, vq-landmark broker then
26.
store consumers subscription information;
27.
if M is an East Appr (or West App) message and u 1
(or u 0) then
28.
update the target landmarks name in M to pu ` 21l , vq
(or pu 21l , vq);
29.
forward M to the neighbor with a smaller hop distance
from the pu ` 21l , vq-landmark (or pu 21l , vq-landmark)
broker by 1;
30.
end if
31.
else
32.
forward M to the neighbor with a smaller hop distance from
the pu, vq-landmark broker by 1;
33.
end if
34. end if

y v. For example, the p 24 , 44 q-landmark denotes the p 21 , 22 qlandmark. Note that the messages containing the consumers
interests are delivered along the eastern and western directions,
and the messages containing the producers data are delivered
along the southern and northern directions.
A. Submission of Consumers Interests
If a consumer in the px, yq-landmark region is not the px, yqlandmark broker, the consumer generates a My Broker Appr
message containing the consumers subscription information
including the consumers interests, ID, and landmark name
equal to px, yq. This message is forwarded from one node
to another node with a smaller hop distance from the px, yqlandmark broker by 1 until the px, yq-landmark broker is
reached. Algorithm 3 describes how a consumer in the px, yqlandmark region submits its interests to all pu, vq-landmark
brokers with v y in a distributed manner. If a consumer, c,
is this landmark broker, or this landmark broker receives this
message generated by a consumer, c, in the px, yq-landmark
region, then an East Appr message and a West Appr message
are generated. The East Appr message, which contains the
consumers subscription information and the name of the target
landmark initially set to px ` 21l , yq, traverses the px ` 21l , yq-,
px` 22l , yq-, , p1, yq-landmark brokers one by one (lines 18
19, 2829, 32), where l denotes the architecture level dened
in Section II. The West Appr message, which contains the

consumers subscription information and the name of the target


landmark initially set to px 21l , yq, traverses the px 21l , yq-,
px 22l , yq-, , p0, yq-landmark brokers one by one similarly
to that for the East Appr message.
B. Distribution of Producers Data
The data produced by a producer in the px, yq-landmark
region is rst delivered to the px, yq-landmark broker similarly
to that for submitting the interests to the landmark broker of
the consumer, if the producer is not the px, yq-landmark broker.
Subsequently, this landmark broker generates a North Appr
message and a South Appr message; each contains the producers data. The North Appr message traverses px, y ` 21l q-,
px, y ` 22l q-, , px, 1q-landmark brokers one by one, and the
South Appr message traverses px, y 21l q-, px, y 22l q-, ,
px, 0q-landmark brokers one by one each similarly to that for
the East Appr message.
C. Delivery of Matched Data
When the px, yq-landmark broker matches the producers data and the interests of a consumer, c, in the
pw, yq-landmark region, the landmark broker generates a
Matched Data Delivery message containing the matched data,
the ID of the consumer c, and the name of the landmark
of the consumer c equal to pw, yq. This message traverses
the landmark brokers one by one until reaching the pw, yqlandmark broker similarly as the East Appr message. As
the pw, yq-landmark broker obtains the matched data, it delivers the matched data to the consumer c by generating
a Matched Data Broadcast message containing the matched
data, the ID of the consumer c, and the name of the landmark
of the consumer c equal to pw, yq. The landmark broker then
broadcasts the message to the nodes in the pw, yq-landmark
region.
D. Handoff of Consumers Interests
When a consumer, c, moves from the px, yq-landmark region
to the pp, qq-landmark region, the consumer c resubmits its
interests to all pu, qq-landmark brokers with 0 u 1 by the
methods described in Section V-A and cancels cs subscriptions
to all pu, yq-landmark brokers with 0 u 1 similarly to
submitting cs interests.
VI. P ERFORMANCE E VALUATION
In the simulations, we rst employed the Geometrically Distributed Graph Model [18] to generate the initial networks with
a density of 10, where the network density denoted the average
number of neighbors per node and the transmission range of
each node was set to 1 unit. Subsequently, the reference point
group mobility (RPGM) [19] was used as the node mobility
model, where all the nodes were grouped into a single group
and the speed deviation ratio and angle deviation ratio were
set to 0.1.4 The speeds of the nodes were set to 0.004, 0.02,
4 The group mobility model are more realistic in many mobile network
applications such as a land mine search in a battleeld, a large scale disaster
recovery, and a group of attendees roaming in an exhibition [20].

125
100
75
50
25
0
0.004

0.02

0.04
0.06
Node Speed (units/sec)

(a)
Fig. 3.

0.08

100

90

90

80

80

70
60
50
40
30

MobiMark (l=0)
MobiMark (l=1)
MobiMark (l=2)

20
10
0
0.004

0.02

0.04
0.06
Node Speed (units/sec)

100

70
60
50
40
30

MobiMark (l=0)
MobiMark (l=1)
MobiMark (l=2)

20
10

0.08

(b)

Data Retrieval Latency (sec)

Construction Time (sec)

150

100

Data Retrieval Rate (%)

MobiMark (l=0)
MobiMark (l=1)
MobiMark (l=2)

175

Data Replication/Retrieval Message Overhead

200

0
0.004

0.02

0.04
0.06
Node Speed (units/sec)

(c)

0.08

MobiMark (l=0)
MobiMark (l=1)
MobiMark (l=2)
10

0.1
0.004

0.02

0.04
0.06
Node Speed (units/sec)

0.08

(d)

Impact of the node speed on construction time, data replication and retrieval message overhead, data retrieval rate, and data retrieval latency.

0.04, 0.06, and 0.08 units/sec, which are equal to 1, 5, 10, 15,
and 20 m/sec, respectively, for a typical nodes transmission
range (250 m). To study the impact of the nodes locality, the
maximum permitted distance deviations of a node centered at
the reference point of the node in RPGM, called node moving
radii, were set to 5, 10, 15, 20, and 25 units, respectively.
(In our simulations, the maximum distance between any two
nodes in the network with 500 nodes is approximately equal
to 27 units.) The simulation execution time for each network
was 1,500 seconds, where the rst 1,000 seconds were used to
avoid the initialization problem of the mobility model [21]. For
each network, data were generated by a Poisson process, with
the intensity of 0.8 during 1,000 to 1,100 seconds and expiry at
1,500 seconds. The node that produces the generated data (the
producer) was randomly chosen from the nodes in the network.
The probability of each node interested in each generated data
was 5%. Each consumer submitted its interests right after the
interested data was generated. Time to lives (TTLs) of each
packet were set to 20, 40, 60, 80, and 100, respectively. Each
node could transmit 20 messages per second. Nodes exchanged
Hello messages each second. The empirical data were obtained
by averaging the data of 10 networks.
We evaluate the performance of MobiMark architecture,
in terms of construction time, data replication and retrieval
message overhead, data retrieval rate, data retrieval latency, and
extra broadcast overhead in Section VI-A, and the performance
of MobiMark, SAW [11], and DRIP [16], in terms of data
retrieval rate and data retrieval latency, in Section VI-B. In
SAW, each producer has a xed number of copies per data. The
node that has more than one data copy delivers half of the data
copies to an encountered node and each consumer retrieves the
producers data from an encountered node that has the data. The
construction time is the time required to construct the MobiMark system architecture. It denotes the time elapsed between
when the leader node generated the Leader Hop message and
when each node received one message from each broker. The
data replication and retrieval message overhead denotes the
total number of transmissions and broadcasts of messages
containing the consumers interests or the producers data for
each producer-consumer pair. The data retrieval rate denotes the
ratio of the successful to total submissions for interests retrieval
made by consumers. The data retrieval latency denotes the
time elapsed between the submission for interests retrieval and
receipt of matched data by the consumer. The extra broadcast

overhead denotes the number of broadcast generated by the


leader node and the landmark brokers during the construction
time.
A. Empirical Results of MobiMark Architecture
Fig. 3 shows the simulation results for the construction
time, data replication and retrieval message overhead, data
retrieval rate, and data retrieval latency of Mobimark under
different number of landmark brokers and node speeds. In
this simulation, we established 4 landmark brokers (l 0), 9
landmark brokers (l 1), and 25 landmark brokers (l 2) of
MobiMark architecture with 300 nodes. The node speed was
set to 0.04 units/sec, the node moving radius was set to 15
units, the TTL was set to 60, and the broadcast expiration time
was set to 0.
The increasing architecture level yielded three observations.
First, more ordinary landmark brokers were elected in order, resulting in greater construction time of MobiMark architecture.
Second, the landmark regions became smaller. In a smaller
landmark region, the number of consumers changing the
landmark region increased, resulting in a lower data retrieval
rate and greater maximum data retrieval latency. Third, the
distance between the consumer or producer and its landmark
broker decreased; thus, both the data replication and retrieval
message overhead and the average retrieval latency decreased.
The increasing node speed yielded two results. First, the
construction messages spread to all nodes faster, resulting in
smaller construction time. Second, the number of consumers
changing the landmark region increased; thus, both the data
replication and retrieval message overhead and the data retrieval
latency increased and the data retrieval rate decreased.
Fig. 4 shows the simulation results for the construction time
and the extra broadcast overhead of MobiMark under different
number of landmark brokers and different values of broadcast
expiration time. In general, as anticipated, the greater the value
of the broadcast expiration time, the smaller the construction
time and the greater the extra broadcast overhead; the greater
the number of landmark brokers, the greater the construction
time and the extra broadcast overhead. In addition, if the
broadcast expiration time is set to at least 6 seconds, it takes
less than 6 seconds to construct the MobiMark architecture
with architecture level l 0, 1, 2. Moreover, if the broadcast
expiration time is set to less than 3 seconds, it requires the
same extra broadcast overhead to construct the MobiMark
architectures with different architecture levels l 0, 1, 2. This

100

120

Construction Time (sec)

90

Extra Broadcast Overhead

MobiMark (l=0)
MobiMark (l=1)
MobiMark (l=2)

60
30
15
12
9
6

MobiMark (l=0)
MobiMark (l=1)
MobiMark (l=2)

80
60
40
20

3
0

0
0

2
4
6
Broadcast Expiration Time (sec)

(a)

2
4
6
Broadcast Expiration Time (sec)

(b)

Fig. 4. Impact of the broadcast expiration time on construction time and extra
broadcast overhead.

is because the broadcast messages are only generated by the


polar landmark brokers and their candidates in the rst 3
seconds during the construction of the MobiMark architecture.
B. Empirical Results of MobiMark, SAW, and DRIP
Figs. 5 and 6 show the simulation results concerning the
data retrieval rate and data retrieval latency, respectively of
MobiMark, SAW, and DRIP. In this simulation, MobiMark
employed 25 landmark brokers. In such case, MobiMark replicated or retrieved the messages on 5 landmark brokers; thus,
we compared SAW with 5 copies and DRIP with 25 brokers
and 5 copies. In addition, the number of nodes was set to 300,
node speed was set to 0.04 units/sec, node moving radius was
set to 15 units, TTL was set to 60, and the broadcast expiration
time was set to 0.
The comparison of MobiMark, SAW, and DRIP yielded
the following two observations. First, MobiMark has the best
performance in terms of data retrieval rate and average data
retrieval latency. In SAW (or DRIP), to match the data, the
consumer (or the broker of the consumer) is required to contact
the node (or the broker) with (the copy of) the producers data.
In other words, the data are matched through message transmission and node movement in SAW and DRIP. In contrast,
the data are matched only through message transmission in
MobiMark. Because the node speed is considerably smaller
than the message transmission speed, the performance of
SAW and DRIP is worse than that of MobiMark in terms of
average/maximum data retrieval latency and worse than that of
MobiMark in terms of data retrieval rate as TTL is bounded.
It is remarkable that the data retrieval rate of MobiMark
approaches 100%; this is because MobiMark employs a doubleruling technique to replicate and retrieve data. Second, because
a consumer obtains the producers data only when the consumer
encounters one of the ve nodes with (the copy of) the
producers data and the node speed is considerably smaller than
the message transmission speed, the performance of SAW is
the worst in terms of data retrieval rate and average/maximum
data retrieval latency.
There are four observations. First, as the number of nodes
increases, the average hop count between two nodes in the
network increases, resulting in smaller data retrieval rates of
MobiMark, SAW, and DRIP; and greater average/maximum
data retrieval latency of MobiMark and SAW. It is noteworthy
that as the number of nodes increases, the average/maximum

data retrieval latency of DRIP decreases. It is because the consumer changes the broker region less frequently as the number
of nodes increases. Second, as the node speed increases, the
network topology changes more frequently. The impact of node
speed on Mobimark has been discussed in Section VI-A and is
omitted here. In SAW (or DRIP), as the node speed increase,
the consumer (or the broker of the consumer) has a higher
probability of contacting the node (or the broker) with (the copy
of) the producers data, resulting in greater data retrieval rate
and average/maximum data retrieval latency. Third, because
the message transmission speed is considerably faster than the
node moving speed, the node moving radius has a negligible
impact on the data retrieval rate and average/maximum data
retrieval latency of MobiMark, SAW, or DRIP. Fourth, the
TTL has a negligible impact on the data retrieval rate and
average/maximum data retrieval latency of SAW and DRIP.
This is because the data are matched through node movement
and message transmission in SAW and DRIP, whereas the node
speed is considerably smaller than the message transmission
speed. Besides, since TTL 40 is sufcient for most messages
to retrieve data in MobiMark, the TTL has a negligible impact
on the data retrieval rate and average/maximum data retrieval
latency of MobiMark as TTL 40.
VII. R ELATED W ORKS
Recently, many data replication and retrieval schemes are
proposed in delay/disruption tolerant networks (DTNs). The
schemes in [22], [23] require xed infrastructure to help
forward data, and the scheme in [24] needs to have the
geographic location information of all producers and consumers in advance. None of these schemes can be used in our
network model. In addition, the schemes in [25], [26] require
historical contact information to predict and guide the message
forwarding. In our network model, the contact probabilities
among nodes vary with time; thus, these schemes are unsuitable
for our network model. Moreover, in [27], [28], the schemes
guide the message forwarding with the help of the properties
extracted from the specic network traces, such as similarity
and centrality, and thus, the schemes can perform well only in
the corresponding specic networks.
VIII. C ONCLUSIONS
DRIP [16] is a location-free data replication and retrieval
scheme with remarkable performance in the environment where
nodes have small inter-contact time or data can tolerate delays.
However, in several scenarios, such as military reconnaissance,
the nodes may move locally and the data should be retrieved
as soon as possible. In such an environment, DRIP is unsuitable because in DRIP, the consumers broker can retrieve
the producers data only when it is near one of the brokers
who has the producers data. In this paper, we designed a
distributed method to construct structured landmark brokers
using the landmark-labeling algorithm, and propose a zeroknowledge location-free data replication and retrieval scheme
based on the double-ruling technique, termed MobiMark, for
MANETs. By simply aggregating the information used with the

Data Retrieval Rate (%)

60
50
40
30
20
10

70
60
50
40
30
20
10

200

300
Number of Nodes

400

(a)

Data Retrieval Latency (sec)

Data Retrieval Latency (sec)

10
1
0.1

0.04
0.06
Node Speed (units/sec)

0.08

30
20

300
Number of Nodes

60
50
40
30
20
10
0

10
15
20
Node Moving Radius (units)

10000

MobiMark (l=2)
SAW
DRIP
3DLS

1000
100
10
1

400

500

25

20

40

60
Time to Live

80

100

(d)

10000

MobiMark (l=2)
SAW
DRIP
3DLS

1000
100
10
1
0.1

0.004

(a)

0.02
0.04
0.06
Node Speed (units/sec)

0.08

(b)

Fig. 6.

70

(c)

0.1
200

40

0
0.02

10000

100

100

50

MobiMark (l=2)
SAW
DRIP
3DLS

80

Impact of the number of nodes, node speed, node moving radius, and TTL on data retrieval rate.

MobiMark (l=2)
SAW
DRIP
3DLS

1000

60

(b)

Fig. 5.
10000

70

10

0
0.004

500

90

MobiMark (l=2)
SAW
DRIP
3DLS

80

Data Retrieval Latency (sec)

0
100

100

90

MobiMark (l=2)
SAW
DRIP
3DLS

80

Data Retrieval Latency (sec)

Data Retrieval Rate (%)

70

100

90

MobiMark (l=2)
SAW
DRIP
3DLS

80

Data Retrieval Rate (%)

100

90

Data Retrieval Rate (%)

100

MobiMark (l=2)
SAW
DRIP
3DLS

1000
100
10
1
0.1

10
15
20
Node Moving Radius (units)

(c)

25

20

40

60
Time to Live

80

100

(d)

Impact of the number of nodes, node speed, node moving radius, and TTL on data retrieval latency.

periodic Hello messages, MobiMark can be easily implemented


in MANETs. In addition, simulations show that under different
architecture levels, the elected landmark brokers have a short
construction time, and MobiMark performs well in terms of
data replication and retrieval message overhead, data retrieval
rate, and data retrieval latency. We also compared MobiMark
with SAW and DRIP. The simulation results show that the
performance of MobiMark is the best in terms of data retrieval
rate and data retrieval latency.
ACKNOWLEDGEMENT
This work was partially supported by the National Science
Council, Taiwan, R.O.C. under Contract No. NSC 100-2628E-007-033-MY3.
R EFERENCES
[1] A. Carzaniga, M. J. Rutherford, and A. L. Wolf, A routing scheme for
content-based networking, in IEEE INFOCOM, 2004.
[2] F. Fabret, H. A. Jacobsen, F. Llirbat, J. Pereira, K. A. Ross, and
D. Shasha, Filtering algorithms and implementation for very fast publish/subscribe systems, in ACM SIGMOD, 2001.
[3] J. Li, J. Jannotti, D. S. J. D. Couto, D. R. Karger, and R. Morris,
A scalable location service for geographic ad hoc routing, in ACM
MOBICOM, 2000.
[4] R. Sarkar, X. Zhu, and J. Gao, Double rulings for information brokerage
in sensor networks, IEEE Transactions on Networking, vol. 17, pp.
19021915, 2009.
[5] Q. Fang, J. Gao, and L. J. Guibas, Landmark-based information storage
and retrieval in sensor networks, in IEEE INFOCOM, 2006.
[6] C.-H. Lin, J.-J. Kuo, and M.-J. Tsai, Reliable GPS-free double-rulingbased information brokerage in wireless sensor networks, in IEEE
INFOCOM, 2010.
[7] N. Patwari, J. N. Ash, S. Kyperountas, A. O. H. III, R. L. Moses, and
N. S. Correal, Locating the nodes: Cooperative localization in wireless
sensor networks, IEEE Signal Processing Magazine, vol. 22, pp. 5469,
2005.
[8] K. Wu, J. Xiao, Y. Yi, M. Gao, and L. M. Ni, FILA: Fine-grained indoor
localization, in IEEE INFOCOM, 2012.
[9] D. Last, Gnss: The present imperfect, Inside GNSS Magazine, vol. 5,
pp. 6064, 2010.
[10] J. Burgess, B. Gallagher, D. Jensen, and B. N. Levine, MaxProp: Routing for vehicle-based disruption-tolerant networks, in IEEE INFOCOM,
2006.

[11] T. Spyropoulos, K. Psounis, and C. S. Raghavendra, Spray and Wait: An


efcient routing scheme for intermittently connected mobile networks, in
ACM WDTN, 2005.
[12] Z. Li, H. Shen, H. Wang, G. Liu, and J. Li, SocialTube: P2P-assisted
video sharing in online social networks, in IEEE INFOCOM, 2012.
[13] Y. Huang and H. Garcia-Molina, Publish/Subscribe in a mobile environment, Wireless Networks, vol. 10, pp. 643652, 2004.
[14] P. Costa and G. P. Picco, Semi-probabilistic content-based publishsubscribe, in IEEE ICDCS, 2005.
[15] R. Friedman and N. Mori, 3DLS: Density-driven data location service
for mobile ad-hoc networks, in ACM MOBIHOC, 2009.
[16] Q. Yuan and J. Wu, DRIP: A dynamic Voronoi regions-based publish/subscribe protocol in mobile networks, in IEEE INFOCOM, 2008.
[17] V. Ramasubramanian, R. Chandra, and D. Mosse, Providing a bidirectional abstraction for unidirectional ad hoc networks, in IEEE INFOCOM, 2002.
[18] S. K. S. Gupta and P. K. Srimani, Adaptive core selection and migration
method for multicast routing in mobile ad hoc networks, IEEE Transactions on Parallel and Distributed Systems, vol. 14, pp. 2738, 2003.
[19] F. Bai, N. Sadagopan, and A. Helmy, Important: a framework to
systematically analyze the impact of mobility on performance of routing
protocols for adhoc networks, in IEEE INFOCOM, 2003.
[20] X. Hong, M. Gerla, G. Pei, and C.-C. Chiang, A group mobility model
for ad hoc wireless networks, in ACM MSWiM, 1999.
[21] T. Camp, J. Boleng, and V. Davies, A survey of mobility models for ad
hoc network research, Wireless Communications and Mobile Computing,
vol. 10, pp. 483502, 2002.
[22] D. J. Goodman, J. Borr`as, N. B. Mandayam, and R. D. Yates, INFOSTATIONS: a new system model for data and messaging services, in
IEEE VTC, 1997.
[23] T. Small and Z. J. Haas, The shared wireless infostation model - a new
ad hoc networking paradigm (or where there is a whale, there is a way),
in ACM MobiHoc, 2003.
[24] R. C. Shah, S. Roy, S. Jain, W. Brunette, and G. Borriello, Data
MULEs: modeling and analysis of a three-tier architecture for sparse
sensor networks, Ad Hoc Networks, vol. 1, pp. 215233, 2003.
[25] A. Balasubramanian, B. N. Levine, and A. Venkataramani, DTN routing
as a resource allocation problem, in ACM SIGCOMM, 2007.
[26] X. Tie, A. Venkataramani, and A. Balasubramanian, R3: Robust replication routing in wireless networks with diverse connectivity characteristics, in ACM SIGCOMM, 2011.
[27] P. Hui, J. Crowcroft, and E. Yoneki, BUBBLE Rap: Social-based
forwarding in delay-tolerant networks, in ACM MobiHoc, 2008.
[28] P. Costa, C. Mascolo, M. Musolesi, and G. P. Picco, Socially-aware
routing for publish-subscribe in delay-tolerant mobile ad hoc networks,
IEEE Journal on Selected Areas in Communications, vol. 26, pp. 748
760, 2008.

You might also like