Leveraging Social Behaviour of Users Mobility and Interests for

Improving QoS and Energy Efficiency of Content Services in Mobile

Community Edge-clouds

Vu San Ha Huynh

and Milena Radenkovic

School of Computer Science, University of Nottingham, Nottingham, U.K.

Keywords: Mobile Community Edge-clouds, Network Coding, Content Caching, Multilayer Spatial-temporal Locality.

Abstract: Community network edge-clouds have been attracting significant interests over the recent years with the

emergence of ubiquitous networked devices embedded in our daily activities and increasingly widespread

fully-distributed heterogeneous networks of smart edges offering various applications and services in real

time. This paper proposes EdgeCNC, a novel joint multilayer adaptive opportunistic network-coding

algorithm integrated with adaptive opportunistic content caching service. EdgeCNC exploits the multilayer

spatial-temporal locality of users’ mobility and interests in community network edge-clouds in order to select

highly suitable set of contents to forward, cache and network code to highly suitable set of nodes in order to

enhance QoS, reduce data transmissions and improve energy efficiency. We perform a multi-criteria

evaluation of EdgeCNC performance in realistic Foursquare New York scenario of mobile community edge-

clouds against the benchmark and competitive protocols in the face of dynamically changing users’ publish-

subscribe and mobility patterns. We show that EdgeCNC achieves higher success ratio and data transmission

efficiency while keeping lower delays, packet loss and energy consumption compared to the competitive and

benchmark protocols.

1 INTRODUCTION

Community network edge-clouds (Selimi et al., 2019)

aim to respond to the rapidly increasing demands for

network connectivity and real-time distributed services

in rural and urban communities. According to (Cisco

Index, 2017), global network traffic will be threefold

over the next few years and the provision of intelligent

and adaptive content services (Radenkovic et al., 2018)

closer to the local interest of dynamic geo-temporal

clusters of mobile users will be needed to deal with the

increasing traffic demand. This paper aims to further

improve the reliability and scalability of the intelligent

dynamic edge-cloud networks by reducing data

transmission and improving energy efficiency which

play a vital role regarding the resource constraints and

battery of users’ mobile devices.

We propose EdgeCNC, a novel integration

scheme which combines adaptive and opportunistic

network coding service CafNC (Radenkovic and

Zakhary, 2012) and adaptive edge caching

https://orcid.org/0000-0002-5472-5328

https://orcid.org/0000-0003-4000-6143

CafRepCache (Radenkovic et al., 2018) to improve

the QoS and energy efficiency of content services in

mobile community edge-clouds. We argue that

adaptive network coding (Radenkovic and Zakhary,

2012) and adaptive content caching (Radenkovic et

al., 2018) will enable high-performance efficiency of

content services while reducing the number of

sending packets, avoiding congestion and minimising

energy consumption in dynamic mobile community

edge-clouds. We envisage that EdgeCNC is desirable

in heterogeneous dynamic geo-temporal clusters of

mobile user scenarios such as mobile personal clouds

in pervasive health services (Radenkovic and Huynh,

2016) and cognitive privacy for personal mobile

clouds (Radenkovic, 2016). Building on CafNC and

CafRepCache, EdgeCNC leverages the social power-

law behaviour of users’ mobility and interest to select

highly suitable set of contents to forward, cache and

network code in highly suitable set of relaying nodes

while being adaptive to dynamically changing

resource and energy availability. We evaluate and

382

Huynh, V. and Radenkovic, M.

Leveraging Social Behaviour of Users Mobility and Interests for Improving QoS and Energy Efﬁciency of Content Services in Mobile Community Edge-clouds.

DOI: 10.5220/0009415103820389

In Proceedings of the 10th International Conference on Cloud Computing and Services Science (CLOSER 2020), pages 382-389

ISBN: 978-989-758-424-4

compare EdgeCNC with state-of-the-art in realistic

Foursquare New York scenario of mobile community

edge-clouds. We show that EdgeCNC enhances QoS

of content services such as edge-to-edge delay, packet

loss as well as improves the energy efficiency of

communications between edge-clouds.

The paper begins by providing an overview of the

related work in Section 2. We then introduce the

adaptive multilayer opportunistic network coding

mechanism integrated with adaptive caching in

Section 3. Section 4 evaluates the performance of

different network-coded forwarding and caching

protocols in opportunistic networks. The conclusion

is given in Section 5.

2 RELATED WORK

Network coding approaches (Kouvelas et al., 1998;

Radenkovic, 2004; Radenkovic and Zakhary, 2012)

for dynamic communities (i.e. clusters) have attracted

significant interest in recent years due to their

potential to improve network throughput, reduce

delay and increase data transmission efficiency.

HubCode (Ahmed and Kanhere, 2009) is proposed as

a static forwarding and network coding approach that

utilises highly central nodes as message relays in

order to improve the probability of message delivery.

While using highly central nodes or hubs as relays

may help to deliver messages to the destinations,

CafNC (Radenkovic and Zakhary, 2012) shows that

these hubs may become overloaded when there is an

increasing congestion level, resulting in significant

packet loss and achieve even lower delivery success

ratios. In this paper, we integrate CafNC’s solution

and further refine it with energy awareness to choose

the relaying nodes that are capable of receiving coded

contents without becoming congested and being

energy depleted. (Muhammad and Kang, 2018) states

that network coding is beneficial in content-centric

networks which fundamentally consider multiple

publish-subscribe content-delivery solutions, then

propose a linear-coding techniques CCCN

(Muhammad and Kang, 2018). However, CCCN does

not support capturing and predicting the spatial-

temporal locality of content requests, thus the

network coding rate calculation is not sufficiently

fine-grained.

Authors in (Manzoor et al., 2019) propose a

proactive caching which measures mobility

prediction uncertainty and content request frequency

to predict the most promising prefetching node, thus

eliminating redundancy and improving cache

resource efficiency. The proposed approach achieves

improvement compared to reactive schemes;

however, these gains are limited due to the dynamics

of users’ mobility patterns and content traffic

demand. Authors in (Said et al., 2018) propose a

proactive caching algorithm that caches the contents

from highly influential users within a close group or

community which is discovered by clustering

coefficient based genetic algorithm. (Said et al.,

2018) assumes the assist of infrastructure-based

knowledge and does not support the dynamic network

topologies due to the users’ mobility.

In order to identify and predict more accurately

dynamic spatial-temporal clusters of mobile social

publisher-subscriber, thus improve caching and

network coding performance, exploitation of users'

contextual information for social power-law

behaviour of users’ mobility and content interests is

needed. Research on today’s social networks

(Dabirmoghaddam et al., 2014; Oliveira et al., 2017)

has questioned the validity of the Independent

Reference Model (IRM) and Zipf’s law model (Cha

et al., 2007) of content requests which suggested that

the content is requested scattered randomly and

independently over time. Instead, (Dabirmoghaddam

et al., 2014; Oliveira et al., 2017) argue that content

requests often exhibit both temporal and spatial

locality which is shown in the real-world Foursquare

New York dataset that we used in this paper.

3 JOINT NETWORK CODING &

CACHING SERVICES IN

COMMUNITY EDGE-CLOUDS

Figure 1: Community Edge-Clouds.

Community edge-clouds (Fig. 1) are fully-distributed

self-organised networks of smart edges which form

heterogeneous dynamic geo-temporal clusters of

Leveraging Social Behaviour of Users Mobility and Interests for Improving QoS and Energy Efﬁciency of Content Services in Mobile

Community Edge-clouds

383

users. In this paper, we propose EdgeCNC, a novel

joint multilayer network coding-caching mechanism

which integrates adaptive opportunistic linear

network coding in CafNC (Radenkovic and Zakhary,

2012) with adaptive edge caching service in

CafRepCache (Radenkovic et al., 2018) to improve

the QoS and energy efficiency of multilayer

communications in mobile community edge-cloud

networks.

3.1 Overview of Joint Network Coding

and Caching Services - EdgeCNC

Adaptive congestion-aware forwarding, replication

and caching CafRepCache (Radenkovic et al., 2018)

is a multilayer adaptive caching framework in mobile

Opportunistic Networks. CafRepCache is able to

capture and predict the dynamic spatial-temporal

locality of users’ mobility, content request patterns

and resources as well as the interdependences

between them in order to more accurately and more

responsively cache contents from dynamic clusters of

subscribers (Huynh and Radenkovic, 2019) with

dynamic mobility, complex content requests and

dynamic resource availability. CafRepCache utilises

multi-layer predictive analytics to manage complex

dynamic trade-offs between maximising content

delivery while reducing delay and resource

consumption. This paper extends CafRepCache with

adaptive opportunistic network coding to reduce the

number of sending packets, improve network

throughput and enhance data transmission efficiency.

Adaptive congestion-aware forwarding and

opportunistic network coding CafNC (Radenkovic

and Zakhary, 2012) is a real-time multilayer network

coding mechanism in mobile Opportunistic

Networks. CafNC is able to capture, predict and adapt

to dynamically changing patterns of users mobility,

content interests and resources via multilayer

predictive analytics which drives network coding to

the most suitable set of contents in the most suitable

set of carriers in order to increase content delivery

success ratio while minimizing delay, reducing

redundancy and waste of network resources in multi-

source, multi-path and multicast forwarding

algorithm. CafNC adaptively balances the complex

dynamic trade-off between under and over-utilised

coding nodes, as missing encoding opportunities may

potentially cause increased delays and lower success

ratios while extensive coding at the time of high

congestion may result in significant packet loss

(Radenkovic and Zakhary, 2012). CafNC encodes

together contents that share the same interest or are

sent to the same dynamic geo-temporal cluster of

subscribers. This paper refines CafNC with an

energy-aware network coding rate controlled by the

network coding threshold metric in order to improve

the data transmission and energy efficiency of

communications in community edge-clouds.

EdgeCNC is built on CafRepCache (Radenkovic

et al., 2018) to allow integration with CafNC

(Radenkovic and Zakhary, 2012) and enable energy

awareness which enhances the QoS while reducing

the number of sending packets and improving the

energy efficiency of mobile community edge-cloud

networks. In line with CafNC (Radenkovic and

Zakhary, 2012) and CafRepCache (Radenkovic et al.,

2018), EdgeCNC is adaptive, fully-distributed,

opportunistic, collaborative and is able to perform

multilayer spatial-temporal predictive analytics and

heuristics of multivariate mixed data (e.g. mobility,

content and resource) as well as collaborate and

exchange its local observations with other neighbour

nodes in

order to detect and exploit coding and

caching opportunities in the accurate and responsive

manner without the need of global knowledge.

3.2 Energy-Aware Network Coding in

EdgeCNC

We refine CafNC (Radenkovic and Zakhary, 2012)

with energy consideration by proposing a novel

energy-aware network coding threshold metric in

order to improve data transmission and energy

efficiency. In line with CafNC (Radenkovic and

Zakhary, 2012), we model linear network coding

process in EdgeCNC system as a network G

consisting of a set of nodes N and a set E of edges. A

set of contents that can be requested in the network is

denoted as O. A node n



∈N receives c





(the

encoded form of content o



∈O) directly from a relay

node and linearly combined encoded symbol



,τ



+τ



) from another node in which requested

content is segmented linearly into the encoded

symbols and the encoded symbols are received

simultaneously. The coding vector is defined as a set

of coefficients associated with each coding point. We

define the transform matrices T as follow:

T



(1)

We consider φ



, as the symbol received by the

caching point or intermediary node n



. The node can

then recover all N encoded symbols by resolving the

linear equation:









T













(2)

CLOSER 2020 - 10th International Conference on Cloud Computing and Services Science

384

Energy efficiency plays a vital role in EdgeCNC

solution due to the constraint in the battery life of

users’ mobile devices. Thus, we propose to refine the

NCThreshold from CafNC (Radenkovic and

Zakhary, 2012) to include energy awareness (in

addition to mobility, resources and contents).

The EnergyNCThreshold parameter to control the

coding rate is resolved based on energy availability



. EnergyNCThreshold is a configurable parameter

between 0 and 1 measured as:

Energ

NCThreshold

1,E



0

cost





,E



0

(3)

in which E



is the currently available energy level of

a node n



, cost



is the energy cost to transmit signals,

forward and code the content o



. γ is a control

parameter which will be set depending on the

importance of a node such that 0 < γ ≤ 1. For example,

γ will be set to low if a node n



is important and has

to be protected from being battery drainage.

EdgeCNC justifies the number of coding a node

carries out based on EnergyNCRate parameter:

Energ

NCRateEdgeCNCTotalUtil

,



∗1

Energ

NCThreshold

(4)

in which EdgeCNCTotalUtil is the total multi-layer

utility of EdgeCNC in line with (Radenkovic and

Zakhary, 2012) and (Radenkovic et al., 2018).

Table 1: EdgeCNC pseudocode.

EdgeCNC – Arrival of Content Request/Interest

When the request/interest of a content is received at a node:

for each Contact c do:

c.socialHeur = exchSocHeur (Contact, Contacts.reputation)

c.resourceHeur = exchResHeur (Contact, Contacts.reputation)

c.contentHeur = exchPopularity(Contact, Contacts.reputation)

c.calculateUtility(c.heuristics)

ListUtils.insert(Contact.Utility)

end for



,x



) = in-networkCaching&NetworkCoding(ListUtils)

if x



== 1

Node is set to be a cache candidate of the Content

if(EnergyNCRate >= 0.5 & lenth(contents) ≥

EnergyNCThreshold) do NCodeCache(contents) else

cache(contents)

else if x



== 1

forward Interest to Contact with highest utility value

if(EnergyNCRate >= 0.5 & lenth(contents) ≥

EnergyNCThreshold) do NCodeForward(content) else

forward(content)

end if

In line with CafNC (Radenkovic and Zakhary,

2012), we argue that when EnergyNCRate is high,

the level of congestion is relatively low, EdgeCNC

increases the probability of coding with low risk of

packet loss. When EnergyNCRate is low, there may

be a high level of congestion, EdgeCNC stricts the

coding criteria as the risk of high packet loss. Packets

of the content are compared to the

EnergyNCThreshold. If the number of packets for

the content exceeds the threshold, then the node

network codes the contents and sends its coded

contents to the corresponding neighbour node. If the

threshold is higher than the number of content

packets, EdgeCNC forwards the contents without

coding them. The caching node generates a coding

vector and performs a linear combination of the

packets cached that are targeted to the same dynamic

cluster of subscribers. The subscribers will collect

packets and when the number of these packets

reaches a certain number, the subscribers will decode

them and deliver them to the upper network layer. We

provide the pseudo-code of EdgeCNC in Table 1.

4 EVALUATION

This section discusses the performance of EdgeCNC

across multiple metrics against state-of-the-art

network coded forwarding and caching protocols

including HubCode (Ahmed and Kanhere, 2009),

CafNC (Radenkovic and Zakhary, 2012), CCCN

(Muhammad and Kang, 2018) and CafRepCache

(Radenkovic et al., 2018). The overall performance is

measured by different criteria: success ratio, latency,

packet loss, number of nodes coding, network coding

cost (measured as the ratio between

encoding/decoding time and total time) and average

energy consumption.

We use real-world mobility trace (Scott et al.,

2009) and real-world content requests Foursquare

New York trace (Dingqi et al., 2015) in ONE

simulator (Keränen et al., 2009). We run our

simulation for 7 days with a total of 100 nodes and

contents in the network. In line with (Cha et al.,

2007), we vary the content size from 1MB to 8.4MB

while the request packet size ranges from 8kB to 128

kB. In this paper, we fix the number of content

publishers (15% of nodes) while varying the number

of Foursquare users (subscribers) who have interests

in the contents. All experiments are repeated ten times

and averaged to give us statistically sufficient

diversity to evaluate EdgeCNC performance in

different publish-subscribe contexts.

Leveraging Social Behaviour of Users Mobility and Interests for Improving QoS and Energy Efﬁciency of Content Services in Mobile

Community Edge-clouds

385

Figure 2a: Spatial locality of mixed popular content

requests in Foursquare New York.

Figure 2b: Temporal locality of mixed popular content

requests in Foursquare New York.

Fig. 2a & b show the spatial and temporal

correlation of content traffic (i.e. temporal requests

pattern of mobile subscribers) for different contents

in different locations in Foursquare dataset. The

locations of mobile subscribers feature different

degrees of similarity in the content request such that

two locations which are relatively close to each other

have similar request patterns compared to that of

another location which is far away from others (Fig.

2a). This shows that there is a strong correlation

between the geographical diversity of the users and

their requested contents. Fig. 2b shows the temporal

locality of content traffic during weekdays and

weekend that the content is not requested randomly

and independently, rather a content might be of

particular interest at a certain period of time interval

before its popularity gradually decreases.

Figure 3: Edge-to-edge success ratio vs. Number of

Foursquare users.

Figure 4: Average edge-to-edge latency vs. Number of

Foursquare users.

As EdgeCNC integrates multilayer adaptive

opportunistic CafNC and CafRepCache, it is able to

utilise multilayer multidimensional predictive

analytics of in-network delays and congesting rates of

nodes and their ego networks in order to minimise

delays of dynamic edge-to-edge content retrieval in

community edge-cloud services. EdgeCNC extends

CafRepCache with adaptive network coding

mechanism to reduce the number of sending packets

and thus improve overall network throughput.

EdgeCNC refines network coding rate in CafNC with

energy awareness (in addition to mobility, resources,

contents) which enhances the QoS while reducing the

number of sending packets and avoiding energy

depletion. Furthermore, EdgeCNC’s integrated

CLOSER 2020 - 10th International Conference on Cloud Computing and Services Science

386

adaptive content caching mechanism allows the

requested contents to be mainly found in the caching

points rather than getting the content from the

publishers, thus reduce significantly the edge-to-edge

latency. HubCode and CCCN have the worst

performance (below 60% success ratio, above 6.1 min

delay) due to its static and non-adaptive network

coding in the face of increasing congestion levels.

Figure 5: Edge-to-edge packet loss vs. Number of

Foursquare users.

Fig. 5 shows the packet loss that EdgeCNC and

other competing protocols experience in the face of

increasing congestion levels through the increasing

number of mobile Foursquare users. EdgeCNC,

CafRepCache and CafNC achieve the lowest packet

loss rates (below 18%) followed by CCCN (43%) and

HubCode (56%). This is because EdgeCNC,

CafRepCache and CafNC utilise fully-distributed

predictive analytics and heuristics of how likely the

nodes and their ego networks are about to congest,

size of the buffer, the bandwidth available and

computational resources of a node in order to avoid

forwarding and network coding to congested nodes or

regions where congestion may happen. CafRepCache

without network coding has relatively lower packet

loss compared to EdgeCNC and CafNC as dropping

network-coded packets, even occasionally, may

result in increased packet loss. HubCode has lowest

QoS performance in terms of packet loss rates

(55.4%) as it constantly network codes the contents,

even in the event of congestion which increases

significantly the packet loss compared to only

forwarding a single packet. CCCN has significantly

higher packet loss rate compared to EdgeCNC,

CafRepCache and CafNC as CCCN does not support

congestion-aware and energy-aware network coding.

Figure 6: Average % of node coding vs. Number of

Foursquare users.

Figure 7: Average % of Time Coding vs. Number of

Foursquare users.

Fig. 6 and 7 show the average percentage of

network coding nodes and the average percentage of

time duration out of total simulation time a node

performs network coding in EdgeCNC, CCCN,

CafNC and HubCode in the face of the increasing

number of mobile Foursquare users. We observe that

EdgeCNC and CafNC use 18-21% of the nodes to

encode/decode traffic 29-35% of the time. HubCode

is 400% worse off than others as it uses 10% of the

nodes to network code traffic all of the time (99%).

HubCode, as a static forwarding coding approach,

misses many coding opportunities even when

congestion is low. When the congestion level is

increased, HubCode forwards and codes at a static

rate that increases the risk of packet loss.

Leveraging Social Behaviour of Users Mobility and Interests for Improving QoS and Energy Efﬁciency of Content Services in Mobile

Community Edge-clouds

387

Figure 8: Average Energy Consumption (J) vs. Number of

Foursquare users.

Fig. 8 shows that EdgeCNC consumes

significantly less average energy (5095J) compared to

HubCode (9824J) and CCCN (8012J) due to

EdgeCNC utilising multilayer predictive analytics

and heuristics for dynamic resources and energy (in

addition to mobility and contents) which allows it to

improve data transmission efficiency, reduce waste of

network resources and energy consumption (while

the HubCode and CCCN do not support congestion

and energy awareness). EdgeCNC’s energy

consumption is less than 10-20% compared to

CafRepCache (5412J) and CafNC (6039J). This is

because EdgeCNC extends CafRepCache and CafNC

with adaptive network coding and caching which

reduce significantly the number of sending packets,

thus improve energy efficiency. EdgeCNC profits

from its refined energy-aware network coding rate in

order to cache and network code contents of dynamic

spatial-temporal clusters of subscribers in a more

energy-efficient manner, thus minimise redundancy

and decrease average energy consumption.

5 CONCLUSIONS

This paper proposes EdgeCNC, a novel adaptive

opportunistic joint network coding and caching

mechanism in mobile community edge-clouds.

EdgeCNC integrates multilayer adaptive network

coding scheme CafNC (Radenkovic and Zakhary,

2012) and adaptive content caching service

CafRepCache (Radenkovic et al., 2018) in order to

improve QoS and energy efficiency for community

edge-clouds. EdgeCNC is able to exploit the

multilayer spatial-temporal locality of users’

mobility, interests and resources in order to select

highly suitable set of contents to forward, cache and

network code in highly suitable set of nodes. Note

that our focus is to design an algorithm across trusted

collaborators in the community network edge-clouds,

thus malicious-behaviour protection is out of the

scope of this paper. We performed an extensive

evaluation of EdgeCNC in real-world connectivity

traces and use a realistic dynamic changing content

request dataset in order to compare EdgeCNC

performance versus competitive network coded

forwarding and caching algorithms: HubCode

(Ahmed and Kanhere, 2009), CCCN (Muhammad

and Kang, 2018), CafNC (Radenkovic and Zakhary,

2012) and CafRepCache (Radenkovic et al., 2018).

We show that our proposal performs better than the

state-of-the-art solutions, it improves the content

delivery success ratio while reducing the number of

data transmissions, resulting in lower packet loss,

delay and energy consumption.

We envisage that the jointly adaptive network

coding-caching mechanism proposed in this paper

will be applied to many application scenarios of

mobile community edge-clouds that include both

social and other complex types of networks.

REFERENCES

Ahmed S. and Kanhere S. Hubcode: message forwarding

using hub-based network coding in delay tolerant

networks. Proceedings of the 12th ACM international

conference on Modeling, analysis and simulation of

wireless and mobile systems ACM, p. 288–296, 2009.

Cha M., Kwak H., Rodriguez P., Ahn Y.-Y., and Moon S. I

tube, you tube, everybody tubes: analyzing the world’s

largest user generated content video system.

Proceedings of the 7th ACM SIGCOMM conference on

Internet measurement, ser. IMC ’07, p. 1–14. 2007

Dabirmoghaddam A., Barijough M. M., and Garcia-

LunaAceves J. Understanding optimal caching and

opportunistic caching at ‘the edge’ of information-

centric networks. Proc. ICN. USA, pp. 47–56. 2014.

Dingqi Y., Daqing Z., Vincent W. Z., Zhiyong Y. Modeling

User Activity Preference by Leveraging User Spatial

Temporal Characteristics in LBSNs. IEEE Trans. on

Systems, Man, and Cybernetics: Systems, (TSMC),

45(1), p. 129-142, 2015.

Huynh, V. S. H. and Radenkovic, M. Interdependent Multi-

layer Spatial Temporal-based Caching in

Heterogeneous Mobile Edge and Fog Networks.

Proceedings of the 9th International Conference on

Pervasive and Embedded Computing and

Communication Systems PECCS. P. 34 – 45. 2019.

CLOSER 2020 - 10th International Conference on Cloud Computing and Services Science

388

Index, C. V. N. Global mobile data traffic forecast update,

2016–2021 white paper. Cisco: San Jose, CA, USA.

2017.

Karbaschi G., Viana A., Martin S., and Al Agha K. On

using network coding in multi hop wireless networks.

Personal, Indoor and Mobile Radio Communications,

IEEE 20th International Symposium. p. 420–424. 2009.

Keränen A., Ott J. and Kärkkäinen T. The ONE Simulator

for DTN Protocol Evaluation. SIMUTools'09: 2nd

International Conference on Simulation Tools and

Techniques. Rome. 2009.

Kouvelas, I., Hardman, V. and Crowcroft, J. Network

adaptive continuous-media applications through self

organised transcoding. Proceedings of the Network and

Operating Systems Support for Digital Audio and

Video. 1998.

Manzoor, S., Mazhar, S., Asghar, A., Mian, A. N., Imran,

A. and Crowcroft, J. Leveraging mobility and content

caching for proactive load balancing in heterogeneous

cellular networks. Transactions on Emerging

Telecommunications Technologies. 2019.

Muhammad B., Kang S. G. Network-Coding Approach for

Information-Centric Networking. IEEE Systems

Journal, p.1–10, 2018.

Nicosia V., Tang J., Mascolo C., Musolesi M., Russo G.,

and Latora V. Graph metrics in temporal networks.

Springer Temporal Networks, p. 15–40. 2013.

Oliveira E. M. R, Viana A., Naveen K., Sarraute C. Mobile

data traffic modeling: Revealing temporal facets.

Computer Networks, 112:176–193, 2017.

Radenkovic, M. A framework for building and deploying

the multiparty audio service for collaborative

environments. Presence: Teleoperators & Virtual

Environments, 13(6), pp.708-725. 2004.

Radenkovic, M. Cognitive privacy for personal clouds.

Mobile Information Systems, 2016.

Radenkovic, M. and Huynh, V. S. H. Low-cost mobile

personal clouds. International Wireless

Communications and Mobile Computing Conference

(IWCMC) (pp. 836-841). 2016. IEEE.

Radenkovic M., Huynh V.S.H. and Manzoni P. Adaptive

real-time predictive collaborative content discovery

and retrieval in mobile disconnection prone networks.

IEEE Access, 6, pp.32188-32206. 2018.

Radenkovic M. and Zakhary S. Flexible and dynamic

network coding for adaptive data transmission in dtns.

IEEE 8th International Wireless Communications and

Mobile Computing Conference (IWCMC), p. 567–573,

2012.

Said, A., Shah, S., Farooq, H., Mian, A., Imran, A. and

Crowcroft, J. Proactive Caching at the Edge

Leveraging Influential User Detection in Cellular D2D

Networks. Future Internet, 10(10), p.93. 2018.

Scott, R. Gass, J. Crowcroft, P. Hui, C. Diot, A. Chaintreau.

CRAWDAD data set Cambridge/haggle. May 2009.

Selimi, M., Cerdà-Alabern, L., Freitag, F., Veiga, L.,

Sathiaseelan, A. and Crowcroft, J. A lightweight service

placement approach for community network micro-

clouds

. Journal of Grid Computing, 17(1), pp.169-189.

2019.

Leveraging Social Behaviour of Users Mobility and Interests for Improving QoS and Energy Efﬁciency of Content Services in Mobile

Community Edge-clouds

389