A Social-based Strategy for Memory Management in Sensor Networks

Basim Mahmood

and Ronaldo Menezes

Computer and Information Center, University of Mosul, Mosul, Iraq

BioComplex Laboratory, Department of Computer Sciences, Florida Institute of Technology, Melbourne, U.S.A.

Keywords:

Wireless Sensor Networks (WSNs), Social Networks, Human Mobility Model, Social Capital, Memory

Management.

Abstract:

The technological structure of today’s societies enables people to easily exchange and share their information.

This structure contains many sophisticated technologies such as mobile wireless devices (e.g., smartphones

and tablets). These devices are mainly used for connecting people with each other. As these devices grow in

usability, many issues have become apparent such as memory management, security, and power consumption.

In this paper, we propose a novel social-based strategy for memory management in mobile sensor networks.

This strategy is inspired from two concepts, namely, social capital in sociology and preferential return mech-

anism in human mobility. The ﬁndings show that the proposed strategy is quite effective in keeping up-to-date

information in each sensor/device about the sensor connections. We believe that this is the ﬁrst work that

investigates the issue of memory management in this type of networks using concepts form social networks

and human mobility.

1 INTRODUCTION

Wireless devices (e.g., smartphones, tablets, laptops)

have become important tools to many aspects of our

lives. These devices support people in different tasks

related to education, business, and social interactions.

Recently, the use of wireless devices has signiﬁcantly

been increased, especially, with the widespread avail-

ability of the communication technologies (e.g., Wi-

Fi). Therefore, an infrastructure that is formed from

wireless devices and the connections among them al-

ready exists around us. Since these devices are car-

ried by people who tend to be mobile, the considered

infrastructure represents a Dynamic Wireless Sensor

Network (DWSN) in which “sensors” are the mobile

wireless devices and the connections among them are

formed where these devices become in the commu-

nication range of each other. However, many issues

have been introduced when designing applications on

this infrastructure and many challenges have become

apparent such as memory management, power con-

sumption, connectivity, and security.

In the aforementioned DWSN, the connections

among sensors depend on the social relations and

interactions among people. Therefore, understand-

ing the social networks helps the understanding of

how information ﬂows within the network which

eventually may contribute in overcoming connectiv-

ity in these types of networks. The ﬂow of in-

formation within a social network is performed by

passing through the relations among network nodes.

Moreover, the decision of passing information from a

node to another one is more likely based on relation

strength between the individuals. Yet, the strength

of a relation between two individuals depends on the

type of relation they maintain (Burt, 1982)(Borgatti

et al., 1998)(Coleman, 1988). In social networks,

the relations among individuals can be in two types:

strong or weak depending on the behavior and the

level of interactions (e.g., frequency of encounters be-

tween the individual) (Granovetter, 1973). Further-

more, Social Capital is another characteristic in so-

cial networks, by which the importance of a relation

between two individuals can be quantiﬁed.

1.1 Social Capital and Homophily

In sociology, the concept of Social Capital refers to

the beneﬁts (economic, political, cultural, etc.) de-

rived from the social relations and cooperation among

social actors (e.g., groups or individuals) (Putnam,

2000). The social capital of an individual, in its sim-

ple case, is the shared norms and values of a connec-

tion with a particular individual (Burt, 1997)(Burt,

2001)(Coleman, 1988). Moreover, (Lin, 1999) de-

ﬁned social capital as “resources embedded in an in-

Mahmood, B. and Menezes, R.

A Social-based Strategy for Memory Management in Sensor Networks.

DOI: 10.5220/0005672600250034

In Proceedings of the 5th International Confererence on Sensor Networks (SENSORNETS 2016), pages 25-34

ISBN: 978-989-758-169-4

dividual’s social network, resources that can be ac-

cessed or mobilized through ties in the networks”.

Furthermore, social capital is not possessed by people

themselves, but it is embedded in the social relation-

ships among them (Wisconson Center for Education

Research, 2006). (Ehrlich and Carboni, 2005) pre-

sented social capital as the total sum of the resources

that an individual gains as a result of the interactions

and relationships. (Licamele et al., 2005) also deﬁned

the social capital of an individual as the beneﬁts that

are given or received by this individual. Finally, (Jo-

hannessen, 2012) described social capital as the rela-

tions among individuals including the reciprocal trust-

worthiness which makes them more powerful in their

social networks.

The concept of social capital has been used in dif-

ferent studies such as information dissemination in

social networks (Bosen and Gang, 2010). It also de-

scribes many of social ties characteristics (e.g., co-

operation among individuals, and the importance of

a particular social tie), and among these characteris-

tics one can include the classiﬁcation of ties into weak

and strong as deﬁned by (Granovetter, 1973); clearly

a high social capital needs to include some diversity

between weak and strong tie.

Social Capital however is not a uniﬁed concept

and the literature has introduced sub-classiﬁcations of

the concept such as public, private, formal, informal,

bonding, bridging, and linking social capital. How-

ever, the main types of social capital in social network

literature are (Adhikari, 2008)(Larsen et al., 2004): 1)

Bonding social capital exist among homogeneous ac-

tors (e.g., friends or family members) and this type is

comprised primarily of strong ties. Hence, bonding

social capital may be appropriate for internal infor-

mation dissemination (e.g., among same group mem-

bers). 2) Bridging social capital exists among hetero-

geneous actors (e.g., across groups). This type incor-

porates mostly the weak ties of an individual and can

be useful in external information dissemination (e.g.,

among different groups). 3) Linking social capital is

also comprised of weak ties but only long-distance

connections making linking social capital also impor-

tant for external information dissemination.

In social sciences and social networks, Homophily

is the tendency of one to associate and connect

with similar others (McPherson et al., 2001)(Ibarra,

1992)(Marsden, 1988). According to the seminal

work of (McPherson et al., 2001), there are two

main aspects of homophily, namely; status homophily

which refers to the fact that individuals with similar

social status features (e.g., race, gender, religion, and

age) are more likely to form ties, and value homophily

which whereby individuals tend to participate with

those who behave and think in similar ways regard-

less of differences in social status. However, value

homophily has more signiﬁcant impact on homophily

than status homophily as presented in (Yuan and Gay,

2006). Moreover, there are many factors that lead

to homophily such as geography (e.g., people’s co-

location), social ties (e.g., strong and weak ties), or-

ganizational foci (e.g., school or work), isomorphic

sources (e.g., people who occupy equivalent roles),

and cognitive processes (e.g., people who have demo-

graphic similarity) (McPherson et al., 2001). More

importantly, (Lin, 1999) showed that homophily, as a

social characteristic, can affect social capital. (Bor-

gatti and Foster, 2003) and (Borgatti et al., 1998)

found that homophily, tie strength, and high rate of

knowledge transfer (i.e., information ﬂow) between

two individuals have a signiﬁcant impact on their so-

cial capital.

1.2 Human Mobility

In DWSNs, the term mobility refers to the ability of

nodes (sensors) to move in some way. One of the

ways that sensors move is following a mobility model

which describes the general spatio-temporal regulari-

ties of sensor node movements; more precisely, a mo-

bility model describes the movement of mobile nodes

and how their positions, directions, and speed change

over time (Lin et al., 2004)(Kesidis et al., 2003). Typ-

ically, to simulate and evaluate a DWSN, a particular

mobility model should be incorporated in the network

(Musolesi and Mascolo, 2009). In the context of so-

cial movement, we need to use models that have the

ability to describe most of human mobility character-

istics. (Song et al., 2010) proposed a model for human

mobility to that is able to precisely describe human

movement. Their model is based on two mechanisms:

Exploration: The tendency to explore new locations

decreases with time. This mechanism ensures that

the next step of an individual can be completely

independent of the previously visited locations.

Preferential Return: Humans behavior reﬂects an

important property, which is the tendency to re-

turn to the most visited locations in the past (e.g.,

home or work).

(Song et al., 2010) represents the state of the art in

modeling of human mobility. Recently other models

have been proposed to focus on irregularities of

human movement (e.g. (Barbosa et al., 2015)) but in

the context of sensor networks, the regularities are

more important because they represent the “normal”

pattern of movement and hence can be exploited in

information dissemination.

SENSORNETS 2016 - 5th International Conference on Sensor Networks

1.3 Problem Statement

In the beginning of Section 1, we described the frame-

work we are working with. In this framework, wire-

less devices are carried by people who encounter each

other as a part of their social activities. In the context

of sensor networks, as people encounter each other,

their devices also can establish connections each other

by one of the communication technologies available

(e.g., Bluetooth or Wi-Fi). These encounters should

be stored in the devices’ memories if one wants to

make use of the regularities of encounters in infor-

mation dissemination. The tracking of encounters

requires these devices to remember their history of

encounters. The ability to use this past information

of encounters efﬁciently depends considerably on the

size of devices’ history. People however are likely

to encounter many others as part of their daily activ-

ities and keeping track of all these encounters adds

a burden to the sensors since it may not be able to

store all the history due to strong memory constraints

in the devices. In our previous work (Mahmood

et al., 2015), we investigated the issue of memory

requirements and predicted the size of memory that

devices should use for tracking purposes. The ﬁnd-

ings showed that a device needs to be able to track ≈

2.5% of the total number of devices in the environ-

ment (≈0.5% is dedicated for strong ties and ≈2.0%

for weak ties). Moreover, this approximation repre-

sents the maximum number of devices that can be

tracked by a device. When one uses a pre-deﬁned size

of memory for tracking, it may become full and no

space may be available for the new incoming encoun-

ters to be reported during people movement. This

leads to loss some encounters that are perhaps im-

portant to devices and eventually affects, for exam-

ple, data spreading pattern. This situation introduces

the issue of memory management in sensor networks.

To deal with this issue, a decision should be made to

replace one (or more) of the existing item(s) in the

list of encounters with the incoming (new) ones. This

decision should avoid the loss the important encoun-

ters of devices. Therefore, in this work, we propose

a novel social-based approach for memory manage-

ment in sensor networks inspired from two concepts,

namely, social capital in sociology and preferential

return mechanism in human mobility model. In the

former, we calculate the social capital among indi-

viduals based on three indicators: Interaction Level,

Trust Level, and Homophily Level. For the latter, we

involve the recent location of an individual as an indi-

cator to also determine the importance of the current

encounters. Our claim in this work is that using the

aforementioned social concepts can efﬁciently con-

tribute in managing a device’s memory.

This paper is organized as follows: next section

presents the related works, in Section 3 we describe

the details of our memory management model, Sec-

tion 4, we present the main results, and then we con-

clude our paper in Section 5.

2 RELATED WORKS

The concept of social capital has been used in dif-

ferent network applications such as information dis-

semination in social networks (e.g., knowledge trans-

fer among network nodes). (Bosen and Gang, 2010)

explained the effect of social capital on knowledge

transfer and knowledge creation in organizations.

They studied the ability of organizations to use the

existing resources (e.g., knowledge) and the external

resources effectively for their success. In organiza-

tions, team interactions represent an important factor

in knowledge acquisition and creation within the or-

ganization or among organizations. The interactions

among individuals as proposed in (Smith, 2008) and

(Yao et al., 2014) can be incorporated to the forma-

tion social capital. Therefore, team social capital in

an organization represents the intensity of the inter-

actions within a team or among teams. (Bosen and

Gang, 2010) also showed that team social capital af-

fects team knowledge transfer; when teams promote

and develop their social capital using measures such

as trust, they can transfer knowledge effectively. Fur-

thermore, trust can help to decrease risks in the re-

lationships among team’s members and make them

more likely to share their knowledge (Bosen and

Gang, 2010)(Hsu et al., 2007).

(Lin, 1999) investigated the structural features

(e.g., density) of people relations based on the re-

sources that are embedded in them. He found that

networks can provide the necessary conditions for ac-

cessing and using embedded resources (e.g., knowl-

edge). In addition, he proposed a network theory of

social capital that integrates network structural fea-

tures such as density, reciprocity, openness, closeness,

and homophily. For instance, the density of a network

may increase resource sharing among participant in-

dividuals or groups.

Measuring the value of social capital can be

performed in different methods based on network

structure and available parameters. (Burt, 1997)

showed that social capital is affected by social com-

ponent size, density, and hierarchy between individ-

uals. (Larsen et al., 2004) measured the social capi-

tal among neighbors two neighbors interact and trust

each other in the daily activities. (Licamele et al.,

2005) proposed a friendship-event network which is

A Social-based Strategy for Memory Management in Sensor Networks

a speciﬁc form of a social network capturing features

of two inter-related networks (a friendship network

and an event network). The event network describes

events, event participants and organizers. Using these

networks, they infer social capital based on the actor-

organizer friendship relationship. Their dataset con-

tains information from three scientiﬁc conferences for

ten years. In the ﬁndings, they observed that hav-

ing few powerful friends (e.g., high value of social

capital) is more important than having many power-

less friends (e.g., low value of social capital), and this

power can be changed over time. In (Phung et al.,

2013). They proposed an approach to calculate the

social capital of an individual in online social net-

works which uses six indicators: number of friends,

number of community memberships, number of fol-

lowers, number of posts written, number of comments

made per day, and number of comments received per

day. For each indicator, three values are deﬁned: low,

medium, and high based on the activity of an individ-

ual.

(Abdelaal and Ali, 2012) calculated the social

capital in a network out of different wireless networks

by considering three variables: network size, network

density, and the value of a transaction (e.g., collab-

oration) that occur among actors. They found that

social capital is mobilized to empowering communi-

ties to achieve collective telecommunication infras-

tructures. (Bosen and Gang, 2010) and (Zhao and

Wang, 2009) showed how team social capital can sup-

port knowledge transfer more effectively. They used

three social dimensions: structural (e.g., the intensity

of social interactions among team members), cogni-

tive (e.g., shared language, goals, and culture), and

relational (e.g., trust). (Law and Chang, 2012) in-

volved the same social capital factors used in Bosen

and Gang’s work—except the cognitive feature—to

calculate one’s social capital. They observed that

these factors have a signiﬁcant impact on knowl-

edge transfer. (Smith, 2008) distinguished the con-

nections among individuals as follows: Explicit So-

cial Networks (ESNs) and Implicit Afﬁnity Networks

(IANs). ESNs connect actors together based on a

well-deﬁned relationship (e.g., many features in com-

mon). IANs connect actors based on loosely deﬁned

afﬁnities (e.g., less features in common). Smith also

mentioned that social capital is grounded on relation-

ships, individuals’ attributes, and available social re-

sources. Based on the ESNs and IANs, they calcu-

late two types of social capital; bonding and bridg-

ing social capital. (Subbian et al., 2013b)(Subbian

et al., 2013a) calculated social capital based on the

closeness centrality measurement in social networks.

(Sander and Teh, 2014) showed that social capital

can be determined between two individuals based on

three main characteristics: trust, reciprocity, and in-

vestment (e.g., information sharing).

Overall, the descriptions of the works here tell us

that social capital can be estimated. Despite the differ-

ences in the approaches above, what is important for

our work is that the social capital of individuals can

be extracted from their social network or network of

encounters. Our approach is hence quite general be-

cause one may be able to change how the social cap-

ital is calculated and get different results. This may

be useful in special-purpose scenarios in which social

capital may be better deﬁned according to one of the

works above.

3 MODEL DESCRIPTION

As mentioned in Section 1.3, we propose a novel

social-based replacement strategy for managing

memory in sensor networks. This strategy is inspired

from two social concepts: social capital and prefer-

ential return. The main idea behind our approach is

to have a replacement mechanism, by which the im-

portance of an encounter between two nodes can be

determined and eventually helps the decision of re-

placing items in sensors’ memories.

3.1 Calculating Social Capital

(Burt, 2000) pointed that social capital is dynamic and

should consider indicators that also change of time.

Determining these indicators is subject to the nature

of the adopted infrastructure (see Section 2). In our

work, social capital is calculated based on three indi-

cators deﬁned as follows (see also Figure 1):

Social Interactions (ι): This indicator expresses the

level of interactions among network devices that

contributes in measuring the strength of the rela-

tion between two devices. In this work, three as-

pects are used to measure this level:

• Frequency of Encounters (φ): The frequency of

encounters of a pair of device represents how

many times they encounter in a period of time.

• Duration of Encounters (δ): The duration of

a particular encounter represents how long en-

counters tend to last for.

• Regularity of Encounters (ρ): Provides us with

information about the regularity the encounters

of a pair of devices. More precisely, it repre-

sents the time it generally takes until the pairs

encounter each other again (waiting time).

SENSORNETS 2016 - 5th International Conference on Sensor Networks

Figure 1: The deﬁnition of the social capital of a pair of social actors (individuals) A and B. It can be seen that many social

characteristics are embedded in the relation between A and B and the collective value of these characteristics forms social

capital of the pair A,B.

Based on the aforementioned aspects of social in-

teractions, a relation between two devices is con-

sidered strong when they have high rates of fre-

quency and duration, and low rate of regular-

ity. Now, the interaction level (ι

i j

(t)) between a

pair (i,j) can be calculated based on the following

equation:

i j

(t) = φ

i j

(t) + δ

i j

(t) +

i j

(t)

. (1)

Trust (τ): This indicator reﬂects the amount of data

exchanged between two devices. This means how

trustworthy they are to share knowledge (Bosen

and Gang, 2010)(Hsu et al., 2007). In our model,

we measured trust of a pair by counting the num-

ber of messages that two devices have exchanged.

Homophily (η): In Section 1.1, we described the

concept of homophily and showed two aspects

of it; status and value homophily. However, us-

ing both aspects in calculating homophily leads

to our model to be over-parameterized. Based on

the study of (Borgatti et al., 1998) on the factors

that affect social capital, which stated that of sta-

tus homophily on social capital is not signiﬁcant

, we decided to use only value homophily in our

approach:

• Friends In Common (α): Individuals tend to

consider that their friends are like them and also

tend to participate with those who have similar

orientations (McPherson et al., 2001). The ef-

fect of this feature can be on, for example, re-

source sharing among devices.

• Locations In Common (β): Structural posi-

tions of individuals can be used as a reference

of other groups (Festinger, 1950). Yet, indi-

viduals who are more structurally similar are

most likely to have similar vies. Also, they

are more likely to communicate, inﬂuence each

other (Burt, 1982)(Friedkin, 1993) and eventu-

ally form social connections (Cho et al., 2011).

• Distance In Between (γ): The distance between

two individuals plays an important role given

that the distance between individuals is a good

measure of how strong the friendship may be

(Preciado et al., 2012).

According to the description above, the ho-

mophily η

i j

(t) of a pair of devices p(i, j) at time

t is:

i j

(t) = α

i j

(t) + β

i j

(t) + γ

i j

(2)

Now, given the above three indicators the social

capital (SC) of a pair of devices p(i, j) at time t can

be calculated as follows:

i j

(t) = ι

i j

(t) + τ

i j

(t) + η

i j

(t), (3)

3.2 Preferential Return

In the human mobility model that was proposed by

(Song et al., 2010), the movements of an individual

are based on either exploring new locations (explo-

ration) or return to the previously visited locations

(preferential return). However, according to the study

of (Barbosa et al., 2015), the concept of the second

mechanism (preferential return) can be seen in two

different points of view: return to frequently-visited

locations or return to recently-visited locations. Ac-

cording to their ﬁndings, they observed that, in addi-

tion to the tendency to return to the most frequently-

visited locations, the recently-visited locations have

A Social-based Strategy for Memory Management in Sensor Networks

also a high visitation probability. Accordingly, we

incorporated the idea of recently-visited locations as

an indicator in our replacement strategy for memory

management (as we will see in the next section). The

reason of using this indicator in our proposed strategy

is that when two individuals recently visited the same

location, the probability of both to visit this location is

high. Therefore, it is more likely for them to become

friends (Burt, 1982)(Friedkin, 1993)(Preciado et al.,

2012).

3.3 The Anticipatory Strategy

We named our approach Memory Anticipatory Strat-

egy (MAS) because it tries to anticipate whether a cur-

rent encounter is important and hence should be “re-

membered”. The main goal of proposing MAS is to

manage a device’s memory when it reaches the maxi-

mum allowable number of items (device IDs) in mem-

ory. The MAS strategy aims to remove one (or more)

of the current items from device’s memory and re-

place it with one (or more) of the new incoming ones.

Determining the maximum number of items that a

device can store in its memory is based on the number

of devices that can be tracked by a particular device.

In our previous work (Mahmood et al., 2015), we in-

vestigated the issue of memory requirements. We ob-

served that a device can keep track ≈2.5% of devices

in the environment (e.g., a city), we also observed that

strong ties take about ≈0.5% and ≈2.0% for weak

ties. Based on these results, we consider the maxi-

mum number of items that a device can store in its

memory to be ≈2.5% of the total number of devices

in the environment.

While devices move in the simulation environ-

ment, they encounter each other; some of these en-

counters are important while others are not. Ideally

one would prefer to have a a full history of encoun-

ters in the devices’ memories. However, a device’s

memory is limited may become full due to the large

number of encounters. The MAS strategy starts when

a device’s memory contains the maximum allowable

number of items. It performs two basic operations as

follows:

Add Operation: A device adds an item into its mem-

ory if both the current encountered devices re-

cently visited the same location—the most recent

one according to (Barbosa et al., 2015)–this is also

called the adding condition.

Remove Operation: A device removes the item that

has least value of social capital in its memory if

the adding condition holds true.

For a better understanding of how the MAS strategy

works, a scenario is considered as follows: Consider

A as a device with maximum allowable items in mem-

ory equal to 2. The current items in A’s memory are X

and Y with social capital values of 2.5 and 4.0 respec-

tively, and the recent location of A is `

. Furthermore,

consider that A currently encountered two other de-

vices in the environment B and C and their recent lo-

cations `

and `

respectively. The status of A’s mem-

ory is currently full. In such case, our model uses the

MAS strategy to decide whether adding B, C, both,

or neither is necessary. However, to add an item into

A’s memory, the recent-location of B and/or C must

be the same of A’s recent-location. In this scenario,

B has the same recent location of A which is `

. This

means, B is a candidate to be added into A’s memory.

Now, the item that should be removed from A’s mem-

ory must be chosen. MAS chooses the item that has

least value of social capital among A’s items: X in this

case. Finally, MAS removes X then adds B. The rea-

son of choosing B rather than C is the probability of B

to encounter A is high in the future (more important).

This means, the probability of them to be friends is

also high (as explained in Section 3.2). There are also

some other cases that MAS strategy performs, such

that, if two items D and E have the same recent loca-

tion of A, MAS removes two items from A’s memory

that have least value of social capital among A’s items

and adds both D and E.

Algorithm 1 shows the process of removing and

adding item(s) into a device’s memory. In this algo-

rithm, A is a device in the environment. The maxi-

mum number of items that each device can store is

Max

limit

. A contains items that were previously added

items

. Also, E

items

is a list that contains the devices

that are currently near the device A (the list of devices

encountered by A).

Algorithm 1: Illustrates the main steps of MAS strategy for

managing the memory of a device.

1: INPUT: E

items

, A

items

, and Max

limit

2: for each item ∈ (A

items

∩ E

items

) do

3: Update SocialCapital using Equation 3

4: end for

5: MinScItem = Minimum Social Capital ∈ A

items

6: if Length(A

items

) = Max

limit

then

7: for each item ∈ E

items

8: if RecentLocation(item) =

RecentLocation(A) then

9: remove MinScItem from A

items

10: add item into A

items

11: Initiate the social capital of item

12: end if

13: end for

14: end if

SENSORNETS 2016 - 5th International Conference on Sensor Networks

The MAS strategy can be summarized as follows:

• Only local information is used in the processes.

This means a device does not request any external

information (e.g., network level parameters) that

cannot be accessed directly by the device.

• The MAS strategy always provides devices with

weak ties, and avoid keeping only strong ties in

memory. More precisely, when adding an item to

a device’s memory, MAS does not take into con-

sideration its social capital value. The item that is

added may have a smaller social capital than the

removed one.

• MAS calculations are dynamic and on-the-ﬂy, this

means at every time step a device uses the MAS

strategy if an encounter occurs.

• MAS is a social-inspired approach, in which two

social concepts are involved; social capital and the

social behavior of human movements.

4 EXPERIMENTAL RESULTS

4.1 Simulation Environment

The simulation environment we designed is imple-

mented as follows: a squared city of 10 × 10 km,

the city consists of 100 × 100 squared blocks. About

2000 mobile nodes are exponentially distributed in

the city because most metropolises follow this distri-

bution (Grossman-Clarke et al., 2005). Each mobile

node represents an individual who carries a wireless

mobile device (e.g., smartphone, tablet, or laptop).

Wi-Fi technology is used for connecting devices with

range of ≈50 meters, the communication type is peer-

to-peer based. In the environment, each node moves

at a ﬁxed velocity of 1 block per tick. Given the en-

vironment dimensions, a tick is equal to 1.2 minute in

real time considering that each device is carried by

humans who have an average walking speed is ≈5

km/h (Metta et al., 2006). Finally, the nodes move

based on human mobility model (Song et al., 2010)

(as described in Section 1.2).

4.2 Benchmarking Approaches

In this work, two well-known approaches (FIFO and

LRU)

—that are used in page replacement algorithms

in operating systems—were used to benchmark the

our proposed approach (MAS):

We benchmark our approach against these two approaches

because they ﬁt the social framework we are dealing with.

FIFO Algorithm: The ﬁrst-in-ﬁrst-out is the most

popular algorithm when it comes to memory man-

agement in operating systems. In this algorithm,

an item that is added ﬁrst will be removed ﬁrst

(Hopcroft et al., 1983). It is widely used as a base-

line by researchers to benchmark their approaches

(Galvin et al., 2013).

Marking-LRU Algorithm: Marking algorithms

represent a general class of replacement al-

gorithms that are based on the reference (e.g.,

reference bit) to recent use of a page (O’neil et al.,

1993). Least-Recently-Used (LRU) is a marking

algorithm in which a page that is recently used

is marked (e.g., reference bit is set). LRU is

also used for benchmarking other approaches. It

works based on two mechanisms; 1) it memorizes

the pages that has recently been used. 2) Replace

the pages that have not been used for longest time.

In our model, we implemented LRU by applying

its mechanisms as follows. For each device,

we use a list of recent encounters (recent-list),

in which we store only the IDs of the recent

encountered devices without considering other

encounter information

. When device’s memory

reaches the maximum limit and a new encounter

occur, LRU removes the least recent encounter

from device’s memory and adds an item from the

recent-list of that device under the condition that

there is at least one of the new encountered items

in the recent-list (candidate item). Practically,

this means that the candidate item is used more

frequently than the one that is removed.

4.3 Experimental Results

We have implemented the proposed approach (MAS)

plus two others used for comparison purposes:

Marking-LRU and FIFO. The replacement rate is the

metric used for benchmarking MAS. The replacement

rate represents the average replacements of all sensors

over times. Consider the number of replacements of

sensor i at time t is R

(t), then:

ReplacementRate(t) =

∑

i=1

(t)

, (4)

where n is the number of devices that are deployed in

the environment.

Figure 2 depicts the cumulative replacement rate

of the modeled approaches. Clearly, MAS reﬂects

lower replacement rate because it infrequently re-

places items in memory. This is a positive result be-

cause it reﬂects the known fact that our weak and

The recent-list is used only to support the decision of

which item in this list will be stored in the memory of that

device.

A Social-based Strategy for Memory Management in Sensor Networks

          































Figure 2: Cumulative replacement rate of MAS, Marking-

LRU, and FIFO approaches.

MAS

Figure 3: Showing the variations (variance) in dissemina-

tion distance for each of the modeled approaches. The vari-

ance for each approach is obtained from all the runs (100

runs for each approach).

strong ties do not to tend to change frequently. FIFO

replaces an item at every new encounter and Marking-

LRU is more likely to replace an item when encoun-

tering new devices leading to them displaying higher

rate of replacements.

Furthermore, we investigate the behavior of the

three approaches in terms of variance. Figure 3 shows

the variance of each approach. Note that MAS has a

smaller variance than the competitive ones.

However, the above ﬁndings are not signiﬁcant be-

cause they do not provide details on the differences

among the approaches. Therefore, we decide to statis-

tically conﬁrm these results. One-Way ANOVA is used

to show whether the means µ of the approaches are

similar. Our null hypothesis states that all the means

are equal while the alternative hypothesis states that

they are are not equal as follows:

: µ

(MAS)

= µ

(Marking-LRU)

= µ

(FIFO)

: µ

(MAS)

6= µ

(Marking-LRU)

6= µ

(FIFO)

The output of ANOVA shows that F-statistics =

11.26 and the p-value = 0.001. Given these results, we

cannot accept the null hypothesis which conﬁrms that

there is a difference in the means of the approaches.

Now, we need to verify if these differences are sta-

tistically signiﬁcant. To this end, we compute a pair-

wise multiple comparisons among the modeled ap-

proaches using Bonferroni’s Test (Bonferroni, 1936)

with 95% of conﬁdence. It can be observed that the

difference between the pair (MAS, FIFO) with a p-

value of 0.004 is signiﬁcant. Besides that, the p-

value of the pair (MAS, Marking-LRU) equals 0.003

Figure 4: The distribution of social capital among devices

follows a power-law distribution.

which is also signiﬁcant. Therefore, it can be inferred

that our proposed approach (MAS) outperforms the

benchmarking approaches in terms of replacement

rate and variance.

Figure 4 shows the distribution of social capital

among devices in the environment. Clearly, it follows

a power-law distribution. Based on the framework we

are working on, the social interactions of a pair of

devices is symmetric (see Section 3.1). For example,

consider two devices A and B, and also that they are

in each other’s memory. After a while, A removed

B from its memory (A still in B’s memory) due to

an important encounter that causes B to be removed.

However, when A and B encounter again leading to A

adding B back into its memory, it can retrieve from B

all information about their history. This characteristic

is important insofar as it contributes in keeping a more

historical tie strength.

In our previous work (Mahmood et al., 2015), we

involved the interactions among nodes (e.g., frequen-

cies, duration, and regularities of encounters) in or-

der to measure the strength of relations among nodes.

In this work, the calculations of social capital also

include the interactions among network nodes—the

strength of a tie is embedded in the social capital of

the relation itself (Lin, 1999)(Borgatti et al., 1998).

This work then builds from our previous results to

generalize the concepts of interactions into the con-

cept of social capital value.

Our ﬁndings show that the Remove Operation in

the MAS approach avoids losing strong ties from

memory while the Add Operation provides memory

with weak ties in the same time. The overall outcome

is a good balance between weak and strong ties as

shown in Figure 4. The distribution shows few sen-

sors with high social capital (strong ties) and many

values with low social capital (weak ties).

5 CONCLUSIONS

In this work, we proposed a novel social-based re-

placement strategy (Memory Anticipatory Strategy)

SENSORNETS 2016 - 5th International Conference on Sensor Networks

for memory management in sensor networks. This

strategy is based on two social concepts: the idea

of Social Capital in sociology and the Preferen-

tial Return mechanism in human mobility model-

ing. The proposed strategy is benchmarked with two

well-known approaches (FIFO and Marking-LRU) in

memory management literature. The ﬁndings show

that our approach outperforms the comparative ap-

proaches in terms of replacement rate and variations

and that it can successfully maintain a 80-20 ratio be-

tween weak and strong ties. Our approach avoids los-

ing strong ties that are important to a particular device

and provides memory with weak ties.

ACKNOWLEDGEMENTS

The authors would like to thank Marcello Tomasini

for his invaluable inputs during the preparation of this

manuscript.

REFERENCES

Abdelaal, A. and Ali, H. (2012). Measuring social capital in

the domain of community wireless networks. In Sys-

tem Science (HICSS), 2012 45th Hawaii International

Conference on, pages 4850–4859.

Adhikari, K. P. (May 2008). Bridging, linking, and bonding

social capital in collective action. Collective Action

and property rights CAPRi.

Barbosa, H., de Lima Neto, F. B., Evsukoff, A., and

Menezes, R. (2015). The effect of recency to human

mobility. European Physical Journal Data Science (in

press).

Bonferroni, C. E. (1936). Teoria statistica delle classi e cal-

colo delle probabilita. Libreria internazionale Seeber.

Borgatti, S. P. and Foster, P. C. (2003). The network

paradigm in organizational research: A review and ty-

pology. Journal of management, 29(6):991–1013.

Borgatti, S. P., Jones, C., and Everett, M. G. (1998).

Network measures of social capital. Connections,

21(2):27–36.

Bosen, L. and Gang, Q. (2010). Relationship between

team social capital and knowledge transfer: The me-

diated effect of tms. In E-Product E-Service and E-

Entertainment (ICEEE), 2010 International Confer-

ence on, pages 1–4.

Burt, R. S. (1982). Toward a structural theory of action:

network models of social Structure, Perception, and

Action. Academic Press.

Burt, R. S. (1997). The contingent value of social capital.

Administrative science quarterly, pages 339–365.

Burt, R. S. (2000). Decay functions. Social networks,

22(1):1–28.

Burt, R. S. (2001). Structural holes versus network closure

as social capital. Social capital: Theory and research,

pages 31–56.

Cho, E., Myers, S., and Leskovec, J. (2011). Friendship

and mobility: user movement in location-based social

networks. In KDD ’11 Proceedings of the 17th ACM

SIGKDD international conference on Knowledge dis-

covery and data mining, pages 1082–1090, New York.

ACM.

Coleman, J. S. (1988). Social capital in the creation of hu-

man capital. American journal of sociology, pages

S95–S120.

Ehrlich, K. and Carboni, I. (2005). Inside social network

analysis. Boston College.

Festinger, L. (1950). Informal social communication. Psy-

chological review, 57(5):271.

Friedkin, N. E. (1993). Structural bases of interpersonal

inﬂuence in groups: A longitudinal case study. Amer-

ican Sociological Review, pages 861–872.

Galvin, P. B., Gagne, G., and Silberschatz, A. (2013). Op-

erating system concepts. John Wiley & Sons, Inc.

Granovetter, M. S. (1973). The strength of weak ties. Amer-

ican journal of sociology, pages 1360–1380.

Grossman-Clarke, S., Zehnder, J. A., Stefanov, W. L., Liu,

Y., and Zoldak, M. A. (2005). Urban modiﬁcations in

a mesoscale meteorological model and the effects on

near-surface variables in an arid metropolitan region.

Journal of Applied Meteorology, 44(9):1281–1297.

Hopcroft, J. E., Ullman, J. D., and Aho, A. V. (1983). Data

structures and algorithms. Addison-Wesley Boston,

MA, USA:.

Hsu, M.-H., Ju, T. L., Yen, C.-H., and Chang, C.-M. (2007).

Knowledge sharing behavior in virtual communities:

The relationship between trust, self-efﬁcacy, and out-

come expectations. International Journal of Human-

Computer Studies, 65(2):153–169.

Ibarra, H. (1992). Homophily and differential returns: Sex

differences in network structure and access in an ad-

vertising ﬁrm. Administrative science quarterly, pages

422–447.

Johannessen, M. (2012). Social capital and the networked

public sphere: Implications for political social media

sites. In System Science (HICSS), 2012 45th Hawaii

International Conference on, pages 2573–2582.

Kesidis, G., Konstantopoulos, T., and Phoha, S. (2003).

Surveillance coverage of sensor networks under a ran-

dom mobility strategy. In Sensors, Proceedings, vol-

ume 2, pages 961–965. IEEE.

Larsen, L., Harlan, S. L., Bolin, B., Hackett, E. J., Hope, D.,

Kirby, A., Nelson, A., Rex, T. R., and Wolf, S. (2004).

Bonding and bridging understanding the relationship

between social capital and civic action. Journal of

Planning Education and Research, 24(1):64–77.

Law, S. and Chang, M.-K. (2012). Social capital and knowl-

edge sharing in online communities: A mediation

model. In System Science (HICSS), 2012 45th Hawaii

International Conference on, pages 3530–3539.

Licamele, L., Bilgic, M., Getoor, L., and Roussopoulos, N.

(2005). Capital and beneﬁt in social networks. In Pro-

ceedings of the 3rd International Workshop on Link

A Social-based Strategy for Memory Management in Sensor Networks

Discovery, LinkKDD ’05, pages 44–51, New York,

NY, USA. ACM.

Lin, G., Noubir, G., and Rajaraman, R. (2004). Mobil-

ity models for ad hoc network simulation. In INFO-

COM 2004. Twenty-third AnnualJoint Conference of

the IEEE Computer and Communications Societies,

volume 1. IEEE.

Lin, N. (1999). Building a network theory of social capital.

Connections, 22(1):28–51.

Mahmood, B., Tomasini, M., and Menezes, R. (2015). Es-

timating memory requirements in wireless sensor net-

works using social tie strengths. In Local Computer

Networks Workshops (LCN Workshops), (SenseApp)

2015 IEEE 40th Conference on. IEEE.

Marsden, P. V. (1988). Homogeneity in conﬁding relations.

Social networks, 10(1):57–76.

McPherson, M., Smith-Lovin, L., and Cook, J. M. (2001).

Birds of a feather: Homophily in social networks. An-

nual review of sociology, pages 415–444.

Metta, G., Fitzpatrick, P., and Natale, L. (2006). Yarp: yet

another robot platform. International Journal on Ad-

vanced Robotics Systems, 3(1):43–48.

Musolesi, M. and Mascolo, C. (2009). Mobility mod-

els for systems evaluation. In Middleware for Net-

work Eccentric and Mobile Applications, pages 43–

62. Springer.

O’neil, E. J., O’neil, P. E., and Weikum, G. (1993). The lru-

k page replacement algorithm for database disk buffer-

ing. ACM SIGMOD Record, 22(2):297–306.

Phung, D., Gupta, S., Nguyen, T., and Venkatesh, S. (2013).

Connectivity, online social capital, and mood: A

bayesian nonparametric analysis. Multimedia, IEEE

Transactions on, 15(6):1316–1325.

Preciado, P., Snijders, T. A., Burk, W. J., Stattin, H., and

Kerr, M. (2012). Does proximity matter? distance de-

pendence of adolescent friendships. Social networks,

34(1):18–31.

Putnam, R. D. (2000). Bowling alone: The collapse and

revival of american community. In Proceedings of the

2000 ACM Conference on Computer Supported Coop-

erative Work, CSCW ’00, pages 357–, New York, NY,

USA. ACM.

Sander, T. and Teh, P.-L. (2014). Determining the indica-

tors of social capital theory to social network sites. In

User Science and Engineering (i-USEr), 2014 3rd In-

ternational Conference on, pages 264–268.

Smith, M. S. (2008). Social capital in online communities.

In Proceedings of the 2Nd PhD Workshop on Informa-

tion and Knowledge Management, PIKM ’08, pages

17–24, New York, NY, USA. ACM.

Song, C., Koren, T., Wang, P., and Barab

asi, A.-L. (2010).

Modelling the scaling properties of human mobility.

Nature Physics, 6(10):818–823.

Subbian, K., Sharma, D., Wen, Z., and Srivastava, J.

(2013a). Finding inﬂuencers in networks using so-

cial capital. In Advances in Social Networks Analy-

sis and Mining (ASONAM), 2013 IEEE/ACM Interna-

tional Conference on, pages 592–599.

Subbian, K., Sharma, D., Wen, Z., and Srivastava, J.

(2013b). Social capital: The power of inﬂuencers in

networks. In Proceedings of the 2013 International

Conference on Autonomous Agents and Multi-agent

Systems, AAMAS ’13, pages 1243–1244, Richland,

SC. International Foundation for Autonomous Agents

and Multiagent Systems.

Wisconson Center for Education Research (2006). Families

and school together. social capital: social capital as a

theoretical construct. Technical report, University of

Wisconsin-Madison.

Yao, C.-Y., Tsai, C.-C., and Fang, Y.-C. (2014). Under-

standing social capital, team learning, members’ e-

loyalty and knowledge sharing in virtual communities.

Total Quality Management and Business Excellence,

pages 1–13.

Yuan, Y. C. and Gay, G. (2006). Homophily of network ties

and bonding and bridging social capital in computer-

mediated distributed teams. Journal of Computer-

Mediated Communication, 11(4):1062–1084.

Zhao, Y. and Wang, Y. (2009). Research on the knowl-

edge transfer among enterprise network: A perspec-

tive from social capital. In Knowledge Acquisition and

Modeling, 2009. KAM ’09. Second International Sym-

posium on, volume 2, pages 379–382.

SENSORNETS 2016 - 5th International Conference on Sensor Networks