0% found this document useful (0 votes)

19 views

Clustering Techniques

This article surveys clustering techniques for topology management in Internet of Things (IoT) networks that resemble ad-hoc networks like Wireless Sensor Networks (WSNs). It discusses how WSN clustering can potentially be applied to IoT networks given shared characteristics of transmitting data hop-by-hop. However, clustering IoT networks is challenging due to heterogeneity, mobility, and integration with technologies like edge computing and 5G. The survey aims to investigate applicability of existing WSN clustering to IoT networks and examine clustering in modern contexts.

Uploaded by

jyothi

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

19 views

Clustering Techniques

Uploaded by

jyothi

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 32

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TNSM.2020.3035315, IEEE
Transactions on Network and Service Management
IEEE TRANSACTIONS ON NETWORK AND SERVICE MANAGEMENT, VOL., NO., 1

A Survey and Future Directions on Clustering:

From WSNs to IoT and Modern Networking
Paradigms
Amin Shahraki, Amir Taherkordi, Øystein Haugen, and Frank Eliassen

Abstract—Many Internet of Things (IoT) networks are created WSNs, e.g., Internet of Vehicles (IoV). In such networks, IoT
as an overlay over traditional ad-hoc networks such as Zigbee. nodes communicate with each other as an overlay network
Moreover, IoT networks can resemble ad-hoc networks over on top of an existing ad-hoc network protocol, e.g., Zigbee.
networks that support device-to-device (D2D) communication,
e.g., D2D-enabled cellular networks and WiFi-Direct. In these The basic building blocks of these types of IoT networks
ad-hoc types of IoT networks, efficient topology management are often WSNs in which sensors, along with multitude of
is a crucial requirement, and in particular in massive scale everyday objects, communicate, interact and share data on a
deployments. Traditionally, clustering has been recognized as a massive scale [5]. In addition, thanks to the possibility of
common approach for topology management in ad-hoc networks, device-to-device (D2D) communication in networks such as
e.g., in Wireless Sensor Networks (WSNs). Topology management
in WSNs and ad-hoc IoT networks has many design common- cellular networks, WiFi and Bluetooth, the IoT nodes in such
alities as both need to transfer data to the destination hop networks can communicate in an ad-hoc manner [6]. This will
by hop. Thus, WSN clustering techniques can presumably be allow more efficient interaction with co-located IoT nodes and
applied for topology management in ad-hoc IoT networks. This reducing the network overhead. Moreover, there are many IoT
requires a comprehensive study on WSN clustering techniques applications that run over ad-hoc and MANET networks, such
and investigating their applicability to ad-hoc IoT networks.
In this paper, we conduct a survey of this field based on the as healthcare [7], smart cities [5], [8], vehicular networks [9],
objectives for clustering, such as reducing energy consumption military applications [10], and smart agriculture [11].
and load balancing, as well as the network properties relevant In ad-hoc IoT networks, topology management is a critical
for efficient clustering in IoT, such as network heterogeneity requirement for efficient and scalable management of the
and mobility. Beyond that, we investigate the advantages and network, as well as the applications deployed over such
challenges of clustering when IoT is integrated with modern
computing and communication technologies such as Blockchain, networks. In ad-hoc networks like WSNs, clustering has
Fog/Edge computing, and 5G. This survey provides useful insights been introduced as the most popular approach for topology
into research on IoT clustering, allows broader understanding of management. Clustering techniques divide the network to
its design challenges for IoT networks, and sheds light on its groups of nodes and distribute network functions among the
future applications in modern technologies integrated with IoT. group members to improve efficiency in, e.g., collecting and
Index Terms—IoT, Clustering, WSNs, Survey, 5G, SDN, Edge, forwarding data, resource management, and supporting QoS.
Fog, Blockchain, NFV Many clustering techniques have been proposed for WSN
topology management, such as LEACH [12], HEED [13] and
I. I NTRODUCTION TEEN [14], to name a few.
Smart devices have facilitated the pervasive presence of Like WSNs, ad-hoc IoT networks can use clustering for
various things, interacting and cooperating with each other topology management to meet the above performance needs,
through unique addressing schemes—Internet of Things (IoT). as well as IoT-specific challenges, e.g., scalability of the
IoT, introduced first in 2008-2009 [1], connects billions of network [1]. Sharing many fundamental characteristics with
devices around the world on top of different network infras- WSNs implies that ad-hoc IoT networks can potentially utilize
tructures, mainly the Internet. IoT aims to integrate different WSN clustering techniques without having to design them
traditional and next-generation network technologies to work from scratch. However, clustering in IoT can be challenging
simultaneously in a common infrastructure and support differ- due to high heterogeneity and mobility of IoT nodes, and
ent ubiquitous applications [2]. Compared to other networks integration of IoT with recent computing and networking
such as WSNs, IoT nodes are highly heterogeneous [3] thanks paradigms such as Edge computing and 5G networks. There-
to their broad usage in divers application domains and being fore, conducting a comprehensive study on existing WSN
a key component of Cyber-Physical Systems (CPS) [4]. clustering techniques and investigating their applicability to
Many IoT networks resemble ad-hoc networks, following IoT networks would be significantly advantageous for the IoT
the same pattern of data transmission to the Internet as research community. Such a survey work can be further com-
pleted with reviewing clustering techniques that are already
Amin Shahraki, Amir Taherkordi and Frank Eliassen are with the Depart- being used in IoT networks. In conducting such an extensive
ment of Informatics, University of Oslo, Norway survey, it is crucial to define clear goals on what aspects of the
Amin Shahraki and Øystein Haugen are with Faculty of Computer Sciences,
Østfold University College, Halden, Norway existing techniques should be investigated, and how the scope
of applicability to IoT should be envisioned. Considering the

1932-4537 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Gothenburg. Downloaded on December 19,2020 at 10:33:07 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TNSM.2020.3035315, IEEE
Transactions on Network and Service Management
IEEE TRANSACTIONS ON NETWORK AND SERVICE MANAGEMENT, VOL., NO., 2

performance needs of ad-hoc IoT networks, the objectives II. C LUSTERING : BASIC C ONCEPTS AND TAXONOMY
of clustering, such as reducing energy consumption, load Topology management is one of the main challenges in es-
balancing, improving connectivity, etc. should be central to tablishing networks, especially in ad-hoc networks [23]. Clus-
such a survey. tering, as a type of topology management technique, improves
Existing surveys. Most existing clustering surveys study the efficiency by dividing the network to groups of nodes and
and compare WSN clustering techniques and efficiency of distributing network functions among the group members, e.g.,
them [15], [16], [17], [18], not comparing the objectives of collecting and forwarding data, and resource management.
those techniques and more importantly their applicability in Various network types have utilized clustering for topology
IoT [19]. Most of them cover only main clustering techniques, management, such as MANET [24], VANET [25], WSN [15],
such as LEACH [12], HEED [13] and FLOC [20], proposed in and IoT [26]. From the application layer viewpoint, clustering
different forms and extensions. Some other surveys consider techniques have been introduced for different types of needs
only one parameter in designing clustering techniques or such as resource allocation [27], applying reputation models
network infrastructures like unequal clusters [18]. Another [28], service discovery [29], intrusion detection [30], fault
category of surveys covers the techniques derived from the monitoring [31], and anomaly detection [32]. A cluster is
above main techniques to improve clustering performance composed of a number of nodes (i.e., members) and has
or support application-specific clustering requirements. For one or more Cluster Heads (CH) to manage the members
example, in [21] the authors indicate that there are more than and shared resources. Moreover, CHs can collect, fuse, and
60 extended versions of the LEACH protocol in the literature. process members’ data, and transfer it to gateway(s). Each
Besides, in some survey papers, e.g., [22], clustering has been network can have one or more gateways (also known as base
studied from the viewpoint of reducing energy consumption, stations or sinks) that connect CHs to outside of the network.
while it is not the sole objective of clustering. Clustering can address several conventional quality-related
This survey work is distinguished from the others by objectives, such as reducing resource consumption, improving
focusing on applicability of clustering in IoT networks and load balancing and QoS, and fault tolerance. We discuss those
feasibility of migrating existing WSN clustering techniques to objectives in detail in Section V.
IoT networks. The main contributions of this paper include:
• reviewing existing clustering techniques in WSNs from the A. Clustering Structural Models
objectives viewpoint and providing an insightful statistical Two primary aspects of clustering include grouping nodes
analysis (cf. Section III); and allocating responsibilities. Grouping methods are gen-
• investigating applicability of relevant WSN clustering tech- erally based on the structure of Voronoi diagrams, but also
niques to IoT networks with respect to network properties can be non-Voronoi like chain or spectrum structures. In the
and identified objectives (cf. Section V); Voronoi structure, a 2D or 3D network environment is divided
• for each clustering objective: studying existing IoT clus- into several unequal sections, called clusters. Each cluster
tering solutions and discussing the associated challenges in possesses some nodes and possibly interacts with other clusters
using clustering in IoT (cf. Section V); or gateways directly or through neighbor clusters. In the chain
• investigating the advantages of clustering when IoT is inte- structure, nodes in a cluster connect to each other to reach
grated with modern computing technologies and paradigms the corresponding CH. Each node has only two connections
(i.e., Blockchain, Fog and Edge, Software-Defined Network- with neighbors in the chain to reach the CH: one outgoing
ing and Network Function Virtualization, and 5G), as well connection to the next hop, and one incoming connection
as exploring the challenges in applying clustering techniques from another node to the current node as the next hop. In the
in these types of integration (cf. Section VI). spectrum structure, angles of nodes to base stations (BSs) are
Review methodology. The methodology adopted for conduct- as important as the distance to BS for cluster establishment.
ing this survey consists of the following steps. First, we Node angles are generally captured by the Scanning Sweep
extracted the list of main WSN clustering techniques from the method [33]. In both spectrum and chain structures, layering
relevant papers in reputable conferences and journals, such of the network can also be performed to enable multi-hop
as ICCCN, WCNC, GLOBCOMM, ICPS, CNCS, SECON, data transmission and improve the efficiency of the network,
IPDPS, ICDCS, INFOCOM, EWSN, PerCom, SenSys, IT- especially in terms of resource consumption. Figure 1 depicts
PDS, IEEE IoTJ, ITWC, ITN, ITVT, and ATSN. Based on the above three clustering structures.
the extracted relevant papers, we then checked their references Cluster establishment methods: There are generally two
and related work to find any other papers that were concealed. methods to establish clusters in a network: i) Determining
Having the main clustering techniques compiled, we finally clusters by grouping nodes and then selecting one or more
searched for all other research works that either cited the nodes as CH(s), and ii) Selecting CHs first and then inviting
main techniques like LEACH or proposed their own clustering other nodes to join a neighbor CH. Joining of nodes to a
technique. For that, we read and refined about 500 papers. cluster is mainly based on the physical proximity of nodes
Finally, over 250 papers were extracted as the distinguished and other parameters such as cluster size, number of nodes,
clustering techniques proposed for WSNs and IoT. We filtered and balancing network load and resource consumption. In
the papers based on their quality, publication channel, and addition, high-level parameters can be defined as clustering
number of citations. criteria, such as application-specific requirements, local data

because of resource depletion, thereby the CH role should

Chain be rotated among appropriate nodes during the network life-
Structure time. With respect to re-organization of clusters, events like
network congestion, dead nodes, and unbalanced network
Structure
Voronoi
load can also trigger re-clustering. There are two methods
to trigger re-clustering: i) Time-based method: The network
will be re-clustered at a certain time to balance resource
consumption among nodes. This method is generally used
in homogeneous networks with predetermined network load,

S ec
t
Sp
ru trum
Member and ii) Event-based method: An event triggers part of or

ctu
re
CH the whole network to re-select CHs and possibly re-cluster,
Inter cluster routing
Intra cluster routing e.g., exceeding resource usage thresholds like energy, CPU,
bandwidth consumption, or high resource consumption in a
Fig. 1. Different structures of clustering techniques
certain time. The combination of time-based and event-based
methods can be used for re-clustering as well. Depending on
processing needs (e.g., data fusion and compression), and
which re-clustering conditions hold first, the corresponding
resource sharing. With respect to CH selection, there are
method will be triggered.
various techniques as described below:
Data forwarding in clustering techniques: The network data
• In some clustering techniques, resource-rich nodes are can be transferred in its raw format or as a fused value (i.e.,
predetermined as fixed CHs for the whole network life data fusion). In clustered networks, CHs are used to gather
time [34]. In networks with homogeneous or resource- and fuse the members’ data. The CH can transmit individual
constrained nodes, this method is not efficient. Even in data items to BS or send the fused values. Since common
a heterogeneous network, being CH for a long period of clustering techniques are based on the proximity of the nodes
time will drain the node power quickly, leading to node providing data for the same application, it would be possible
death. Moreover, in case of fixed CHs, mobile nodes and to perform data aggregation in CHs and reduce the amount
dynamicity of the network can unbalance the number of of data to be transmitted. In addition, in some cases, CHs are
members and/or cluster load, causing network congestion able to compress data to reduce the data volume. Last but not
and inefficient resource consumption. least, data can be processed locally in CHs and results can
• In some clustering techniques, randomness is the solution to be sent to the gateway(s). There are two methods to transmit
circulate the CH responsibility among nodes [12]. Although packets from CHs to BS(s):
this is beneficial in homogeneous networks, accidental unfair
• CHs can send data directly to BS. In the direct commu-
randomness can lead to network congestion and chronic
nication, each CH transmits data directly to BS(s) causing
energy consumption in some CHs.
energy depletion in case of a long distance (d) based on
• The most common solution for such issues is the conscious
Eq.1 [37].
CH selection method in which CHs are selected based on
Etransmit = F (d2 ) (1)
the circumstances of the nodes and the network [35], [36].
In this method, appropriate nodes are actively selected as • CHs can use middle nodes (often other CHs) to forward data
CHs based on parameters like available resources, location, to the gateway(s), called inter-cluster routing. Although it
and number of neighbors. reduces energy consumption, it increases delay. This method
The CH selection process can be performed either in a also needs establishing efficient routes from CHs to the
centralized or distributed way. In the centralized model, CH gateway(s) causing network overhead.
selection parameters are gathered, compared, analyzed and Besides the communication between CHs and BS(s), nodes
processed in a central node (i.e., generally BS). Although in in a cluster can connect to their CHs directly or indirectly,
this model network-wide comparison of parameters can be called intra-cluster routing. However, it is used in special cases
performed, it often imposes high overhead due to transmitting e.g., blind spaces or big-sized clusters. This routing method
many management packets, in particular in large and/or highly can also impact QoS, e.g., increasing delay [37], [38].
dynamic networks with the CH re-selection capability. On the
other hand, distributed methods impose less overhead, but due III. C LUSTERING IN WSN S
to relying on local CH selection criteria, selected CHs cannot WSNs are a network of devices, denoted as nodes, which
always fulfill the whole network requirements. Due to this fact, can sense the environment and transmit the sensed data
distributed methods can cause network inefficiency in different gathered from the monitored field wirelessly. The sensed data
performance aspects, e.g., QoS and load balancing. is transmitted, possibly via multiple hops, to a BS node
Re-clustering methods: Re-clustering refers to any action that can process it locally or forward it through a gateway
related to re-selecting CHs or re-organizing existing clusters. to other networks, e.g., the Internet. There exist different
CH selection methods are mostly designed to react against types of WSNs, such as Terrestrial WSNs [39], Underground
any unforeseeable circumstances by re-selecting or replacing WSNs [40], Underwater WSNs [41], Multimedia WSNs [42],
CHs dynamically with more appropriate nodes. Moreover, and Mobile WSNs and Wireless Sensor and Actuator Net-
not every node can serve as CH for a long period of time works (WSANs) [43].

WSNs are often considered as infrastructure-less resource- between the sources and the sink. To apply existing WSN
constrained networks [44] such that the nodes should clustering techniques to IoT, such techniques would be more
cooperate to establish a network, and gather and transfer the advantageous because in IoT networks, services can run inside
data, calling for efficient topology management solutions. the network on intermediary nodes, e.g., hierarchical data
Clustering is recognized as a popular technique for this processing in Fog [49]. In addition, the number of gateways
purpose, e.g., to improve efficiency of routing methods and may be more than one node in IoT, implying the need for inter-
reduce energy consumption due to wireless data transmission cluster routing to access the most efficient gateway in terms
using hierarchical data fusion [17]. From about 2000 to 2019, of, e.g., bandwidth and delay. It should be noted that we do not
tens of WSN clustering techniques have been proposed. In consider intra-cluster routing as a parameter in applicability of
this section, we study all significant clustering methods by WSN clustering techniques to IoT as it is used in special cases
reviewing around 125 papers. Unlike other survey papers, such as in blind spaces or for big-sized clusters, in addition
we do not focus on the design details of clustering methods to the fact that it may also degrade QoS.
like algorithm complexity, methodology, etc. as it is not the Mobility: Based on their applications, IoT networks may con-
goal of our study. We review the most significant clustering tain more mobile nodes [50] than WSNs [51], e.g., wearable
techniques in WSNs based on their objectives and the specific devices and IoV. Clustering techniques that support mobility
properties of networks. Considering the former, the primary would be more appropriate for IoT, thus we include mobility as
goal of clustering in WSNs is retaining and/or improving a network property in reviewing WSN clustering techniques.
a pre-defined set of quality-driven objectives. A clustering Table I lists the studied WSN clustering techniques, sorted
technique is designed to support one or multiple objectives in chronological order. Each technique has been examined
simultaneously. The objectives include: based on the aforementioned network properties and ob-
jectives. For example, LEACH [12], proposed in the year
• Energy Consumption (E) • Load Balancing (L) 2000, neither supports heterogeneity, nor mobility. CHs in
• Fault Tolerance (F) • Reliability (R) this method are mainly in charge of data fusion and they
• Physical Layer Support (U) • Jitter (J) use direct inter-cluster routing to transmit data to BS. In
• Throughput (T) • Scalability (A) addition, the sole clustering objective of LEACH is to reduce
• Coverage (O) • Connectivity (C) energy consumption (E). Some techniques support objectives
• Mobility Management (M) • Stability (B) that are not among the main 17 objectives. For them, the
• Packet Delivery Ratio (Y) • Security (S) table includes a short description of the targeted objectives,
• Mutli-Sink Support (K) • Delay (D) e.g., EACLE [52] reduces packet collision, in addition to
• Number of Packets Received by BS (P) reducing energy consumption (E). Moreover, for techniques
We also consider a set of network and clustering properties supporting heterogeneity and mobility, the table clarifies how
in our review that are crucial for the applicability of clustering these properties are supported. If there is no comment for a
techniques to IoT networks. These properties include: hetero- technique supporting heterogeneity, its heterogeneity aspect is
geneity, role of CH, inter-cluster routing, and mobility. Below, by default the energy resource. The last row of Table I shows
we explain the reasons for choosing these properties. the statistics of the studied techniques. As shown, most of
Heterogeneity: Contrary to WSNs which are often homoge- the literature focuses on improving the energy consumption,
neous [19], IoT networks are known to be heterogeneous [3]. in addition to load balancing and scalability. Moreover, it
Thus, network management in IoT needs to deal with nodes indicates that many techniques support multi-hop inter-cluster
heterogeneity as it affects the performance of the network routing. With respect to the network properties, there are
from different aspects, e.g., stability, connectivity, and QoS not many clustering techniques supporting heterogeneity and
[45], [46]. Clustering is a network management technique mobility. As these are two most important parameters in IoT
that can address heterogeneity issues in the network. Thus, clustering, more research effort in this area is needed.
WSN clustering techniques that support heterogeneity are To the best of our knowledge, this table compiles the most
more compatible with the nature of IoT networks. well-known clustering techniques in WSNs, their properties
Role of CH: Generally, CHs in WSNs are not in charge of and objectives. It allows filtering the techniques based on
performing complicated tasks, thus the parameters to select the desired properties and objectives for clustering. Regarding
CHs are often limited to energy and data forwarding resources. their applicability to IoT networks, it allows finding the
However, some clustering techniques utilize CHs for data suitable techniques based on the chosen network properties
fusion, in which the CH selection criterion is the computation and clustering objectives. In the next section, we study those
power of nodes [15]. Such techniques can be suitable for types of techniques, in the context of IoT, according to the
clustering in IoT networks as CHs in IoT can be used to aforementioned clustering objectives, and discuss existing IoT
execute computationally demanding tasks [47], [48], e.g., clustering solutions addressing each objective.
edge-level image processing in smart cities. In Table II, objectives of clustering are reviewed based
Inter-cluster routing: In some WSN clustering techniques, on their correlation. The white cells on the diameter of the
data transmission to the sink is performed through a 2-hop table show the number of techniques that support an objective
communication model (i.e., node to CH and CH to sink) over on x/y-axis, e.g., 44 techniques support load balancing (L).
a multi-point to point network infrastructure. However, many Other numbers, in each column, show the percentage of the
other clustering techniques allow data routing with more hops techniques that support an objective on the y-axis.

Mobility
Heterogeneity

Inter-cluster
Routing
Num Clustering
Technique
Year
Role
of
CH
Objectives

Fusion
Relay
E L R D J T Y P C O F B A M S U K
1 LCA[53] 1981 No X DM Yes X
X X
2 [54] 1997 No X X DM Yes support multimedia applications
X X
network management, fault tolerance (processor failure),
3 CLUBS[55] 1998 No X M Yes resource sharing (distributed computing)
CHs are leaders to allocate tasks
4 LEACH[12] 2000 No X D No X
X
5 TEEN[14] 2000 No X M No support time-critical data sensing applications
6 HCC[56] 2001 No X X M Yes X X
X X
7 MBC[57] 2001 No X X M Yes support group mobility
8 PEGASIS[58] 2002 No X D No X X
X X X X
9 RCC[59] 2002 No X D Yes support high speed mobility
X
10 APTEEN[60] 2002 No X M No support query based WSNs
X X X
11 GS3[61] 2003 Yes X D Yes support high degree of mobility and dynamicity
12 EEHC[62] 2003 No X DM No X X
13 HEED[13] 2004 No X DM No X X X
14 ACE[63] 2004 No X D Yes X X X
15 FLOC[20] 2004 No X D Yes X X
16 SEP[64] 2004 Yes X D No X X X
X X
17 DWEHC[65] 2005 No X D No possible inter-cluster communication
X X X
18 [66] 2005 No X X DM No fault tolerance (CH failure)
X X
19 UCS[67] 2005 Yes X M No CHs are mobile
X X X X X
20 TTDD[68] 2005 No X X M No support multiple mobile sinks
support query based WSNs, directed diffusion
21 BCDCP[69] 2005 No X M No X X
22 EECS[70] 2005 No X X D No X X X
X X X X X
23 MOCA[71] 2006 No X M No reduce processing and message complexity, CH failure tolerance
24 [72] 2006 No X X M No X X X
25 DWCA[73] 2006 No X M Yes X X
26 CCS-PEGASIS[74] 2007 No X M No X X
27 EEDC[75] 2007 No X M No X X X
X X
28 EcoMapS[76] 2007 No D No task scheduling, parallel processing, task mapping in CHs
X X
29 EACLE[52] 2007 No X M No reduce packet collision
X X X X
30 [77] 2008 Yes X X M No reduce transmission cost
31 ICDB[78] 2008 No X M No X X
32 [79] 2008 No X X M No X
X
optimize message ratio to discover services
33 C4SD[80] 2008 Yes M Yes service discovery protocol based on clustering
CHs are directory of services
34 EEMC[81] 2008 No X M No X X
35 MRPUC[82] 2008 No X M No X X
X X
36 MHP[34] 2008 Yes X M No support query based data gathering, CHs are predetermined
37 EEDUC[83] 2008 No X M No X X
X
38 BSIDR[84] 2008 Yes X D No support computation & communication heterogeneity
data compression in CHs
39 PEBECS[85] 2009 No X M No X X X
40 UCR[86] 2009 No X M No X X
X X X X X X X
41 [87] 2009 Yes X M Yes support heterogeneity in transceivers, node type
sensor type and energy
42 KOCA[88] 2009 No X M No X X X
43 ACHTH-LEACH[89] 2010 No X M No X X
X X X
44 CBR-Mobile[90] 2009 No X Yes CHs are statiornary
CHs are responsible to do time scheduling for data aggregation
45 HGMR[91] 2010 No X M No X X X X X
X X X
46 PANEL[92] 2010 No X M No data aggregation, reliable data storage
support asynchronous applications
47 ARC[93] 2010 No X M No X X X X
X X X X
48 EDFCM[94] 2010 Yes X D No support computation resource heterogeneity
X
49 MMCRA[95] 2010 No X DM Yes support vehicular WSNs
X X
50 EAUCF[96] 2010 No X X M No X X X
X X
51 [97] 2011 No X D No track the targets and recover lost targets
52 Spatial-clustering [98] 2011 No X M No data aggregation
53 LUCA[99] 2011 No X M No X X
54 EEEPSC[100] 2011 No X D No X X
55 EADUC[101] 2011 Yes X M No X X
56 EDUC[102] 2011 Yes X X DM No X X X
57 ACT[103] 2011 No X M No X X
X X X
58 CCR[104] 2011 Yes X M Yes support WSAN, support delay sensitive applications
support heterogeneity in energy and transmission rate
59 MBC[105] 2011 No X M Yes X X
60 EBCAG[106] 2012 No X M No X X
61 LEACH-SWDN [107] 2012 No X D No X X X
62 ECPF[108] 2012 No X M No X X
63 LEACH-ERE[109] 2012 No X D No X X X
64 DECSA[110] 2012 No X M No X

Mobility
Heterogeneity

Inter-cluster
Routing
Num Clustering
Technique
Year
Role
of
CH
Objectives

Fusion
Relay
E L R D J T Y P C O F B A M S U K
X X X
65 EPCR[111] 2012 No X M Yes fault tolerance (packet loss recovery)
66 E-LEACH[112] 2012 No X M No X X
67 IFUC[113] 2013 No X M No X X
68 COCA[114] 2013 No X M No X X
X X X
69 ACDA[115] 2013 No X M No support directional antenna
70 DSBCA[116] 2013 No X D No X X X
71 LCM[117] 2013 No X X M No X X X
72 MODLEACH[118] 2013 No X M No X X
73 Q-LEACH[119] 2013 No X D No X X X X
74 EDDEEC[120] 2013 Yes X D No X X
X X X
75 PASCCC[121] 2014 Yes X D Yes optimize packets received by BS and CHs, congestion control
support application priority models
76 [122] 2015 No X M No X X X X
X
77 DECA[123] 2014 Yes X D No CHs are predetermined
78 FUCP[124] 2015 No X M No X X X
79 FAMACROW[124] 2015 No X M No X X X
80 E-OEERP[125] 2015 No X M No X X X X
X
81 SCCH[126] 2015 No X D No reduce data loss, fault tolerance (CH failure)
X X
82 TCBDGA[127] 2015 No X M No support mobile sinks, support heterogeneous sensory data
X X
83 DFCR[128] 2015 No X M No fault tolerance (CH failure)
84 PEECR[129] 2015 No X M No X X
85 BTCWSN[130] 2015 No X D No X X X
X
86 [131] 2016 No X M No data compression, support hexagon-clustered networks
X X
87 PKF-ST [132] 2016 No X D No support temporal and spatial data aggregation
88 Improved EADUC[133] 2016 No X M No X X
89 FBUC[134] 2016 No X M No X X
90 DUCF[135] 2016 No X M No X X
91 GFTCRA [136] 2016 No X M No X X X
92 nCRO-UCRA[137] 2016 No X M No X X X
X X X
93 UMBIC[138] 2016 Yes X M No fault tolerance (CH failure)
94 EDDUCA[139] 2016 No X M No X X
95 UCCGRA[140] 2016 No X M No X X
X X X
96 MHCDA[141] 2016 Yes X M Yes (support mobile sinks)
X X X
97 [142] 2016 No X M No support mobile sinks
X X X
98 NDCMC[143] 2016 No X D No CHs are mobile as mule
99 DECUC [144] 2017 No X M No X X X
100 PSO-ECHS[145] 2017 No X D No X X
101 GEEC[146] 2017 No X D No X X
X X X X
102 K-SCC[147] 2017 No X M No improve security by providing authentication
103 HDMC[148] 2017 No X M No X X X
104 GASONeC[149] 2017 No X D No X X
105 LEACH-SF[150] 2017 No X D No X X X
X X X
106 EPMS[151] 2017 No X D No support mobile sinks
107 EA-CRP[152] 2017 No X X M No X X X
108 nCRO-ECA [153] 2017 No X M No X X
X X X
109 EECDRA [154] 2017 No X M No support multiple mobile sinks
X X X X X
110 DCMDC [155] 2017 No X M Yes reduce mobility management cost
111 HSCA[156] 2018 Yes X M No X X X
112 FUCA[157] 2018 No X D No X X
113 COARP[158] 2018 No X D No X X
114 H-kdtree[159] 2018 Yes X M No X X X X
115 HiTSeC[160] 2018 Yes X D Yes X X
116 ENEFC[161] 2018 No X M No X X X
117 DHRP[162] 2018 No X M No X X
X
118 CMS2TO[163] 2018 No X M No support mobile sinks
X
119 [164] 2018 No X M No support mobile sink mule
X X X X X X X
120 FEEC-IIR[165] 2018 Yes X M No balance channel load, improve buffer occupancy
X
121 VCHFBG [166] 2019 Yes X D No fault tolerance (CH failure)
X X X X
122 EECSR[167] 2019 No X M No data compression in CHs
X X
123 BPA-CRP [168] 2019 No X X M No data compression in CHs
124 NEP-ECGD[169] 2019 No X M No X X
125 MLBC[170] 2019 No X M No X X X X X X X X
Yes=22 Direct=34 Yes=21
Results N/A 28 108 113 44 3 21 3 15 20 20 8 8 13 8 32 2 4 3 2
No=103 M-hop=83 No=104
TABLE I: Comparing the existing well-known WSN clustering techniques

For example, out of 44 papers on L, 97,7% of them support rapidly being introduced for massive scale applications such
E as other objective, while only 4,5% of them support R. as smart cities and vehicular networks, which are often built
The heatmap highlights which pairs of objectives in existing over ad-hoc infrastructures, as mentioned above.
clustering techniques have received more attention by the Contrary to the centralized models with the capability of
researcher. In addition, the statistical analysis of Table I shows direct connectivity to the server or the Cloud, in the decen-
that the number of papers that support 1, 2, 3, 4, 5, 6, 7 and tralized model, topology management is a critical requirement
8 objectives simultaneously are 15, 53, 40, 10, 4, 0, 2, and for efficient and scalable management of IoT networks. Being
1, respectively. This indicates that clustering techniques are the primary solution for topology management, clustering can
normally used to support multiple objectives simultaneously, be applied to decentralized IoT networks in order to improve
where the majority of them support two objectives. resource usage, QoS, and load balancing, to name a few.
The other useful insight is to find out which pairs of Therefore, the survey is carried out for this type of IoT
(objective, networkproperty) have been considered more in networks.
the design of clustering techniques. For this purpose, out of
the 125 techniques in Table I, we counted the number of tech-
niques that support a given (objective, networkproperty) as Server
Cloud
Cloud

shown in Table III, e.g., 92 of the techniques with energy

efficiency (E) as an objective can not support heterogeneity. Server Cloud

Server Server

As shown in the table, most of the literature does not support

heterogeneity. Considering the role of CH, the table shows that With On-Premise Servers With Cloud Platforms

data fusion has been the common role of CH in achieving most Centeralised IoT Networking Model Decentralised IoT Networking Model

of the objectives such as E, L, Y and P. Besides the obvious

Fig. 2. General IoT networking models
reason of fusing data for energy consumption reduction or load
balancing, for Y and P sending fused data instead of individual
data items will lead to a lower rate of data loss in the network.
Considering routing, multi-hop methods are quite common for A. Clustering in WSNs vs. in IoT: Technical Differences
achieving E, in addition to improving network coverage (O) Existing solutions and technologies for WSN clustering
and connectivity (C). In addition, it can be seen that there have the potential to be applied to IoT networks. However,
is a limited amount of literature that supports mobility when clustering in IoT introduces new challenges that are discussed
the objective is energy consumption improvement or load below. The summary of the technical differences is presented
balancing (13 and 2 respectively). This implys that mobility in Table IV.
support can disrupt achieving these popular objectives. The 1) Heterogeneity: A fundamental challenge in IoT is sup-
above statistical evaluation shows that designing clustering porting heterogeneity. Compared to WSNs, IoT network man-
techniques for IoT networks needs more attention as support- agement techniques need to consider heterogeneity as a con-
ing mobility and heterogeneity are two key requirements for tingent characteristic of the network. In IoT, in addition to
IoT clustering. WSN nodes, other types of smart devices (e.g., smartphones
and cameras) may be used in one application scenario [172].
Even compared to typical WSNs, WSNs-assisted IoT [173]
IV. M IGRATING FROM WSN S TO I OT are essentially different in terms of heterogeneity [174]. As
WSNs essentially connect sensor nodes to the Internet shown in Table I, most WSN clustering techniques are based
through a gateway, contrary to smart devices in IoT that on homogeneous networks. Those supporting WSN hetero-
are often able to connect directly to the Internet. Figure 2 geneity focus mostly on energy resources [175], but in IoT,
shows different types of IoT networking models. In centralized heterogeneity is somewhat different with respect to clustering:
networking solutions, nodes connect to on-premise servers • Energy: Heterogeneous WSN clustering techniques
or cloud platforms to process data. By integration IoT with (HeWCTs) mostly support heterogeneous networks that
cloud computing platforms, smart devices connect to virtually have predetermined initial energy, though all nodes have not
infinite resources to store and process data centrally. In the the same initial energy. However, they have not considered
decentralized model, Machine-to-Machine (M2M), D2D, and the networks with high diversity in node energy levels,
ad-hoc networks allow IoT networks to connect nodes directly. or nodes with rechargeable energy resources. The fact is
The main difference between D2D and M2M is the physical that, In IoT networks, nodes may have different energy
proximity of devices. In D2D a pair of devices must be in capacities, ranging from nodes with unlimited energy
close physical proximity, but in M2M they can be distant, but resources, e.g., plugged PC or cameras, to nodes with
they are able to communicate directly without the need for a flexible energy resources, e.g., outdoor sensors equipped
central node (e.g., servers). Communication technologies such with photovoltaic solar panels.
as wide area networks and cellular networks have also further • Computing power: Similar to heterogeneity in energy, in
realized the decentralized networking model in IoT, such as some cases, clustering techniques in HeWCTs support a few
LoRA [171] and NB-IoT [171], respectively. To conclude, predefined levels of computing power like in [94]. However,
many IoT networks resemble ad-hoc networks with respect in heterogeneous IoT (HetIoT) [3], the smart devices may
to topology and the communication model. Moreover, IoT is have divers computing power levels. Such diversity can

TABLE II
C OMPARING CLUSTERING OBJECTIVES AND THEIR CORRELATIONS IN THE REVIEWED CLUSTERING TECHNIQUES

Objectives
E L R D J T Y P C O F B A M S U K

E 113 97,7 % 100,0 % 95,2 % 100,0 % 100,0 % 90,0 % 100,0 % 62,5 % 62,5 % 57,1 % 87,5 % 75,0 % 0,0 % 100,0 % 66,7 % 100,0 %

L 38,1 % 44 66,7 % 9,5 % 25,0 % 40,0 % 20,0 % 25,0 % 12,5 % 37,5 % 35,7 % 37,5 % 25,0 % 0,0 % 0,0 % 0,0 % 50,0 %

R 2,7 % 4,5 % 3 4,8 % 50,0 % 13,3 % 10,0 % 0,0 % 0,0 % 12,5 % 0,0 % 12,5 % 6,3 % 0,0 % 0,0 % 0,0 % 0,0 %

D 17,7 % 4,5 % 33,3 % 21 50,0 % 13,3 % 10,0 % 0,0 % 0,0 % 12,5 % 0,0 % 12,5 % 6,3 % 0,0 % 0,0 % 0,0 % 0,0 %

J 3,5 % 2,3 % 66,7 % 14,3 % 4 40,0 % 35,0 % 5,0 % 12,5 % 0,0 % 7,1 % 25,0 % 28,1 % 0,0 % 33,3 % 66,7 % 50,0 %

T 13,3 % 13,6 % 66,7 % 28,6 % 75,0 % 15 5,0 % 0,0 % 0,0 % 0,0 % 7,1 % 0,0 % 6,3 % 0,0 % 0,0 % 33,3 % 0,0 %

Y 15,9 % 9,1 % 66,7 % 33,3 % 25,0 % 33,3 % 20 10,0 % 0,0 % 12,5 % 7,1 % 25,0 % 12,5 % 0,0 % 33,3 % 33,3 % 0,0 %
Objectives

P 17,7 % 11,4 % 0,0 % 4,8 % 0,0 % 13,3 % 15,0 % 20 12,5 % 12,5 % 14,3 % 37,5 % 12,5 % 0,0 % 0,0 % 0,0 % 0,0 %

C 4,4 % 2,3 % 0,0 % 4,8 % 0,0 % 0,0 % 5,0 % 0,0 % 8 0,0 % 0,0 % 0,0 % 3,1 % 0,0 % 0,0 % 33,3 % 0,0 %

O 4,4 % 6,8 % 33,3 % 0,0 % 0,0 % 6,7 % 5,0 % 0,0 % 50,0 % 8 0,0 % 12,5 % 12,5 % 0,0 % 0,0 % 0,0 % 50,0 %

F 7,1 % 11,4 % 0,0 % 4,8 % 25,0 % 6,7 % 10,0 % 0,0 % 12,5 % 12,5 % 14 12,5 % 12,5 % 0,0 % 0,0 % 0,0 % 0,0 %

B 6,2 % 6,8 % 33,3 % 9,5 % 0,0 % 13,3 % 15,0 % 0,0 % 12,5 % 12,5 % 0,0 % 8 18,8 % 50,0 % 0,0 % 66,7 % 0,0 %

A 21,2 % 18,2 % 66,7 % 42,9 % 50,0 % 26,7 % 20,0 % 5,0 % 50,0 % 50,0 % 42,9 % 25,0 % 32 0,0 % 0,0 % 0,0 % 0,0 %

M 0,0 % 0,0 % 0,0 % 0,0 % 0,0 % 0,0 % 0,0 % 0,0 % 0,0 % 0,0 % 7,1 % 0,0 % 6,3 % 2 33,3 % 66,7 % 50,0 %

S 2,7 % 0,0 % 0,0 % 4,8 % 0,0 % 6,7 % 0,0 % 0,0 % 0,0 % 0,0 % 0,0 % 0,0 % 3,1 % 0,0 % 3 0,0 % 0,0 %

U 1,8 % 0,0 % 0,0 % 9,5 % 6,7 % 6,7 % 0,0 % 12,5 % 0,0 % 0,0 % 14,3 % 0,0 % 6,3 % 0,0 % 0,0 % 3 0,0 %

K 1,8 % 2,3 % 0,0 % 4,8 % 0,0 % 0,0 % 0,0 % 0,0 % 12,5 % 0,0 % 0,0 % 0,0 % 3,1 % 0,0 % 0,0 % 0,0 % 2

TABLE III discoverable by each other in physical proximity [87], and

T HE CORRELATIONS BETWEEN OBJECTIVES AND PROPERTIES OF THE ii) in inter- and intra-cluster routing there is a trade-off in
NETWORK IN THE REVIEWED CLUSTERING TECHNIQUES
choosing the best network interface with respect to, e.g.,
Support
Heterogeneity
Role of CH Routing Mobility cost, energy, transmission range, etc. [87].
Yes No Fusion Relay Both Direct Multi-hop Both Yes No
E (113) 21 92 101 20 11 30 77 5 13 100 2) Mobility: Supporting mobility of nodes in IoT net-
L (44) 6 38 43 7 6 6 34 3 2 42
R (3) 2 1 3 0 0 0 3 0 3 0
works has been an important issue, e.g., in MANET-IoT
D (21)
J (4)
4
4
17
0
17
3
6
2
3
1
4
0
17
4
0
0
5
1
16
3
networks [178]. The authors in [179] review mobility manage-
T (15)
Y (20)
6
5
9
15
12
18
3
2
0
4
2
7
12
13
0
0
4
8
11
12
ment challenges in IoT and show how mobility can make IoT
P (20) 2 18 20 1 1 7 12 0 0 20 networks unstable. Mobility in WSNs has often been proposed
C (8) 1 7 5 5 2 1 6 1 2 6
O (8) 1 7 6 3 1 0 8 0 0 8 for rather simple scenarios such as mobile sinks, unlike
F (14) 3 11 9 6 2 5 7 2 7 7
B (8) 2 6 7 2 1 4 4 0 1 7 highly dynamic IoT applications such as smart cities [5], and
A (32) 4 28 25 12 5 12 17 3 8 24
M (2) 1 1 1 2 1 1 1 0 2 0 V2V [180]. Additionally, mobile nodes not only change their
S (3) 1 2 3 0 0 2 1 0 1 2
U (3) 1 2 2 2 1 0 2 1 2 1 locations in the IoT network, but also may switch between
K (2) 0 2 1 2 1 0 2 0 0 2
different networks which is related to handover mechanisms
in IoT networks [181][182]. These issues make designing
clustering techniques in IoT networks challenging [183].
make clustering more complicated when combined with 3) Device Identity: The other issue is that the IoT network
other clustering criteria for, e.g., selecting CH and assigning includes not only ID-based devices but also IP-enabled de-
tasks. Nodes with high computing power are a potential vices [184]. This implies that clustering techniques should
candidate to gather, fuse and process data. be designed in such a way that they can support clustering
• Storage: Having smart devices with different amounts of both these types of nodes. For instance, IP-enabled devices
data storage in the network is another aspect of hetero- can basically communicate with ID-based devices, while the
geneity that should be taken into account in IoT network opposite direction of communication may not be straightfor-
clustering, in particular because of the big data applications ward [185]. Therefore, clustering techniques can adopt two
of IoT [176]. To the best of our knowledge, there is only different approaches for clustering: clustering devices with
one clustering technique [177] that supports heterogeneity the same network type, or clustering both IP-enabled and ID-
in storage. enabled devices.
• Network interface controller (NIC): In IoT networks, nodes 4) Device-to-Device (D2D) Communication: D2D commu-
may be equipped with different network interfaces such nication involves direct short-range communication between
as Wi-Fi, Zigbee, XBee, Bluetooth, BLE, and LTE. In IoT devices without the support of network infrastructures,
HeWCTs, researchers consider mainly low-power commu- such as BSs or access points [186]. Many IoT applications and
nication technologies such as Zigbee and 802.15.4 [111]. services are realized through D2D communication networks
Considering clustering, network nodes with different types such as Zigbee, Bluetooth Low Energy (BLE), Radio Fre-
of interfaces can pose two challenges: i) they may not be quency Identification (RFID) and Near Field Communication

TABLE IV Fig. 3. All conventional clustering objectives and exiting solutions

C OMPARING THE TECHNICAL D IFFERENCES OF WSN S AND I OT TO USE CH Duty
C LUSTERING Rotation
Energy Hierarchical
Consumption Clustering
Technical Differences WSNs IoT
Energy Occasionally Common Balanced
Clusters
Computing Power Selodm Common
Heterogeneity Densified
Storage Seldom Common
Clusters
NIC Mostly single NIC Common Multi-NIC
Balanced
Mobility Occasionally Very common Clusters
Communication Type Mostly ad-hoc Possibly ad-hoc
Load Balancing Cluster-Based
Deployment model Single application Multi-application Congestion
Applications
Design model data-centric Service-based Control
Mechanisms
Balanced Energy
Consumption

(NFC). 6LoWPAN is an advanced D2D enabling technology Cluster-based

Failure Detection
(the scaled-down version of the IPv6 standard for LoW- Fault Tolerance Spare CHs

PANs) [187]. Given the wide usage scope of 6LoWPAN, IoT Re-Clustering

clustering can take advantage of D2D communication through Reliability

Network Stability
Improvement
6LoWPAN and exploit the flexibility and diversity of D2D Reducing
Number of Hops
communications to provide more efficient and application-
Delay Load Balancing
driven clustering techniques.
Routing
5) Applications: Although clustering is targeted toward the Optimization
Quality of
network layer requirements, there are challenges, raised by Service Network Stability
Improvement
IoT applications, that should be considered in IoT clustering: Throughput Stable Number
of CHs
• Deployment model: Contrary to WSNs which usually host a Data Aggregation
single application [188], IoT network infrastructures may be Conventional Tree-based
Clustering Clustering
offered to host multiple applications with their own quality Objectives Jitter
and Solutions Splitting
requirements. As most WSN-based clustering techniques Data Streams

rely on the network layer parameters for clustering, e.g., en- Distributed
Management
ergy, supporting application-level requirements by clustering Scalability
Reducing
Communication
is crucial, in particular when applications’ requirements are Overhead
Network
in conflict with each other, e.g., QoS and resource demands. Management Stability Avoiding
Node Death
• Design models: Unlike WSNs which are mostly data-centric, Inter-Cluster
Mobility
Handover
IoT applications are often service-based [1], in which ser- Management
Mechanisms
vices can run on various devices and platforms, from local Improving Radio Range
Network Coverage Management
computing platforms to the Cloud. To further facilitate IoT Connectivity Improving Nodes Establish Stable
services development and improve their efficiency, new Connectivity Connection

computing technologies such as Edge/Fog computing [189], Network

Reliability
[2] have been recently emerged. With respect to the network Improve Packet
delivery Ratio
Improvement

Reducing
infrastructure, Fog/Edge can possibly change the essence Data Delivery Data Loss

of the network topology from multi-point to point (like Improvement

Optimizing
Network
Lifetime
in WSNs) to multi-point to multi-point due to distribution number
of Packets
Improvement

of IoT services across the network. This will pose new received by BS Data Aggregation

Distributed Trust-
challenges in designing IoT clustering routing algorithms Security
based Models

and CH selection as they have to adhere to the target service Physical

Distributed
Resource
Layer Support
distribution model. Management

MP2MP
Multi-Sink
Clustering-
Support
Routing
V. F ROM WSN S TO I OT: C ONVENTIONAL C LUSTERING
O BJECTIVES
A clustering technique can support a single objective, or clustering, we split existing solutions to pure WSN clustering
multiple objectives simultaneously. The IoT networks that techniques, and IoT-related clustering techniques (i.e., WSN-
resemble ad-hoc networks share several commonalities with assisted IoT and pure IoT). The former category is compiled
WSNs regarding conventional quality-driven objectives. The as a list in Section III, while the prominent works in the latter
nature of ad-hoc IoT introduces new challenges in achieving category are discussed per objective in this section.
clustering objectives in IoT networks. In this section, we study Before discussing the objectives in detail, in Figure 3 we
each objective in detail. In particular, for each objective and illustrate an overview of of clustering objectives and existing
the associated general clustering solutions, we discuss existing solutions. Figure 4 also shows the challenges in achieving
IoT clustering techniques, and also design concerns raising those objectives in IoT through clustering techniques. These
due to applying clustering to IoT for that specific objective. It figures together provide a brief overview of what we present
should be noted that, due to the high volume of the literature in in this section.

Fig. 4. IoT-Specific concerns in achieving clustering objectives (HetIoT), more replacement candidates are available when
Application-
Aware Energy
looking for a node for data forwarding (aggregated) or
Energy Saving processing energy-intensive tasks. In [196], the authors use
Consumption Complicated
Routing the k-means algorithm in IoT networks based on the residual
Techniques
energy to select new CHs. EnergIoT [197] is a hierarchi-
Dispersed
Destinations cal clustering method to assign duty cycle (i.e., switching
Cross Network
Communication
between sleeping and active modes) ratios to devices for
Load Balancing
Dynamicity improving energy consumption. However, in IoT, CHs are
Two Way not only in charge of data aggregation and forwarding but
Communication
Faulty
also responsible to perform more complicated tasks such as
Fault Tolerance
Connections
hosting services, data compression, network management,
Faulty Services
etc. This makes CH selection more complicated as the
Quality of
Quality of
Experience residual energy cannot be the only parameter for selecting
Service Application
Requirements
the new CH like in WSNs. Therefore, CH selection becomes
IoT-Specific
Supporting QoS a multi-objective problem with new parameters such as
in Scalable
Clustering
Concerns
Network Networks location and density of the nodes in Ultra Dense Networks
Management
Highly Dynamic (UDN) and computation power, network interfaces, and
IoT Applications

Heterogeneous
storage in HetIoT. In [198], the authors use the residual
Network
Network
Interfaces
energy of IoT nodes together with the density and location
Connectivity Issues in of alive nodes as the parameters to rotate duty of CH.
Network
Infrastructures • Hierarchical clustering: This is a method that organizes
Data Delivery Unreliable
Network
the network in a number of layers. Each layer can com-
Improvement
Infrastructures municate with neighbor layers, which saves more energy
Highly
Dynamic IoT compared to transferring data directly to far destinations.
Security
Resource- In hierarchical clustering techniques, middle nodes in upper
Constrained
IoT Devices layers (closer to the BS) consume more energy because of
Physical
Layer Support
Ultra Dense
Networks
forwarding massive volume of data from the lower layers to
Multi-Sink Best Gateway the upper layers. One popular method to balance the load
Support Selection
is establishing hierarchical unbalanced clusters. Clusters
in the lower layers have more members compared to the
upper layers [152]. Since receiving data consumes a fixed
A. Energy Consumption Improvement
amount of energy, CHs of lower layers can have more
As the most important objective of clustering, clustering members but do not need to forward/aggregate many packets
techniques can balance energy consumption and improve the from the other layers; therefore unbalanced layers/clusters
network lifetime. Since data communication is the prominent would be an appropriate option to balance energy con-
source of energy consumption in low-power networks [37], sumption. In IoT networks, hierarchical clustering has been
clustering can mitigate it through inter/intra-cluster routing by exploited to balance energy consumption. In [199], Tang
middle nodes, as well as data aggregation in CHs. et al. introduce a hierarchical clustering index tree based
Existing solutions and techniques include: on EnerGy-eFficiency Hierarchical Clustering index tree
• CH duty rotation: In clustering techniques, CHs consume a (ECH-tree) that organizes regions, called grid cells in IoT
large amount of energy compared to cluster members [190]. networks. The method ensures that the upper level grid cells
As a multi-objective selection technique, CH selection is have less dead spaces as compared to the lower levels.
a complicated task which can be performed using Fuzzy Using this technique, the energy consumed to collect and
logic [191] [192], AI methods [193], and heuristic meth- forward the data is reduced as compared with traditional
ods [194] [195]. CHs as a central node should support index tree based techniques. In [200], the authors introduce
various objectives of networks such as QoS, reliability, and hierarchical unbalanced layers for industrial IoT networks
balancing energy consumption. CH rotation—rotating CH to balance the energy consumption as the layers closer to
duty—is a common solution to achieve such objectives. BS contain smaller clusters.
One key criterion for triggering CH rotation is the threshold • Balanced clusters: Although unbalanced clustering (through
for energy consumption or the remaining energy of nodes. hierarchical clustering) is a solution for balancing energy
In homogeneous networks, CH duty rotating is a more consumption, it would be challenging in ad-hoc networks
complicated task because all nodes have the same energy in which nodes are mobile or not uniformly distributed.
resources [36]. On the contrary, in heterogeneous networks, Unbalanced clusters in such networks can lead to more
CH re-selection would be easier thanks to the presence of energy consumption due to gathering data from many mem-
energy-rich nodes limiting the number of CH node candi- ber nodes. In such cases, to have fair energy consumption,
dates. Being highly heterogeneous compared to WSNs [3], clusters should be balanced based on the number of nodes
CH rotation is more natural in IoT, keeping the performance and the volume of generated data, not the size of the areas
of the network at the expected level. In Heterogeneous IoT covered by the clusters [201]. In IoT, balanced clusters have

been proposed to improve energy and resource consumption. which cause network congestion. Generally, load balancing
In [200], Zhang et al. introduce a method that partitions is a critical issue in IoT due to the massive scale, huge
industrial IoT networks into grid clusters in each layer to volume of data and mobility of IoT networks which can cause
balance energy consumption. The size of the grid clusters inefficiency, e.g., in [212] a technique is introduced to balance
are calculated by solving polynomials. In [202], the authors the load of the network to reduce interference among nodes1 .
introduce a service-aware clustering method to improve the Existing solutions and techniques include:
energy consumption. The proposed method uses a traffic • Densified clusters: means creating more clustering layers
engineering model to design the clustering technique and and establishing more clusters in each layer. Having more
provide profiles of services running on sensor nodes to CHs will allow to distribute the network load among more
improve the network lifetime. destinations (middle nodes). Moreover, having more layers
IoT-specific concerns. Whereas the above techniques have in hierarchical clustering can extend the number of hops
been somewhat considered for energy saving in IoT, there are as in each layer more nodes are available as the next hop
a number of IoT-specific issues that should be studied when for data transmission [214]. Considering layering in IoT,
applying clustering for energy saving in IoT, as discussed the authors in [215] propose a multi-path routing based on
below: clustering techniques, providing more routes for areas that
• Application-aware energy saving: From a different view- need to transfer higher volume of data.
point, unlike WSNs, IoT applications and their services • Balanced clusters: Balancing clusters in terms of volume
can affect the design of clustering techniques and energy of generated data would be a solution especially in direct
consumption improvement as the nodes may connect to communication. If nodes are homogeneous, the number
different hosts (local or remote) for service invocation. of nodes can be a parameter to balance the load among
While in many WSN network management techniques, clusters [216]. As an example, in [217] the authors introduce
data-aware energy consumption has been neglected [148], a Software Defined Network (SDN) model to establish a
IoT services serve as a new factor in energy consumption Particle Swarm Optimization (PSO) based clustering table
measurement [203], thus designing a clustering technique that is used by the SDN controller to balance the clusters,
cannot be performed only by considering the network layer resulting in load balancing. The authors in [218] integrate a
requirements [204] [205]. In [206], the authors introduce clustering technique with compressive sensing in IoT. They
a clustering method for multimedia IoT, based on the K- introduce the CSLB-CS method which is a new cluster size-
means algorithm. They establish a tree to divide multime- based load balancing to optimize compressive sensing in
dia sessions and connect clusters of nodes to support the IoT-based sensor networks. In heterogeneous networks like
requirements of multimedia applications. IoT, the packet rate per node can be a parameter to balance
• Routing techniques: The presence of different network in- clusters, but the packet rate is a challenging and resource-
terfaces in IoT makes the implementation of clustering- consuming parameter to predict and process on-the-fly as it
based routing techniques complicated—as a solution to re- needs network traffic classification algorithms [219].
duce energy consumption. Specifically, each node can have • Congestion control mechanisms: Some clustering techniques
multiple network interfaces with different energy consump- are aimed to address the congestion control problem in
tion in data transmission, connection speed, and cost. In the network, resulting in load balancing by dividing the
addition, as IoT networks generally contain more gateways network load among existing resources. In [220], the authors
due to their scale, data volume, and QoS requirements [5], introduce a clustering technique to solve the congestion
routing protocols should support point-to-point, point-to- control problem in M2M IoT networks, resulting in load
multipoint, multipoint-to-point and multipoint-to-multipoint balancing. They assume the IoT network is based on M2M
communications, making clustering-routing techniques more communication in which clustering provides a slot alloca-
complicated compared to WSNs. Routing Protocol for Low- tion mechanism based on Q-Learning to avoid congestion.
Power and Lossy Networks (RPL) as the standard routing In [221], the authors propose a clustering technique to
protocol has been introduced to solve the problem of IoT manage multimedia big data in mobile IoT. They propose a
routing [207], [208], but there are doubts that RPL is able to new approach in MANET-IoT based on a modified LEACH
satisfy requirements of IoT in real world applications [209]. clustering method. They also propose a congestion control
method by link utilization based on the link status.
B. Load Balancing • Balanced energy consumption: As data transmission is the
most important cause of energy depletion, fair distribution
Load balancing is the second most important conventional of energy consumption among nodes can indirectly improve
objective of clustering. Generally, clustering techniques use load balancing and vice versa. In [222], the authors in-
a type of divide and conquer method to transfer data from troduce a clustering technique that balances the network
nodes to BSs. A high load of data to transfer can result in load based on a cost function. LiMCA is a IoT clustering
unbalanced resource consumption and influence QoS support. technique, introduced in [223], to balance the energy con-
Load balancing can also solve the “Hot Spot problem” which sumption in CHs and members by training them about their
is a common problem in WSNs [210] [164] and IoT [211].
The Hot Spot problem is a situation in which some nodes in 1 In [213], a comprehensive review of load balancing techniques in IoT
the network transfer high volume of data compared to others including clustering-based techniques, is presented.

coarse-grain location. They analyze energy consumption by help to detect failures and recover the infrastructure quickly.
evaluating the lognormal shadowing channel, inter-cluster In [228], the authors introduce a routing method for IoT
and intra-cluster traffic, and the packet error rate. They networks based on the bio-inspired particle multi-swarm
consider that both CH and members of clusters have a optimization (PMSO) strategy which is able to detect faulty
predetermined location based on the stochastic deployment nodes quickly and replace routes with new ones. They use
scheme. super nodes serving as a CH for information gathering.
IoT-specific concerns. In order to use clustering for load • Spare nodes (CHs): To tolerate faulty nodes, some tech-
balancing in IoT, we face a number of challenges: niques use spare nodes to take over responsibilities and
• Dispersed destinations: Concurrent applications need to
avoid data loss. Spare CHs can be replaced by faulty CHs
support different rates of packet streams with various vol- as Hot Spots and solve the problem efficiently. In [166], a
ume of data. Moreover, by using new emerging technologies method is introduced to select failure-free CHs to replace
such as Fog and Edge, destinations of streams cannot faulty CHs in IoT. It forms a virtual CH that includes
be only BS(s) or gateways, but also other nodes in the three failure-free CHs based on the Flow-Bipartite Graph
network [224]. This transforms the communication model in Modeling and energy of all failure-free CHs. After detecting
IoT networks to a multipoint-to-multipoint model, making a CH failure by BS, BS will ask failure-free CHs to take
load balancing challenging [225]. over the responsibilities of the faulty CH.
• Cross network communication: Considering load balancing,
• Re-clustering: As spare nodes induce overhead in selecting
the network beyond the gateway can not be controlled. and keeping them updated, the simplest solution is re-
For example, to support low latency in communication, clustering. After detecting failures, clusters can be destroyed
packet streams may be re-routed to a more efficient gate- and re-established, recovering the network infrastructure to
way, leading to dynamic unbalancing of the network load. transmit data. In [229], the authors introduce a clustering
In [226], the authors introduce a new idea called Cognitive based routing technique for IoT networks which replaces
IoT gateways, which can decide which services should be faulty CHs with vice CHs. In case of faulty vice CH, re-
migrated from the Cloud to the IoT network and vice versa, clustering will be performed.
dynamically. Using Cognitive gateways, nodes can monitor IoT-specific concerns. In IoT, fault tolerance is a big
the network performance such as QoS and load balancing challenge because of being highly dynamic and heteroge-
and bring services from outside of the network to inside. neous and also unreliability of the IoT network infrastruc-
• Dynamicity: In IoT, consumers and providers of services ture [230] [231] [232]. Addressing IoT fault tolerance issues
may not be known in advance, appearing and disappearing by clustering introduces the following challenges:
over application lifetime, e.g., multimedia applications. Un- • Faulty connections: In IoT, faulty connections are com-
foreseeable service requests can make the IoT nodes more mon [233]. Reliable communication mediums, such as WiFi
unpredictable with respect to the data load [227]. As the and 4G, can reduce faulty connections, but technologies
load of the network is dynamic, load balancing should be such as cellular-D2D and BLE still suffer from commu-
monitored on a regular basis and optimized accordingly. nication problems [6]. Moreover, mobile IoT devices can
• Two-way communication: The most crucial difference be-
increase the probability of connection faults in IoT networks.
tween WSNs and IoT which affects the network load balanc- These issues can make the clusters and routes between nodes
ing is related to one-way and two-way communications. To unstable resulting in, e.g., data loss.
balance the network load, two-way communication should • Faulty services: Services are the other source of faults in
be considered in IoT and hops should allocate resources to IoT [234], [235]. They can disrupt the network function, in
forward data in both ways. particular in the case of clustered networks. For example,
one solution for handling faulty services is to run the
C. Fault Tolerance services in other nodes [236], but it can influence the
There are different causes for node failure in IoT and WSN efficiency of the clustered network as the routes should be
networks, such as battery depletion, or failure in hardware re-established, affecting load balancing [237].
components like transceiver and processor which can be dam-
aged by external factors. Moreover, connectivity failure can
happen by physical or environmental factors, mobile nodes, or D. Quality of Service
faulty nodes. To handle such types of failures, fault-tolerant QoS is a rather a broad topic, encompassing divers quality
clustering techniques have been introduced to replace faulty aspects of network services, such as packet loss, bit rate,
nodes with other nodes and keep the network stable [166]. throughput, transmission delay, availability, jitter, etc. Al-
Existing solutions and techniques include: though WSNs and IoT share many QoS concerns, the specific
• Cluster-based failure detection: The speed of failed node characteristics of IoT networks introduce new QoS require-
detection methods is a challenge because high latency in ments, such as guaranteeing QoS in simultaneously running
detecting failures can cause data loss especially in data- IoT applications, and higher delay-sensitivity in safety-critical
loss sensitive applications. In addition, connections should IoT applications, such as IoV [238]. Clustering techniques
be checked to determine any fault in connectivity which can contribute to QoS improvement in IoT systems through
can cause delay or data loss. In both cases, clustering can dividing nodes to groups and improving QoS for each group

in a distributed manner. For example, in [212] the authors in- streams to reduce dynamic end-to-end delay. The proposed
troduce a clustering based routing algorithm for IoT networks method analyzes the packet streams on-the-fly and changes
to reduce end-to-end delay. They reduce interference resulting the clusters and routes to reduce delay. In [185], Kumar et
in delay and balance the load among nodes in wireless mesh al. compare K-means, hierarchical clustering and fuzzy C-
IoT networks. In the following, we discuss existing clustering- means clustering techniques to group IoT nodes. They use
based methods that deal with QoS requirements. these techniques to design a response time-aware scheduling
1) Reliability: By reliability, we mean the rate of data model for IoT to improve QoS and show that K-mean is
successfully received by the destination node in the network. more efficient compared to others regarding the prioritized
Clustering can contribute to the reliability of the network by message delivery model.
improving the stability of the network. Below, we discuss • Routing optimization: inter-cluster and intra-cluster routing
existing solutions in this context. can highly influence delay by determining shortest-fastest
Existing solutions and techniques include: routes. As routing is not the main focus of clustering,
• Network stability improvement: Clustering can improve the different techniques are introduced to achieve concurrent
reliability of the IoT network by avoiding node death and routing in clustering techniques [17]. In [243], the authors
maintaining connectivity [239]. As an example, in [170], the establish clusters based on an improved K-Means algorithm.
authors propose a method to reduce inter-cluster commu- They split and merge clusters to restructure the network
nication cost to improve reliability in data transmission by resulting in a non-uniform clustering structure. Then, they
assigning more resources to data forwarding and less amount create a tree for data fusing based on unequal clusters to
of resources to the transmission of network management utilize time slot scheduling, resulting in delay reduction.
data. The proposed method establishes a tree determin- 3) Throughput: It refers to the amount of data that is
ing connections between two nodes based on a weighting successfully transferred through the whole network for a
method and detects more reliable and affordable routes for certain period of time.
data transmission to the BS. In [240], the authors propose Existing solutions and techniques include:
a clustering-based routing algorithm called REMI for RPL-
based IoT networks to provide faster multicast dissemination • Network stability improvement: Infrastructure-independent
of messages, resulting in reliability improvement. In [241], networks are not always stable because of node mobility,
the authors introduce a new clustering technique based node failure, unreliable connections, and node death due
on quantum particle swarm optimization (QPSO) with the to energy depletion, to name the most important ones.
improved non-dominated sorting genetic algorithm (NSGA- By having a more stable infrastructure, the network can
II). The clustering technique introduces a trade-off between have more reliable routes to transfer data which improves
QoS and energy consumption of Industrial IoT. throughput. Clustering serves as an efficient means to im-
prove stability by reducing the time for network recovery in
2) Delay: Optimizing the efficiency of clustering-routing unstable network circumstances. Some clustering techniques
techniques can reduce delay (i.e., end-to-end delay). Gener- (e.g., in [246]) increase the number of alive nodes as a
ally, there is a trade-off between energy consumption and solution to improve the stability of the network. In [247], the
delay [37]. authors introduce a clustering technique for IoT networks to
Existing solutions and techniques include: conserve energy during routing, resulting in less probability
• Reducing number of hops: The optimal solution is sending of node death and so improving stability. They improve
data directly from each node to BS, causing high energy stability of the network by reducing delay, packet drop and
consumption. Therefore, a challenge in designing routing energy consumption, resulting in better throughput.
methods is how to reduce delay and improve energy con- • Stable number of CHs: A certain number of CHs with stable
sumption at the same time. In [242], the authors introduce communication between CHs and BS(s) can improve the
a delay-aware clustering algorithm for data acquisition in throughput of the network and keep the network reliable.
heterogeneous WSN-based IoT networks. In the proposed Having a deterministic number of CHs, the rate of data
approach, the nodes with higher radio range act as relay delivery to the BS will be predictable and the network
nodes to reduce the number of hops [242], [243]. In [52], throughput will become stable accordingly. In [248], the au-
[55], two fixed hops are employed in intra-cluster routing thors improve the throughput of heterogeneous IoT networks
to reduce delay in clusters. by proposing a clustering technique that uses weighted
• Load balancing: Load balancing can optimize lengths of election probabilities to select a stable number of CHs
queues in middle nodes, resulting in less delay. Network among resource-rich nodes.
congestion is a result of an unbalanced load in the net- • Data aggregation: Data compression and aggregation are
work which can cause high latency. Unequal clusters can two methods to optimize the volume of data to transfer.
cause delay due to unbalanced data transmission in dif- Using these methods, the volume of data received by BS(s)
ferent areas of the network [164]. In addition, selecting would be less than the accumulative data generated in all
CHs/forwarders with low amount of resources can further nodes which can improve throughput and decrease energy
increase delay [244]. In [245], the authors introduce a tree- consumption. In [249], the authors introduce a method to
based WSN clustering method which can balance the load eliminate repetitive readings of generated packets during
of the network by reordering clusters based on available intra-cluster and inter-cluster routing to reduce the volume

of data to transfer. E. Network Management

This is an important challenge in IoT networks due to their
4) Jitter: Another parameter that can improve QoS is jitter
large scale of deployments, and heterogeneity and mobility
prevention. It is especially important for multimedia and real-
of nodes. In IoT applications, networks are more dynamic
time applications which should keep the connections stable
compared to WSNs, e.g., IoV and smart cities, making network
and live, and also control network congestion [250].
management difficult. To manage the IoT network, dividing
Existing solutions and techniques include: the network into geographical partitions and assigning the
responsibility of partition management to local nodes is the
• Tree-based clustering: Tree-based clustering structures can primary solution. As network partitioning is the main task of
balance and possibly predict the volume of data in each clustering, different aspects of network topology management
route which can reduce jitter. In [159], the authors intro- are satisfied by clustering techniques as listed below:
duce a hierarchical tree-based routing protocol for cluster 1) Scalability: Due to having many nodes such as sensors,
establishment. They introduce a reactive mechanism that wearables, user equipment, etc., IoT networks have to be
use the history of the network to form CHs and reduce scalable and connect thousands or millions of nodes.
jitter. In [251], the authors introduce an optimized clustering Existing solutions and techniques include:
communication protocol based on intelligent computing. • Distributed management: Centralized network management
The proposed clustering is a tree based technique improves solutions are not efficient in large-scale networks and impose
energy consumption, but also reduces jitter and delay based high overhead. The authors in [253] introduce a scalable
on a cost function for establishing the tree. MQTT model in clustered IoT networks. Each cluster has
• Splitting data streams: Ad-hoc networks use some nodes some nodes as MQTT message brokers. They are used as
as hop to transmit data where each hop can be shared a backend system to service MQTT clients and balance the
among different streams. Different stream rates, within a load of the network in massive scale IoT networks. The
hop, can affect each other by changing the priority of packets authors in [254] introduce a method to search IoT services
in the forwarding queue, leading to jitter creation. By relying on a hierarchical network of semantic gateways.
splitting packet streams in middle nodes or using different They use a recursive agglomerative clustering technique to
middle nodes for different streams, the effect of streams provide a routing table of semantic gateways. It establishes
on each other will be reduced, e.g., effect of a multimedia clusters based on characterizing dissimilarities between ser-
stream on a temperature sensor stream which can be delay, vices and distances. The authors in [199] propose a method
jitter, or low throughput. In [206], the authors introduce a to divide the regions where sensors are deployed into a grid
clustering technique for multimedia IoT networks to divide and use a clustering index tree to manage and establish these
the large multimedia sessions into simple sessions based on grid cells.
the network situation to reduce jitter. • Reducing communication overhead: Communication over-
head is another challenge in supporting scalability. In [130],
IoT-specific concerns. As discussed above, there are some
the authors design a clustering method that reduces com-
clustering-based solutions addressing QoS requirements in
munication overhead to support scalability. Neamatollahi et
IoT, however tackling the QoS issues of IoT systems by
al. [162] introduce a clustering method that supports scala-
clustering introduces the following challenges:
bility by applying an energy-efficient distributed algorithm
• Quality of Experience (QoE): QoE is a novel concept which for all levels of clusters to reduce communication overhead
should be supported by IoT networks. QoE is aimed to based on the number of nodes.
improve the QoE of IoT users by receiving online feedback 2) Stability: Stability is one of the management objectives
from them and the network infrastructure based on avoiding which can make the network infrastructure more reliable in
and solving network problems [252]. In other words, appli- terms of connectivity, QoS, and fault tolerance. Different
cations which interact with a human need to improve QoE, challenges can affect the stability of a MANET like mobility,
however supporting QoE can influence applications that do node failure, etc. The most important existing solution used
not interact with a human but have critical QoS require- by clustering techniques to provide stability is:
ments. Since no work exists on supporting QoE through • Avoiding node death: Clustering techniques reduce energy
clustering techniques in WSNs [19], it becomes challenging consumption of nodes to avoid node death and keep the
to address QoE in clustering techniques introduced for IoT. routes and the cluster more stable. ACHTH-LEACH [89]
• Applications requirements: Many IoT applications demand is a method which provides reliable routes and a more
for critical QoS requirements. In special cases like IoV stable network infrastructure to transfer data by selecting
that deal with safety, QoS must be supported and it is not resource-rich nodes as CHs. In this method, routes are
an optional parameter in improving the efficiency of the established for a long time and nodes can trust routes to
network. Since an ad-hoc IoT network may host multiple be fully connected during their data transmission. In [255],
applications with various QoS requirements, the middle the authors introduce a hierarchical clustering algorithm for
nodes in clustering will face a challenge on how to meet dynamic and heterogeneous IoT. This algorithm considers
the QoS requirements in those applications that may be in battery lifetime of the CHs to improve lifetime of the
conflict. network and consequently make the network more stable.

3) Mobility management: Mobility has always been consid- (directly or indirectly). There are two aspects of connectivity
ered a big challenge in wireless networks. Although mobility that clustering can improve as listed below.
is an advantage for many applications, it can introduce various 1) Improving coverage: A better coverage can connect
challenges in the network such as connectivity, reliability, and more nodes to the network and extend the sensing coverage
stability. area.
Existing solutions and techniques include: Existing solutions and techniques include:
• Inter-cluster handover mechanisms: Mobility of nodes is • Radio range management: In some clustering techniques,
supported by clustering based on handover mechanisms there are specific policies to limit radio range of nodes to
among clusters. Mobile nodes may join and leave the improve coverage. In [115], the authors use a directional an-
clusters during the network lifetime. Some methods are tenna to reduce energy consumption and also coverage of the
introduced to support mobile CHs and BSs for information network by controlling the transmission range of directional
gathering in the network [154] [163]. In [143], CHs are antennas. In [260], the authors introduce a model to analyze
considered mobile and used as mules to gather and transfer clustered D2D IoT networks in three-dimensional spaces
data. In [105], the authors use the speed of each node’s based on Thomas Cluster Process (TCP). They evaluate the
movement as a parameter to select CHs. By using TDMA network in the presence of co-channel interference from
to schedule access to the medium in each cluster, CHs both the same cluster and the other clusters, and investigate
estimate the connection time between members and CHs to the coverage probability.
detect the mobility of nodes between clusters. In [256], the 2) Improving nodes connectivity: A good connectivity
authors introduce a method to enable reliable uplink com- among nodes can make the network infrastructure reliable and
munications for clustered UAV IoT devices considering the fully connected resulting in improving data delivery.
energy consumption. They show that the proposed approach, Existing solutions and techniques include:
can manage mobile IoT nodes as well as provide reliable • Establishing stable connections: In mobile networks, con-
uplink communications. In [257], the authors introduce a nectivity is a big challenge because mobile nodes can loose
clustering technique for smart cities to optimize the planning their connection dynamically. In [105], the authors introduce
of the transportation traffic network among hotel service a method for mobile WSNs to improve connectivity which
centers in Tianjin. They introduce a clustering algorithm is based on an estimated connection time between CHs
using the deep belief network (DBN) to divide regions and and members. If connections between members and a CH
determine their service centers in Tianjin. In [258], Jabeur are lost, the members detect connection loss based on the
et al. introduce a new firefly-based clustering algorithm in estimated connection time and broadcast a request to find
which real-world things (RWTs) organize themselves into a new CH to avoid packet loss. Clusters overlapping can
clusters during the macro clustering phase and integrate improve connectivity of nodes and coverage. However, high
small neighboring clusters. Clusters can ask RWTs to join overlapping can cause various problems, especially delay
or leave the cluster due to events and their impacts. because of having more hops between a node and BS.
IoT-specific concerns. As mentioned above, there are some In [148], the authors introduce a clustering technique that
clustering-based solutions addressing topology management can control the overlapping of the clusters and reduce delay
(i.e., scalability, stability and mobility) in IoT. However, and energy consumption using a hierarchical infrastructure.
addressing topology management in IoT through clustering In [255], the authors introduce a new clustering technique
introduces the following challenges: for heterogeneous IoT networks based on selecting the
• Supporting QoS in scalable networks: As scalability is a optimal number of CHs to improve connectivity of hierar-
critical challenge in large scale IoT systems (e.g., UDN chical communication between nodes and BS. In [261], the
networks, IoV, and smart cities), supporting QoS would authors introduce a cluster-based framework for Unmanned
be realized in a distributed manner in order to reduce the Aerial Vehicle (UAV) IoT networks that is adaptive, self-
network management overhead [259]. In addition, two-way configurable, and resilient. The proposed framework cogni-
communication is considered as a challenge to support QoS, tively adapts with IoT network changes to provide reliable
because in some cases QoS of a stream from a node to a connectivity between spatially dispersed smart devices.
destination is different compared to the QoS of the responses IoT-specific concerns. Connectivity is a challenge due to
from the same destination to the same node. establishing the IoT network over an existing network in-
• Highly dynamic applications of IoT: Mobility is a common frastructure that is perhaps unreliable. Below, we discuss
aspect of several IoT applications such as IoV and smart challenges in applying clustering techniques to tackle IoT
cities. Adopting clustering as the mobility handling solution connectivity concerns.
can impose high overhead when the network undertakes re- • Network interfaces: Each route in IoT can include nodes
clustering due to the high mobility of nodes. with different types of network interfaces and each network
interface may have various disadvantages compromising the
reliability of the whole route. As an example, a Zigbee
F. Network Connectivity
connection between two nodes in a route can be less
Clustering can improve connectivity as each node has at reliable and cause higher delay when all other connections
least one connection to other nodes (directly) and also to BS of the route are WiFi. In addition, although some nodes are

physically close to each other, because of lacking a common reduce network load, and energy consumption. Data aggre-
type of network interface, they can not connect to each other gation, compression and fusion are the solutions applied in
and need middle nodes to connect. CHs and middle nodes to reduce the volume of data [80].
• Network infrastructures: Some technologies like 5G can In [264], the authors introduce a method to fuse data on
serve as the network infrastructure of future IoT systems, clusters established based on RSSI of IoT nodes. Access
even though they still suffer from some connectivity prob- points are the CHs, exploiting RSSI to establish clusters.
lems, e.g., blind spaces and millimeter waves are absorbed The fused data is then sent to the Cloud.
by physical obstacles [262]. The characteristics of the under- IoT-specific concerns. Like WSNs, the packet delivery rate
lying IoT network infrastructure can affect the performance should be improved in IoT to make the network infrastructure
of clustering techniques in connectivity provisioning. more efficient. Below, we discuss the challenge in applying
clustering techniques for packet delivery improvement in IoT:
G. Data Delivery Improvement • Unreliability and instability of network infrastructures: They
It can enhance QoS and reduce energy consumption and are the challenges against packet delivery improvement in
network overhead by reducing packet re-transmission, data IoT. To reduce data loss and improve packet delivery ratio,
loss, etc. There are a number of metrics for packet delivery node failure and connection failure should be reduced as
that can be improved by clustering techniques as listed below: discussed above. The stability of the IoT network is com-
1) Improving packet delivery ratio: It is a parameter that promised due to high dynamicity. In some IoT applications
has a high correlation with throughput. like IoV and smart cities, it is very challenging to keep the
Existing solutions and techniques include: network infrastructure stable compared to WSNs.
• Improving reliability: This refers to data loss and node fail-
ure reduction, and connectivity improvement in the network.
In [165], a method has been proposed to improve the packet H. Security
delivery ratio by selecting CHs based on their remaining There are different types of attacks that can happen due
energy, QoS parameters and the location of nodes. The to ad-hoc characteristics, e.g., hole attacks and DDoS [265].
proposed method uses the immune-inspired optimization Different techniques have been introduced to solve the attacks
algorithm to select routes for delivering packets. Industrial and detect malicious nodes, especially hole attacks which are
IoT is used in [263] as the network infrastructure featuring a very common in ad-hoc networks. In [266], the authors analyse
grid-based clustering algorithm to model the network load. the security mechanisms in IoT networks based on clustering.
The network is divided into unequal grids based on the Existing solutions and techniques include:
optimal grid length to optimize energy consumption and
improve the packet delivery ratio. • Distributed trust-based models: Clustering techniques can
• Reducing data loss: Data loss can also cause serious prob-
enable distributed trust-based models. In such models, each
lems, especially in safety critical applications, such as fire group of nodes includes one or more nodes that is (are)
detection, disaster management, and military-based infor- trustworthy. The trustworthy nodes are responsible to eval-
mation. Clustering can be used to detect node failures and uate the local nodes in case of security attacks. In [267],
reduce data loss. In [126], Izadi et al. introduce a clustering the authors introduce a trust-based CH selection model for
method that can detect node failure in an acceptable time WSN-based Intelligent Transportation Systems (ITS). Each
after failure, while avoiding data loss. In [247], the authors CH is selected based on residual energy, trust value and
introduce a routing protocol for IoT networks based on the number of neighbors. Each node is responsible for
clustering to maximize throughput and decrease packet drop monitoring the behavior of its neighbors. CHs are in charge
ratio together with reducing energy consumption. of evaluating the trustworthiness of each member based on
analyzing other members’ evaluation. In [178], a clustering
2) Optimizing number of packets received by BS: which
technique is introduced to provide security for wearable
has a high correlation with packet delivery ratio, but it refers
IoT devices. They propose a clustering-based anonymity
to the amount of data received by BS in a certain time period
method to preserve privacy of data gathered from wearable
compared to the data generated in nodes. In some papers,
IoT devices and guarantee the usability of the collected
this parameter has been investigated as a metric for evaluating
data. In [268], the authors propose a centralized detection
throughput of the network [161] [21].
system based on the data gathered from clusters to detect the
Existing solutions in WSNs and IoT. Clustering techniques
malicious gateways of clusters in IoT networks using packet
can be leveraged to optimize the number of packets received
drop probability as a means to monitor the gateways.
in BS. In the following, we discuss them:
• Network lifetime improvement: Improving network lifetime
IoT-specific concerns. Security is one of the main concerns
can increase the number of packets received over time by of IoT systems because of having weak and vulnerable nodes.
BS [152] [158]. In [248], the authors introduce a clustering Using clustering techniques for addressing security concerns
technique to improve the network lifetime and consequently in IoT will face a number of challenges as listed below:
the rate of packet transmission to BS. • Highly dynamic IoT: The massive scale of IoT deploy-
• Data aggregation: In some clustering techniques, the num- ments and their high dynamicity can pose more security
ber of packets sent to BS is reduced to improve QoS and challenges. As unknown IoT devices can join the network

in applications like smart cities, applying distributed trust- medium to nodes efficiently, e.g., by using scheduling tech-
based models would be a challenge as trustworthy nodes niques. To reduce medium access interference in an area,
may have not enough time to evaluate unknown nodes. nodes may reduce their radio ranges, which can compromise
• Resource-constrained IoT devices: Security solutions for network connectivity. Thus, to establish clusters, reducing
IoT are mostly focused on authentication and authorization. the interference needs to be considered during CH selection
As supporting security is a heavy task, applying authenti- and joining of the members. Clusters that can not support
cation and authorization techniques is resource demanding. UDN networks may not be a good choice for optimizing
In distributed models, clusters may include heterogeneous resource consumption. Densification is a new definition in
nodes, thereby having the same local authentication and D2D-enabled UDN cellular networks that resembles clus-
authorization techniques among members and the CH would tering, grouping and limiting D2D communication between
be challenging with respect to the heterogeneity of nodes. nodes to reduce interference [272].

I. Physical Layer Support J. Multi-Sink Support

Supporting multi-sink is a great solution to improve the
Although clustering techniques logically belong to the net-
efficiency of the network by increasing the number of data
work layer, they can contribute to the efficiency requirements
routing options to the end node. It can improve throughput,
of the physical layer, e.g., improving the efficiency of band-
packet delivery and fault tolerance, as well as reduce delay.
width reservation methods. Existing approaches in this field
Although multi-sink networks are more efficient compared to
revolve around distributed resource management.
single-sink networks, clustering techniques which are able to
Existing solutions and techniques include:
support multi-sink networks are more complex to design.
• Distributed resource management: Clustering techniques Existing solutions and techniques include:
can group nodes with a local resource management mecha- • Multi-point to multi-point (MP2MP) clustering-routing: As
nism in each cluster in order to more efficiently manage, the main solution, clustering-routing techniques need to
schedule and share physical resources, e.g., medium and connect nodes to the best sink (e.g., with respect to re-
bandwidth allocation to avoid transmission interference. As source consumption and delay) as an optimization problem.
a fundamental clustering method, adaptive clustering [54] Establishing these connections needs to have a multi-point
is used to efficiently allocate bandwidth to members in to multi-point inter-cluster routing. In [154], the authors
each cluster. It uses a virtual circuit definition to assign propose a method to support multiple mobile sinks. It uses
real-time connections and reserve time slots in clusters. CHs to maintain optimal routes which have minimum hops
In [269], the authors introduce a classification technique to to the latest location of mobile sinks. In [68], the authors
establish cells (i.e., clusters) for IoT architectures based on introduce a clustering technique that can support multiple
a mesh of IoT devices, roadside repositories and vehicular mobile sinks to gather data based on queries. They improve
mobile gateways. The model uses routing trees, the network the ’success rate’ which denotes the ratio of the number of
geometries, capacity of wireless connections, and stability successfully received data packets at BSs to the total number
of IoT queues to perform the classification. In [270], the of data packets generated by a node. In [273], the authors
authors introduce a method to utilize bandwidth alloca- introduce an SDN-based solution for resource-constrained
tion based on inter-cluster aggregation which reduces the IoT devices that need to select the best router to connect as
volume of data in a hierarchical clustering technique in the edge node. The network includes multi-sink (i.e., edge
each layer, in addition to reducing energy consumption and nodes) and the clusters use RPL to establish routes between
bandwidth allocation in upper layers. In [245], the authors clusters and edge nodes. They consider different metrics
use a heuristic scheduling model in clusters to adapt the and objective functions for routing, including Minimum
network with end-to-end delay and bandwidth allocation Rank with Hysteresis Function (MRHOF) and Zero (OF0).
requirements. They use a scheduling model that can react to In [274], the authors propose a Shortcut Addition strategy
different data flow changes in routers (CHs) by re-ordering based on the Particle Swarm algorithm for multi-sink IoT
active periods of clusters and optimizing the time-slot to networks. It constructs network topology based on a small-
access shared channels by each cluster. In [271], the authors world network. A fitness function is created by combining
introduce a multiple-input-single-output (CMISO) method the average path length and load of the sink node to
to prolong the lifetime of CHs. In the paper, they formulate evaluate the results of the particle swarm optimization (PSO)
the cooperative coalition selection for CMISO and use a technique. Each node selects the sink closest to itself based
quantum-inspired particle swarm optimization technique to on the results of the PSO algorithm. Sung et al. in [275]
select the optimum cooperative coalition. introduce a multihop clustering technique to minimize the
IoT-specific concerns. Supporting physical layer requirements required Internet connection in an IoT network. A node per
is an important issue in IoT. As listed below, different aspects cluster is responsible for connecting the nodes to the Internet
should be considered for designing IoT clustering techniques and the clustering technique minimizes the distance between
to support physical layer objectives. the end-nodes and those connected to the Internet.
• Ultra Dense Networks (UDN): In several IoT applications, IoT-specific concerns. Although supporting multi-sink is an
the network infrastructure is a UDN which needs to allocate important factor in improving efficiency, some issues should

be considered in designing an IoT clustering technique that and secure by storing transaction records on network nodes
supports multi-sink. We discuss the most important one below: in a distributed manner. General IoT security methods are
• Selecting the best sink (gateway): Supporting multi-sink not distributed and need many-to-one traffic streams which
networks is an important objective that can increase the effi- cause vulnerabilities such as central node failure, delay, and
ciency of ad-hoc networks in different aspects such as QoS unnecessary energy consumption [265], [278].
and resource consumption. Most clustering techniques that Clustering techniques can facilitate integrating IoT and BC
support multi-sink focus on designing a routing method to to improve the security and efficiency of IoT. Using clustering
find the best route to transfer data to outside of the network. techniques, nodes can share their resources in a distributed
Although some clustering techniques consider multi-sink, management model to perform resource consuming tasks of
especially as a scenario in their performance evaluation, BC. In addition, grouping nodes can help BC to provide
there are not many techniques that have a specific solution smaller logical networks that can share resources or work
for multi-sink networks. In IoT, selecting the best sink based on trustworthiness. Mining as a complicated task should
is a multi-objective problem. Two steps are introduced to be performed by several nodes and clustering techniques can
select the best sink: i) selecting the best sink based on enable distributed execution of it.
different costs to transmit data through the sink; ii) selecting Common features of IoT and BC make BC suitable to
the best route from the node to the selected sink, which address the security needs of IoT as listed below:
is a challenging issue in IoT as nodes in a route may • Both technologies are decentralized which can support scal-
be heterogeneous in terms of the properties of network ability and eliminate non-efficient traffic overheads.
interfaces (e.g., bandwidth and delay) [276]. • BC provides security and privacy on top of the overlay net-
work which includes untrusted nodes. Generally, IoT nodes
VI. I OT C LUSTERING AND M ODERN N ETWORKING are resource-constrained, while evaluating their trustworthy
PARADIGMS level is a resource consuming task.
So far, we discussed how clustering can address various • BC nodes are anonymous due to supporting privacy of users

quality-related objectives in IoT. Along this, clustering can which is suitable in IoT use cases.
be exploited to facilitate adoption of recent technologies Although BC can offer many advantages for IoT systems,
integrated with IoT systems, in particular with respect to the there are significant challenges which should be addressed to
barriers originating from networking issues. We focus on four implement BC as listed below:
new pioneering computing and networking paradigms that • BC technology uses some nodes in the network as miners to
are being rapidly integrated with IoT, namely Blockchain, mine the blocks of the ledger. Mining is a highly resource-
SDN/NFV, Fog/Edge, and 5G. First, we describe each of them intensive task that calculates Proof of Work (POW) as a
briefly, especially from the networking viewpoint. Next, we cryptographic puzzle to add new blocks to BC. Each verified
discuss how clustering can be exploited to improve various transaction will be added to a pending block in multiple
design aspects in these technologies. For each technology, miners and all of them should mine the same block, caus-
we discuss the limited number of existing clustering-based ing delay, resource consumption, etc. Since IoT nodes are
approaches. Then, the benefits of clustering is presented per generally resource-constrained, mining is the main challenge
technology. Finally, we identify the challenges raising in in implementing BC in IoT due to resource consumption,
integration of clustering techniques with these technologies. delay and management of distributed mining on different
machines.
A. Blockchain and IoT • BC techniques can be categorized as public and private

Blockchain (BC) technology is fundamentally developed to and also permission-less and permissioned. Public BCs can
store transactions reliably in a distributed manner, especially have global-anonymous participants, but private BCs are
for cryptocurrencies. BC is a distributed ledger among network application-specific and designed to establish a BC among
nodes to store transactions and improve the security, relia- known participants, mainly to reach a specific objective. On
bility and efficiency of a decentralized network. Consensus the other hand, participants of permission-less BC do not
algorithms [277] are used to guarantee that the BC is de- require permission to be part of the BC, but in permissioned
centralized, reliable and secure. In a BC network, all nodes BCs, a control layer controls the actions performed by
have P2P connections similar to BitTorrent and contribute to permissioned participants.
storing, mining, and accessing transactions. Each node has two • BC has high overhead traffic which is not suitable for IoT

keys, private key and public key, to encrypt and decrypt the devices that suffer from bandwidth limitation.
messages from and to the node, respectively. Recently, some works have been carried out to apply BC in
BC can be used in various security-related applications, not IoT, out of which a number of approaches use clustering tech-
only cryptocurrencies. Concerning IoT applications, security niques to integrate IoT and BC. In [279], the authors propose
and privacy are still two major challenges due to the scale a method based on the concept of ’The friend of my friend
and distributed nature of IoT, and the resource scarcity of IoT is my friend’. They use BC for WSN-IoT mobile networks to
nodes. Traditional security methods consume a lot of energy solve authentication problems of resource-constrained nodes in
and impose high processing overhead. Integrating IoT and IoT. The proposed method authenticates mobile nodes that are
BC can promise a network infrastructure that becomes safe moving from one cluster to another. They use BC to reduce the

number of exchanged messages to provide security and also nodes that can trust each other and store blocks, ledgers and
eliminate Key Derivation processing. She et al. [280] propose a transactions. By having clusters, members can trust CHs and
BC trust model in 3D environments to detect malicious nodes CHs can trust each other using multi-signature transactions.
in WSNs based on Consortium Blockchain and clustering Moreover, scalability can be supported by allowing forking in
techniques. BS publishes contracts and CHs use them as verifi- different clusters or different layers of hierarchical clustering
cation nodes and also provide digital certificate-based identity techniques. In [287], the authors propose a secure routing
information for cluster members. In the proposed method, CHs protocol for smart homes. They use a clustering technique to
are pre-selected to detect malicious nodes. In [281], the authors separate indoor spaces by allocating region numbers for each
introduce a security layer in IoT to detect and isolate malicious space like kitchen, bedroom, etc. They create super clusters
nodes based on clustering techniques. They consider CHs as and common clusters. Super clusters include reference nodes
resource-rich nodes which are simultaneously miners and also that can connect to a common cluster of CHs directly. They use
keep a copy of nodes’ firmware. They assume that clusters are a beta-based trust management model to improve the security
already available and if the CH is compromised, the whole of CH selection.
cluster would be untrustworthy. They use an authentication In the case of permissioned BC, Amiri et al. [288] introduce
service as a smart contract on CHs to authenticate nodes. For a cluster-based sharding technique for permissioned BC to
authorization, all nodes in a trusted cluster should be able to support scalability of BC. Their proposed technique, called
authorize other members. SharPer, establishes clusters to assign different shards of data
In [282], the authors proposed a clustering technique for IoV to the clusters. Each data shard is replicated on cluster mem-
to reduce the load of transactions on IoV devices. The pro- bers to support intra-shard, and cross-shard transactions. Their
posed method determines slots which can update BC ledgers proposed BC ledger is based on the directed acyclic graph and
optimally by selecting appropriate CHs. Using the clustering establishes consensus by clusters for cross-shared transactions
technique, they reduce energy consumption for about 40% and based on a flattened protocol. Hyperledger fabric is also
the volume of transactions for about 82%. A distributed ledger introduced by IBM as a modular and permissioned architecture
mechanism is used among CHs and it makes the decision about to assign the roles to nodes in a distributed architecture of BC
when and how the ledger should be updated on CHs. They [289]. In [290], the authors propose a BC-based multi-layer
propose a location-based ledger-offloading model that selects security model for IoT. They divide the IoT network to k-
new CHs to optimize energy consumption. unknown clusters based on Genetic algorithms and the PSO
In [277], the authors use a BC model to provide a distributed algorithm. In each CH, they provide a local authentication and
cyber infrastructure for the future smart grid systems. They authorization technique to improve the security of IoT. They
proposed a model based on public and private keys in which also use BC to provide security for inter-CH communication
each meter-node in the network collects data, encrypts it and and CH-to-BS communications. They also propose a global
then broadcasts it. Other meter node decrypts data and verifies BC to secure the BS communications. An open source BC
the data by a voting method. If the data has enough votes to hyperledger fabric is used to verify the techniques. In addition,
verify, then the data is packaged into a block. Their application Mbarek et al. in [291] introduce a multil-evel BC system to
is a smart meter data aggregation network consisting of re- improve data security and privacy of IoT along with reducing
gional BC and wide-area BC. In this application, smart meters the response time and resource consumption. They propose
are clustered called regional clusters. Then, clusters establish three levels called mirco level, meso level, and macro level,
regional BC networks which create a wide-area BC network. which contain members of clusters, CHs, and servers of BC,
The model establishes a two-tier hierarchical BC where CHs respectively. In addition, the authors introduce the use of
as gateways secure and maintain the wide-area BC network. A mobile IoT devices that move in different levels of IoT and
CH can be a storage or processor of the data sent from cluster BC to perform BC functions, e.g., hashing and encryption.
members. Additionally, there exist some BC research works that
The authors in [283] use BC in IoV networks to improve use clustering methods, while their main contribution is not
the privacy and security of vehicles. They propose a de- proposing new clustering techniques. In [292], the authors use
centralized architecture for the smart vehicle eco-system in CHs to collect, fuse and encrypt data gathered from nodes.
which each node is part of an overlay network and a cluster. They use attribute-based encryption to provide confidentiality
CHs manage BC and maintain the architecture called overlay and access control. CHs are the owners of data and BC miners
block managers (OBM). CHs use changeable public keys and are the other nodes. In [293], the authors use clustering as a
multisig (multi-signature) transactions. Multisig transactions node grouping technique in the supply chain. CHs are selected
are broadcast among OBMs. Each node of OBM has a list of randomly, proposing a new block when needed.
public key pairs to verify signatures. Each OBM node controls 1) Importance of clustering for BC: As mentioned above,
access to the transactions of its cluster members. IoT has some limitations which make implementing BC for
Ali Dorri et al. [284] propose a new BC method called IoT systems challenging. Clustering can be used to overcome
Lightweight Scalable Blockchain (LSB) which replaces the these obstacles.
mining procedure with a new trustworthy model in a clustered • Improve the compatibility of BC with IoT: As mining is
infrastructure. In [285] and [286], A. Dorri et al. propose a resource-intensive process, it can be replaced with other
a method based on LSB to replace the mining model with techniques, e.g., trustworthy mechanisms. Using clustering
a trustworthy mechanism in smart homes. CHs are used as methods, the network can be structured in such a way

that clusters trust each other and do not need to mine are two challenges, which make resource sharing difficult.
blocks. As explained before, A. Dorri et al. [284] propose a Dynamic clustering is needed when a cluster is subject to
trustworthy mechanism rather than mining. Clustering can dynamic changes. Additionally, using task offloading meth-
provide an extra overlay network on top of IoT, which can ods for mobile nodes can address the dynamicity problem.
be used to group nodes. Grouped nodes can evaluate their • Compatibility: As mentioned above, standard BC is not
trustworthiness and detect malicious nodes. Then, clusters applicable in IoT due to resource-intensive processing tasks
can trust each other or have an extra layer that uses BC in BC. Designing a method that can make the standard BC
as well. In other words, by using clustering methods, it compatible with IoT can be useful, e.g., the technique pro-
is possible to have a hierarchical BC which distributes posed in [285]. However, such techniques like trustworthy
responsibilities, not among nodes, rather among groups of based models essentially impose high traffic overhead rather
nodes. than processing overhead. To have compatible models and
• Resource sharing in BC: In implementing a BC, nodes need eliminate mining, distributed techniques should be applied
large amount of resources to mine blocks which are not regarding control of their overheads.
applicable to IoT systems due to resource scarcity. Although • Scalability: Although different techniques can support scala-
mining is a centralized task, it is possible to perform it in bility in BC IoT infrastructures, managing a huge number of
a distributed manner. There exist two different approaches nodes can induce high processing and traffic overheads. In
on distributed mining. Distributed miners [294] is the first large-scale networks, maintaining the efficiency of clustering
approach which divides the mining task into work units would be a resource-intensive task.
and uses different machines to execute them. Bitcoin Plus,
Bitp.it and Stealthcoin are examples of distributed miners.
B. SDN, NFV and IoT
The second approach is called pooled mining [295], in
which miners share their processing power to mine and Software-Defined Networking (SDN) is basically a central-
split the reward based on the amount of their contribution. ized networking paradigm to decouple the control plane from
Both these approaches can be used in IoT for distributed the data plane of data transmission [298]. In SDN, the network
mining based on share resources. Clustering techniques intelligence (i.e., the control function or the control plane)
can be used to propose clusters as pools and CHs will is centralized at one or a set of control entities (i.e., SDN
serve as a coordinator among pools. Especially in Fog controllers), while the data forwarding plane is abstracted for
computing models, which have powerful nodes, it is possible applications and networks services. In the first generation of
to establish clusters to share resources of powerful fog SDN, the OpenFlow protocol [299] is used between the control
nodes. On the other hand, Some BC solutions use solo and data planes. For example, an OpenFlow switch has one
mining [296] in which each miner includes only one node. In or more forwarding tables that are controlled by a centralized
this case, clustering can determine and use CHs as miners. In controller, which realizes programmability in the control plane.
this case, CH is responsible for mining transactions related Network Function Virtualization (NFV) is the concept of
to its members or neighbors. Besides, there are different relocating network functions from dedicated hardware appli-
P2P based mining methods to increase the efficiency of BC ances to software-based appliances running in the cloud envi-
which can be implemented using clustering. As an example, ronment or on general-purpose commodity equipment [300].
P2POOL [297] uses a peer-to-peer network of node miners Using NFV, conventional network functions (NF) will run on
to create a mining pool. Using clustering, a more efficient a virtual machine (VM) as a one-to-one mapping model or is
P2P distributed network can be established to implement decomposed into smaller components called Virtual Network
such methods. Function components running on multiple VMs. Virtual Net-
work Functions (VNF), which represents the implementation
2) Challenges of integrating clustering techniques with BC:
of NFs, are deployed and executed on an NFV Infrastructure
To integrate clustering techniques and BC, we face a number
(NFVI). The NFVI consists of virtual resources, which are
of challenges as listed below:
abstracted and logically partitioned from underlying hardware
• Heterogeneity: In resource sharing techniques, fairness is an resources (computing, storage, and networking) through a
important factor due to the resource-intensiveness of mining. virtualization layer.
Generally, IoT nodes are heterogeneous and the positions of 1) Importance of clustering for SDN and NFV: Clustering
nodes in IoT are unbalanced, causing unfairness. To design has the potential to contribute to the efficiency of SDN
efficient clustering techniques, clusters should be equal in and NFV, such as communication and control. In particular,
terms of available resources. Resource consumption of the clustering can enable a better division of network nodes and
mining process is generally unpredictable which may cause their association with the SDN control domain based on pre-
unbalanced resource consumption on different machines. To defined communication criteria, e.g., nodes vicinity. Clustering
support fairness, all clusters should have balanced resources; can also enable more efficient implementation of the control
thereby, clusters should be composed of heterogeneous component in SDN by facilitating distributed control through
nodes. clustered nodes and their CHs. The main advantage of using
• Dynamicity: To execute tasks on shared resources in a clustering techniques in NFV is that the amount of traffic that
cluster (i.e., mining and storing ledger), the cluster should be needs to be traversed through NFV based nodes in the same
stable between two set-up phases. Mobility and dynamicity cluster do not need to leave the cluster. This can significantly

reduce the amount of traffic routed throughout the network, in which can be implemented in different ways, like data flows.
particular in mobile edge computing networks. Clustering can facilitate the development of orchestration for
However, there are not many works focusing on the ex- NFV by utilizing CHs as the point of orchestration.
ploitation of clustering techniques in SDN scenarios. In [301], • Software-based clustering: a cluster is generally built on
as one of the few works in this category, a new clustering a network topology which adheres to the given clustering
algorithm, The Whale Optimisation Algorithm (WOA) is requirements. From the NFV viewpoint, a given clustering
proposed based on the concept of SDN for IoT systems. WOA mechanism and the associated algorithm can be derived as
considers both sensor resource restrictions and the random virtualized functions installed on, e.g., a general-purpose
diversification of IoT node density in the geographical area. computing device. In this way, more flexibility and dynam-
It divides the sensing area by the SDN controller into virtual icity will be enabled in the implementation of clustering
zones (VZs) to balance the number of CHs according to the functions.
node density in each VZ. Then, it uses WOA to define the
optimal set of CHs based on residual energy, communication
cost and node density. In [302], scalability, fault tolerance C. Fog and Edge
and interoperability in the centralized control of the SDNs IoT devices and services are proliferating into a massive
are challenged and a novel clustered distributed controller scale, thereby, IoT data, services and applications are being
architecture is proposed in the real setting of SDNs. The migrated to the Cloud, enabling powerful processing and
distributed cluster implementation comprises of multiple SDN sharing of IoT data beyond the capability of individual things.
controllers. The clustered controllers perform load balancing Moreover, due to requirements such as mobility support,
by distributing the number of connected OpenFlow switches location-awareness and low latency, the cloud computing
between instances of the controllers in the cluster. Similarly, paradigm has been recently extended from the core of the
in [303] a system is proposed to assign controllers to clusters network to the edge—Fog Computing. The fog nodes, residing
by optimizing the maximal distance between two controllers between the IoT devices and the Cloud can be structured in
in the same clusters in order to reduce the time complexity of various forms, such as hierarchical topologies. More efficient
load balancing in the SDN control plane. In [304], a model organization of fog devices can lead to more efficient data
based on SDN is proposed to prevent different attacks in IoT collection, processing and transmission to the Cloud. Fog
systems. Clusters of SDN controllers are created and a CH computing and edge computing are quite similar in meaning.
selection process is proposed to manage and control different Both are concerned with leveraging the computing capabilities
security issues in a particular domain. In [305], a WSN is in a local network to perform computation tasks that would
divided into multiple clusters, with CHs acting as a controller otherwise have been executed in the Cloud. However, Edge
in each network. Communication between nodes and CHs is computing usually occurs directly on the IoT devices or a
managed by gateway nodes and each cluster is considered to gateway device that is physically close to the sensors and IoT
be an SDN domain. devices. Fog computing moves the edge computing activities
There is also a limited body of work on using clustering to computing nodes that are connected to the LAN or into the
techniques in NFV. In [306], the concept of clustered NFV LAN hardware itself so they may be physically more distant
is proposed. An existing solution for combining clustering from the sensors and IoT devices and spread geographically
and NFV is leveraged to compute the optimal number of from edge devices towards the Cloud.
clusters to minimize the end-to-end time of MEC services. The 1) Importance of clustering in Fog and Edge: Clustering
proposed scheme is applied to form multiple MEC clusters of has the potential to address efficient data collection and pro-
NFV enabled nodes within the radio access networks (RANs). cessing in Fog- and Edge-computing environments. Clustering
In [307], the authors propose a systematic virtual networking can not only improve the robustness of the Fog network but
architecture to perform the virtualization control and monitor- also allow better control of data management. As an example
ing of a CPS, in which NFV configuration and orchestration for the latter, IoT data processing in large deployments is
can be realized. A game-theoretic topology decision approach bounded to a geographically defined area. Figure 5 shows
is proposed to control the topology of the clustering and virtual an overview of how clustering can be applied to fog-based
network functions deployment at run-time in a CPS. scenarios in which fog nodes are organized in a hierarchical
2) Challenges of integrating clustering techniques with model. This model is suitable for large-scale IoT deployments,
SDN and NFV: We identify the following future directions e.g., in smart cities [308]. As can be seen in the figure, a
in using clustering techniques in SDN and NFV: cluster can be composed of a set of IoT devices and the
• Distributed control: thanks to the structure of clusters and associated fog node as the CH. The nodes in a cluster can
the possibility of CH-level decision making, the centralized communicate through D2D or M2M technologies. In this
components of SDN can be decentralized in cluster based model, the fog node is often called as the edge node acting
topologies in order to address more efficiently the capacity as CH. As the alternative scenario, a cluster can also consist
concerns in large IoT applications. This implies that the data of only fog nodes, which fits in scenarios with, e.g., fog-level
forwarding plane is limited to the nodes in a cluster, while load balancing and workload distribution for intensive data
inter-cluster data forwarding is still a challenging issue. processing [309].
• Clustering for orchestrating in NFV: a key function of NFV Recently, a number of works have been reported on the
is the realization of orchestrated virtual network functions, usage of clustering techniques in fog computing scenarios. The

for fog devices that are in charge of collecting and pro-

cessing privacy-sensitive data, e.g., the power consumption
profile of a neighborhood. Using clustering, the collaborative
fog nodes (for data processing) can form clusters with well-
defined privacy constraints. In this way, the privacy issue
will be easier to maintain, in addition to the fact that inter-
cluster communications can be mapped to the privacy levels
of the end-user application.
Fig. 5. A clustering model in hierarchical fog architectural models • Optimization of resource allocation: fog resource allocation
is a complex problem. It refers to how the available fog
resources should be allocated for data processing to satisfy
more popular line of research in this area is related to using quality requirements such as low latency, and cost. To allo-
clustering for cost-efficiency, low power consumption, and cate resources efficiently, optimization algorithms should be
robustness in fog networks. A comprehensive cluster-based leveraged, taking into account the pre-defined optimization
approach in fog computing is proposed in [310]. It proposes parameters in resource allocation. Clustering can be the
an algorithm for cluster formation and load balancing for fog basis for optimization such that the nodes in the cluster are
computing. The algorithm has a customizable design where identified based on the optimization goals [316]. In this way,
metrics, scheduling rules, and clustering objectives can be set the management of resources and execution of optimization
according to specific application and network requirements. algorithms will become more efficient.
Another application of clustering is in data caching. In [311], • Clustering for running data-intensive tasks: in data-intensive
a clustering method is proposed to group spatially proximate applications, tasks may run on different nodes, while the
user devices with mutual task popularity interests with their results of task execution are sent to other nodes in the
serving cloudlets. Then, cloudlets can cache the popular tasks’ network. Clustering can serve as a local network of fog
computations of their cluster members to minimize computing nodes hosting the execution of interrelated tasks and sending
latency. In [312], a communication service between the fog the task execution results to other nodes in the network, e.g.,
layer and the IoT device layer in fog computing is provided, to the Cloud.
in which dynamic IoT device clustering is proposed to reduce • Computation offloading: This is an important feature in fog
the system complexity and the delay for the IoT devices with computing due to QoS requirements or resource limitations.
better channel conditions. The concept of edge-managed clus- A task may be offloaded to a fog device from other nodes
tering is proposed in [313] to supervise the operations among or the Cloud, or uploaded to the Cloud from a fog node,
cooperating devices forming a cluster. The cluster manager respectively. With respect to the former, a critical efficiency
is proposed running on an edge access point, providing both parameter of the offloading process is selecting the best can-
connectivity and fast management response to the attached IoT didate for task offloading. Clustering can provide a suitable
devices. platform for facilitating the selection process. In particular,
There are also contributions to the application of clustering when CH has an overview of the resource capabilities of its
techniques in fog-based systems, e.g., clustering for the pur- nodes, it can be exploited to find the node(s) for offloading
pose of better application software management. In [314], an based on the resource availability information. Moreover,
architecture for edge cloud PaaS is presented for application when the computation offloading is limited to a geographical
and service orchestration to manage and orchestrate applica- area (e.g., due to privacy preservation), the counterpart
tions through containers. They use clustering to create a cluster cluster(s) specify the boundaries for the techniques that
of Raspberry Pi nodes (as fog nodes) to run computationally should select the candidate nodes for offloading.
intensive software. In [315], a distributed consensus algorithm
is proposed for decision making in service-oriented IoT. Clus-
tering is exploited to partition the IoT network into multiple D. 5G and IoT
clusters for a local consensus. The local consensus can be Today, different generations of cellular networks cover at
achieved within each cluster and can then be used to make least 90% of the world’s population. 2G providing voice
consensus decision in knowledge sharing and integration of communication, covers 90%, while 3G used for transferring
functional capabilities. The authors use a distance-based static data and voice covers 65% of the people in the world [317].
clustering scheme for node clustering. In [49], the authors 4G and Long Term Evolution (LTE) are the generations after
investigate container-oriented operational frameworks for IoT. 3G supporting IoT applications not only themselves but also
They propose edge-managed clustering, where a manager su- by establishing concrete overlay network infrastructures like
pervises the operations among cooperating devices. The cluster NB-IoT [317]. There are several issues in supporting IoT
manager runs on an edge access point to allow connectivity applications in the 4G network which need to be addressed,
and fast management response to the attached IoT devices. such as capacity, data rate, end-to-end delay, massive device
2) Challenges of integrating clustering techniques with Fog connectivity, communication cost and QoE [318]. 5G is an
and Edge: We identify the following future directions in using emerging generation of cellular networks designed to enhance
clustering techniques in fog computing: the efficiency of data transmission and address the problems
• Clustering for addressing privacy: clustering can be applied of LTE. Although 5G is not designed specifically for IoT, it

is expected that it will provide the main backhaul of IoT in TABLE V

2025 [319]. A large number of solutions have been proposed T HE SPECIFICATION AND COVERAGE AREA OF DIFFERENT 5G CELL TYPES

on the integration of IoT applications and 5G infrastruc- Cell Type Radius (Km) Power (W) Number of Users Covered area Example
Macro Cell 8 to 30 10 to >50 >2000 Outdoor Cover a City
tures [320]. Third Generation Partnership Project (3GPP) as a Micro Cell 0.2 to 2.0 1 to 10 100 to 2000 Indoor/Outdoor Cover a Street
Pico Cell 0.1 to 0.2 0.25 to 1 30 to 100 Indoor/Outdoor Cover a Building
collaboration among different associations designs standards Femtocell 0.01 to 0.1 0.001 to 0.25 1 to 30 Indoor Cover a Flat

for cellular networks like 5G to establish concrete infrastruc-

tures that support applications with requirements such as low
latency, QoS, energy efficiency, etc. [320] [321] [322] [323]. find users close by to transfer data. Today, Wi-Fi direct is used
Different technologies are emerging to improve the efficiency in such applications (e.g., Zapya [336]) for direct data trans-
of 5G, and different standards are being released based on mission. Natural disasters and crisis management are other
them. Although there exists a large number of such tech- application types of D2D 5G networks that need to establish
nologies, some technologies are the foundation stones of an infrastructure-less network when BSs are not available.
5G. There are 5+1 fundamental technologies which form In IoT, D2D can reduce delay, improve coverage, optimize
5G [324] [325] [326] [327], including Millimeter Waves energy efficiency and support QoS. Different applications like
(mmWave) [328], Small Cell [322], Massive MIMO [328], Vehicle to Vehicle (V2V) [337] can use a combination of
Beamforming [329], Full Duplex [330], and Device-to-Device D2D and Small Cell to improve the efficiency of the network
Communication (D2D) [331]. infrastructure.
Although D2D is not a fundamental technology for 5G, it is From a topology management perspective, combining D2D
emerging as a significant technology to support 5G objectives. and Small Cell yields a network infrastructure which resem-
The above technologies can improve the efficiency of LTE and bles with multi-gateway (i.e., multi-sink) WSNs. Similar to
4G to satisfy the requirements of 5G. Although all these tech- WSNs, this topology needs to be managed in order to satisfy
nologies support each other, Small Cell and D2D are used to different objectives, e.g., energy efficiency, handling band-
form the network infrastructure, logically in the network layer. width limitations, and delay reduction. These two technologies
Almost all other technologies contribute to improvements in have the potential to be used for clustering in 5G networks.
the MAC and physical layers. Small Cell is a technology that To the best of our knowledge, the only paper that studies the
establishes a hierarchical network infrastructure to cover blind effect of clustering in IoT 5G scenarios is [19]. The authors
spaces, optimize energy consumption based on reducing the consider the use of clustering techniques in 5G IoT networks
distance between devices and small cell BSs, and support to support energy efficiency, distributed processing and hierar-
QoS. Small cells break up a cell site into smaller cells, chical management of the network infrastructure. The authors
usable in indoor and outdoor environments called Macro cells, studied five challenges including node heterogeneity, resource
Microcells, Picocells and Femtocells, respectively, as listed in and financial cost of transmission, user utility, smart core
Table V. As an example, a Femtocell as the smallest cell can network, and supporting mobility. However, Small Cell and
connect to a bigger cell or directly connect to the Internet by D2D are not studied as a 5G infrastructure in the paper.
wire. The way of connecting is same for all types of cells. In [338], the authors use a clustering method in D2D-assisted
This hierarchical model can help 5G to cover all blind spaces, LTE-A networks for tie-breaking to maximize fairness and
reduce energy consumption, and improve QoS [332]. throughput. In their method, CH is opportunistically different
Although Small Cell can enhance the efficiency of the at each scheduling frame to manage cellular traffic. In [339],
network infrastructure management in 5G, there are still the authors introduce user-centric cluster-based coordinated
some challenges with respect to supporting QoS, improving multi-points in 5G. Clusters can reduce the overhead of
bandwidth and energy consumption. D2D communication can intercell coordination by limiting it to within the cluster only.
address these challenges. D2D is introduced for short-range In addition, they use the re-clustering method to distribute the
communication between two devices without using BS, result- traffic of cells to neighboring cells. Network densification is
ing in less energy consumption, and improving load balancing also a method that can use clustering techniques to improve
and QoS. Resource-constrained IoT devices normally do not the efficiency of spectrum utilization [340]. In [239], the
support wide coverage of connectivity. D2D technology was authors use a clustering method to achieve green IoT in
first introduced by Qualcomm, called flashlinQ, to enable 5G networks and solve the problem of handover. In [341],
D2D in LTE-A [333]. It is officially addressed in 3GPP LTE Song et al. use a clustering method to minimize network
release 12 [334]. D2D is introduced by 3GPP, also known energy cost and maximize connected IoT devices in 5G small
as Proximity Service (ProSe) which supports multi-hop relay cells based on the k-means algorithm. In [342], a clustering
networks [335]. In release 15, the main objective is to improve method is introduced based on supporting Massive MIMO
the power efficiency of remote User Equipment (UE), such in 5G. In [343], a method is introduced which uses cluster
as IoT devices and wearables, by allowing them to establish forms in 5G cellular networks to support D2D. The above
D2D communication with UEs acting as a relay. Although literature show that clustering improve the efficiency of 5G by
D2D in ISM bands like Wi-Fi direct is well-established, D2D managing nodes and resources in each cluster locally. In this
standardization in cellular networks is still a challenge. paper, we review the opportunities and challenges of clustering
D2D communication in 5G can enable the development of techniques in 5G networks based on Small Cell and D2D.
novel applications and enhance existing mobile applications. 1) Importance of clustering in 5G Networks: As mentioned
For example, social networks apps and games can use D2D to above, in 5G settings that support Small Cell and D2D commu-

Clusters
Example

D2D Connection
5G Infrastructure
Connection

Physical Obstacles
(Wall)

CH Pico Cell Marco Cell

BS BS

Member Micro Cell

(UE) Femto Cell
BS BS

Fig. 6. An illustration of how IoT can be integrated with 5G based on clustering

nication technologies, clustering can be a foundation stone for clustering can reduce overall network energy consumption
topology management. Clustering techniques in 5G can sup- through balancing energy consumption in nodes. Similarly,
port different applications and also improve the efficiency of D2D and Small Cell are introduced to optimize energy
communication based on grouping the nodes and hierarchical consumption. Clustering techniques can be used in a 5G
management. Below, we discuss the importance of clustering D2D network to delegate data transfer to the nodes that
in 5G based on the different efficiency improvements achieved have higher energy resources.
by 5G clustering. • Spectrum efficiency: Spectrum efficiency is an important
factor in 5G which is supported by the mmWave technology.
• Applications: Not all applications over 5G need to use BS By mmWave, UEs use a 10-300GHz band that supports a
for data transmission, e.g., online group games. Based on very fast gigabit-per-second data rate. In mmWave, com-
5G characteristics, D2D communication can be established munication between nodes is limited by their transmission
among nodes that have Line of Sight (LoS). Clustering power and physical obstacles. Using clustering techniques
can connect different UEs to one UE as CH based on in such networks opens up for low range communication.
commonalities of applications or services. This can support low multi-user interference (MUI) in ultra-
• Improving massive MIMO efficiency: Massive MIMO use dense networks like IoT. UEs can communicate with CHs
BSs with very large antenna arrays to communicate with by allocating different bands and adopting LOS or multi-
several users in each time frequency [344]. With a large hop D2D. In addition, in the case of having no LOS between
number of UEs which communicate with BS directly, in- nodes and CH, intra-cluster routing can be provided by D2D
terference can be quite high in BS, which can result in a communication.
smaller number of users being supported. Clustering can be • Resource usage efficiency: Each UE in 5G has specific
considered as a solution to scale down the number of users resources that can be used in the network. Supporting D2D
connected directly, while the resulting multi-hop network allows 5G to use the resources of UEs to support different
topology can lead to increased delay in the network as services and applications. Clustering techniques can be used
discussed in [37]. Given the high speed of D2D commu- to group a set of nodes and assign a task to a node group.
nication in 5G, clustering will not impose significant delay, • Network heterogeneity: In 5G, nodes support different net-
in addition to the fact that the number of UEs connected work interfaces. Clustering techniques can be used to switch
directly to BS can be controlled, resulting in improving the between network interfaces for data transmission. As an
efficiency of Massive MIMO. example, nodes can connect to CH using D2D and then
• Caching: Leveraging small cells and D2D, it is possible to use Wi-Fi direct to transfer data to the next hop.
cache data in CHs. Cluster members can be UEs that need
the same data. There are many types of applications that Figure 6 shows an example of how clustering techniques can
need caching, such as multimedia, online video conferenc- improve connectivity and efficiency of 5G networks.
ing, etc. Caching can help avoid data redundancy in data 2) Challenges of integrating clustering techniques with 5G:
transmission. In hierarchical clustering techniques, caching Although using clustering techniques in 5G-IoT networks has
can reduce the volume of data significantly by reusing data many advantages as mentioned above, it also introduces a
hop by hop. In addition, different services in the Fog can use number of challenges:
cached data in CHs and avoid asking the Cloud to transfer. • Heterogeneity: Most clustering techniques are designed for
• Energy consumption: As one of the main challenges in 5G, homogeneous networks, not considering requirements of

other layers as a parameter in establishing clusters. As an clustering technique.

example, mmWave technology can be confined by physical • Hierarchical management: In UDN, network management
obstacles to connect to UEs. To establish clusters, physical is a challenge due to the scale of the network in terms
obstacles can change the form of clusters, number of mem- of number of nodes and communication load. Hierarchical
bers, etc. which can cause inefficiency. On the other hand, distributed management can address this challenge, appli-
clustering techniques should consider application services. cable to cells of all sizes. Clusters should be established
As traditional clustering techniques are designed to establish based on the different sizes of cells as the possible cost of
clusters based on the network layer objectives like improv- causing cluster overlap. Reducing overlaps is a complicated
ing QoS, new parameters should be considered in cluster process when establishing clusters within cells. Small Cell
creation. As an example, to improve efficiency, clusters can technology divides the network into cells and clustering
be established to unite the nodes that use the same service divides the network into clusters. This implies that two
as data generator or data consumer. These parameters make identical node grouping techniques may simultaneously
establishing and maintaining clusters more complicated and be used over a common network infrastructure. Applying
resource consuming. the same clustering techniques simultaneously can cause
• Interference: As 5G-IoT networks can possibly be a kind unexpected efficiency problems. Overlaps and interference
of Ultra-Dense networks, massive volumes of data in a among clusters and cells may cause overheads and topology
cluster can cause interference among members. Generally, management problems. 5G networks supporting Small Cell
clusters can be established based on common applications technology are similar to multi-sink ad-hoc networks in
among neighbor nodes, but also some applications can have which each BS is a sink. Therefore, selecting the best sink
a high volume of data to transfer which cause interference as the gateway of a packet stream would be challenging due
in an area. MUI reduction techniques might be considered to financial and resource costs.
in cluster establishment and member selection to distribute
potential interference nodes among clusters. Using MUI as VII. C ONCLUSIONS
a parameter to establish clusters, may result in a clustering Topology management is an important issue in the IoT
model in which each cluster contains members belonging to networks that resemble ad-hoc networks, e.g., D2D-based IoT
different applications which again would lead to interference networks over cellular networks. Clustering is the most pop-
reduction in a cluster. On the other hand, having nodes ular method for topology management, especially in WSNs.
which consume or generate data of a specific application Some existing approaches for WSN clustering have the po-
in a cluster can result in a higher chance of performing tential to be applied to IoT networks, even though special
data aggregation, fusion, service migration efficiency, etc. characteristics of IoT networks such as heterogeneity and
Finding an optimal trade-off between reducing interference nodes mobility make clustering a challenging issue in IoT. In
and establishing clusters based on common applications and this paper, we conducted a comprehensive survey on existing
services is a challenge that need to addressed. WSN clustering techniques and investigated their applicability
• Scalability: with thousands of nodes in 5G-IoT networks, to IoT networks. The survey work is conducted based on
scalability is a great challenge. As 5G-IoT networks can all the common clustering objectives (e.g., reducing energy
be a UDN [345], the number of nodes and communication consumption and load balancing), as well as the network
links per area are very high. As the radio range of the UEs properties impacting efficiency of IoT clustering (e.g., mo-
should be limited to improve resource consumption and also bility). Classifying the well-known clustering literature shows
reduce interference, the number of nodes per cluster and that clustering can not only reduce energy consumption as
number of clusters per network would increase which cause its primary objective, but also achieve several other quality-
more complexity of establishing clusters and connecting related objectives. In addition, our investigations show that
them to BSs. In addition, in 5G networks, there are several existing clustering techniques can contribute to better support
gateways in the network, basically BSs, which make the of quality-related requirements of ad-hoc IoT networks, e.g.,
topology management more complicated. In massive-scale QoS and fault tolerance, although their high dynamicity and
IoT applications, having multiple potential destinations, heterogeneity make clustering challenging. Moreover, we fo-
makes finding optimal routes to clusters and balancing the cused on the modern networking and computing paradigms in-
load of network efficiently a complex task. tegrated with IoT (i.e., Blockchain, SDN, NFV, Fog/Edge, and
• Dynamicity: Network dynamicity is a significant challenge 5G) and reviewed existing clustering-based approaches. Our
in 5G networks due to the presence of mobile nodes, faulty thorough study shows that clustering can greatly contribute
nodes, etc. For example, some nodes are moved by humans in better handling of design concerns related to scalability
as attached nodes like vehicles, cellphones, wearable sen- in these computing areas, from distributed control to service
sors, etc. Routes, clusters and connections which transfer orchestration and resource allocation.
data through the mobile nodes need to be re-established
after considerable network topology changes which cause ACKNOWLEDGMENT
high overhead and time complexity of re-clustering tech- Research leading to these results has partially received
niques. Having a dynamic network infrastructure in 5G IoT funding from the EU ECSEL Joint Undertaking under grant
networks, new challenges emerge like channel impairment agreement no 737459 (project Productive4.0) and from the
which should be considered in the design of a dynamic Research Council of Norway

R EFERENCES [27] Leila Shooshtarian et al. A clustering-based approach to efficient

resource allocation in fog computing. In International Symposium on
[1] Jie Lin et al. A survey on internet of things: Architecture, enabling Pervasive Systems, Algorithms and Networks, pages 207–224. Springer,
technologies, security and privacy, and applications. IEEE Internet of 2019.
Things Journal, 4(5):1125–1142, 2017. [28] Ichraf El Haj Hamad et al. BTRMC, a bio-inspired trust and reputation
[2] Mung Chiang and Tao Zhang. Fog and iot: An overview of research model using clustering in WSNs. 2017 Int. Conf. on Smart, Monitored
opportunities. IEEE Internet of Things Journal, 3(6):854–864, 2016. and Controlled Cities, SM2C 2017, 2017.
[3] Tie Qiu et al. How can heterogeneous internet of things build [29] Sameh Ben Fredj et al. A scalable IoT service search based on
our future: A survey. IEEE Communications Surveys & Tutorials, clustering and aggregation. In IEEE Int. Conf. on Green Computing
20(3):2011–2027, 2018. and Communications and IEEE Internet of Things and IEEE Cyber,
Physical and Social Computing, pages 403–410. IEEE, 2013.
[4] Raja Haseeb Javed et al. ApproxCT: Approximate Clustering Tech-
niques for Energy Efficient Computer Vision in Cyber-Physical Sys- [30] Neeraj Kumar et al. An intelligent clustering scheme for distributed
tems. In 12th Int. Conf. on Open Source Systems and Technologies intrusion detection in vehicular cloud computing. Cluster Computing,
(ICOSST), pages 64–70. IEEE, 2018. 18(3), 2015.
[31] Sen Zhou et al. Device clustering for fault monitoring in Internet of
[5] Paolo Bellavista et al. Convergence of manet and wsn in iot urban
Things systems. IEEE World Forum on Internet of Things, WF-IoT
scenarios. IEEE Sensors Journal, 13(10):3558–3567, 2013.
2015 - Proc., 2015.
[6] Oladayo Bello et al. Intelligent device-to-device communication in the
[32] Lingjuan Lyu et al. Fog-empowered anomaly detection in IoT using
internet of things. IEEE Systems Journal, 10(3):1172–1182, 2014.
hyperellipsoidal clustering. IEEE Internet of Things Journal, 4(5),
[7] Muhammad Mahtab Alam et al. A survey on the roles of communi-
2017.
cation technologies in iot-based personalized healthcare applications.
[33] Hanh Le et al. S-Web: An efficient and self-organizing wireless sensor
IEEE Access, 6:36611–36631, 2018.
network model. In Int. Conf. on Network-Based Information Systems,
[8] Muhammad Khalil Afzal et al. Ieee access special section editorial: pages 179–188. Springer, 2008.
The new era of smart cities: Sensors, communication technologies, and
[34] Z. Zhang et al. Energy-efficient multihop polling in clusters of
applications. IEEE Access, 5:27836–27840, 2017.
two-layered heterogeneous sensor networks. IEEE Transactions on
[9] Armir Bujari et al. Would current ad-hoc routing protocols be adequate Computers, 57(2):231–245, 2008.
for the internet of vehicles? a comparative study. IEEE Internet of
[35] D. Jia et al. Dynamic cluster head selection method for wireless sensor
Things Journal, 5(5):3683–3691, 2018.
network. IEEE Sensors Journal, 16(8):2746–2754, 2016.
[10] Frank T Johnsen et al. Application of iot in military operations in a [36] M. Gavhale et al. Survey on algorithms for efficient cluster formation
smart city. In 2018 International Conference on Military Communica- and cluster head selection in manet. Procedia Computer Science,
tions and Information Systems (ICMCIS), pages 1–8. IEEE, 2018. 78:477–482, 2016.
[11] Daniel G Reina et al. The role of ad hoc networks in the internet [37] A. Shahraki et al. A new approach for energy and delay trade-off
of things: A case scenario for smart environments. In Internet of intra-clustering routing in WSNs. Computers & Mathematics with
things and inter-cooperative computational technologies for collective Applications, 62(4):1670–1676, 2011.
intelligence, pages 89–113. Springer, 2013. [38] Marjan Kuchaki Rafsanjani, Atieh Rezaei, Amin Shahraki, and Ar-
[12] Wendi R. Heinzelman et al. Energy-efficient communication protocol sham Borumand Saeid. Qarima: A new approach to prediction in queue
for wireless microsensor networks. In System sciences, 2000. Proc. of theory. Applied Mathematics and Computation, 244:514–525, 2014.
the 33rd annual Hawaii Int. conference on, pages 1–10. IEEE, 2000. [39] E. Stuntebeck et al. Wireless underground sensor networks using
[13] Ossama Younis and Sonia Fahmy. Heed: a hybrid, energy-efficient, commodity terrestrial motes. In Wireless Mesh Networks, 2006.
distributed clustering approach for ad hoc sensor networks. IEEE WiMesh 2006. 2nd IEEE Workshop on, pages 112–114. IEEE, 2006.
Transactions on mobile computing, 3(4):366–379, 2004. [40] Peter J Bennett et al. Wireless sensor networks for underground railway
[14] A. Manjeshwar et al. Teen: Arouting protocol for enhanced efficiency applications: case studies in Prague and London. Smart Structures and
in wireless sensor networks. In ipdps, pages 1–7. IEEE, 2001. Systems, 6(5-6):619–639, 2010.
[15] Ali Sh. Rostami et al. Survey on clustering in heterogeneous and ho- [41] C. Uribe et al. Radio communication model for underwater WSN. In
mogeneous wireless sensor networks. The Journal of Supercomputing, New Technologies, Mobility and Security, 3rd Int. Conf. on, pages 1–5.
74(1):277323, 2018. IEEE, 2009.
[16] M. M. Afsar et al. Clustering in sensor networks: A literature survey. [42] A. Sharif et al. Wireless multimedia sensor network technology: A
Journal of Network and Computer Applications, 46:198–226, 2014. survey. In Industrial informatics, 7th IEEE Int. conference on, pages
[17] Xuxun Liu. A Survey on Clustering Routing Protocols in Wireless 606–613. IEEE, 2009.
Sensor Networks. Sensors, 12(8):11113–11153, 2012. [43] Feng Xia. QoS challenges and opportunities in wireless sensor/actuator
[18] S. Arjunan et al. A survey on unequal clustering protocols in Wireless networks. Sensors, 8(2):1099–1110, 2008.
Sensor Networks. Journal of King Saud University - Computer and [44] A. A. Abbasi et al. A survey on clustering algorithms for wireless
Information Sciences, 31(3):304–317, 2017. sensor networks. Computer communications, 30(14-15):2826–2841,
[19] L. Xu. A Survey of Clustering Techniques in WSNs and Consideration 2007.
of the Challenges of Applying Such to 5G IoT Scenarios. IEEE Internet [45] Taehyeun Park et al. Learning how to communicate in the internet of
of Things Journal, 4(5):1229–1249, 2017. things: Finite resources and heterogeneity. IEEE Access, 4:7063–7073,
[20] M. Demirbas et al. FLOC: A Fast Local Clustering Service for Wireless 2016.
Sensor Networks. First Int. Conf. on Broadband Networks, BroadNets [46] Amin Shahraki, Hamed Taherzadeh, and Øystein Haugen. Last sig-
2004, pages 1–6, 2004. nificant trend change detection method for offline poisson distribution
[21] S. K. Singh et al. A survey on successors of LEACH protocol. IEEE datasets. In 2017 International Symposium on Networks, Computers
Access, 5:4298–4328, 2017. and Communications (ISNCC), pages 1–7. IEEE, 2017.
[22] Shio Kumar Singh et al. A survey of energy-efficient hierarchical [47] Amin Shahraki et al. A comparative node evaluation model for
cluster-based routing in wireless sensor networks. International Journal highly heterogeneous massive-scale internet of things-mist networks.
of Advanced Networking and Application (IJANA), 2(02):570–580, Transactions on Emerging Telecommunications Technologies, 2020.
2010. [48] Amin Shahraki, Marius Geitle, and Øystein Haugen. A distributed fog
[23] M. Younis et al. Topology management techniques for tolerating node node assessment model by using fuzzy rules learned by xgboost. In
failures in wireless sensor networks: A survey. Computer Networks, 2019 4th International Conference on Smart and Sustainable Technolo-
58:254–283, 2014. gies (SpliTech), pages 1–6. IEEE, 2019.
[24] S. M. AlMheiri et al. MANETs and VANETs clustering algorithms: A [49] Roberto Morabito et al. Evaluating performance of containerized iot
survey. In IEEE 8th GCC Conf. & Exhibition, pages 1–6. IEEE, 2015. services for clustered devices at the network edge. IEEE Internet of
[25] C. Cooper et al. A comparative survey of VANET clustering tech- Things Journal, 4(4):1019–1030, 2017.
niques. IEEE Communications Surveys & Tutorials, 19(1):657 – 681, [50] Safwan M Ghaleb et al. Mobility management for iot: a sur-
2017. vey. EURASIP Journal on Wireless Communications and Networking,
[26] Dave Evans. The internet of things: How the next evolution of the 2016(1):1–25, 2016.
internet is changing everything. CISCO white paper, 1(2011):1–11, [51] Ricardo Silva et al. Mobility in wireless sensor networks–survey and
2011. proposal. Computer Communications, 52:1–20, 2014.

[52] Kentaro Yanagihara et al. EACLE: Energy-aware clustering scheme [77] Tal Anker et al. Efficient clustering for improving network performance
with transmission power control for sensor networks. Wireless Personal in wireless sensor networks. In European Conference on Wireless
Communications, 40(3):401–415, 2007. Sensor Networks, pages 221–236. Springer, 2008.
[53] D. J. Baker et al. The Architectural Organization of a Mobile [78] Jiaxi You et al. Budget-based clustering with context-awareness for
Radio Network via a Distributed Algorithm. IEEE Transactions on sensor networks. In Sixth Annual IEEE Int. Conf. on Pervasive
Communications, 29(11):1694–1701, 1981. Computing and Communications (PerCom), pages 306–311. IEEE,
[54] C. Lin et al. Adaptive clustering for mobile wireless networks. IEEE 2008.
Journal on Selected Areas in Communications, 15(7):1265–1275, 1997. [79] Francesca Cuomo et al. Topology formation in ieee 802.15. 4: cluster-
[55] R Nagpal et al. An algorithm for group formation in an amorphous tree characterization. In Sixth Annual IEEE Int. Conf. on Pervasive
computer. In 10th Int. Conf. on Parallel and Distributed Computing Computing and Communications (PerCom), pages 276–281. IEEE,
Systems, pages 1–4, 1998. 2008.
[56] S. Banerjee et al. A clustering scheme for hierarchical control in [80] R. S. Marin-Perianu et al. Cluster-based service discovery for hetero-
multi-hop wireless networks. Proc. IEEE INFOCOM 2001. Conf. on geneous wireless sensor networks. International Journal of Parallel,
Computer Communications., 2:1028–1037, 2001. Emergent and Distributed Systems, 23(4):325–346, 2008.
[57] Beongku An and Symeon Papavassiliou. A mobility-based clustering [81] Yan Jin et al. EEMC: An energy-efficient multi-level clustering algo-
approach to support mobility management and multicast routing in rithm for large-scale wireless sensor networks. Computer Networks,
mobilead-hoc wireless networks. International Journal of Network 52(3):542–562, 2008.
Management, 11(6):387–395, 2001. [82] Bencan Gong et al. Multihop routing protocol with unequal clustering
[58] S. Lindsey et al. Data gathering algorithms in sensor networks for wireless sensor networks. In ISECS Int. Colloquium on Computing,
using energy metrics. IEEE Transactions on Parallel and Distributed Communication, Control, and Management, pages 552–556. IEEE,
Systems, 13(9):924–935, 2002. 2008.
[59] K. Xu. A heterogeneous routing protocol based on a new stable [83] Sungju Lee et al. An Energy-Efficient Distributed Unequal Clustering
clustering scheme. In MILCOM 2002 Proc., volume 2, pages 838– Protocol for Wireless Sensor Networks. Engineering and Technology,
843. IEEE, 2002. 2(12):2715–2719, 2008.
[60] A. Manjeshwar and D. P. Agrawal. APTEEN: A hybrid protocol for [84] Sh. Varma et al. Base station initiated dynamic routing protocol for
efficient routing and comprehensive information retrieval in wireless. In heterogeneous wireless sensor network using clustering. In Fourth Int.
Proc. 16th Int. Parallel and Distributed Processing Symposium, pages Conf. on Wireless Communication and Sensor Networks, pages 1–6.
1–8. IEEE, 2002. IEEE, 2008.
[61] H. Zhang et al. GS3: Scalable self-configuration and self-healing in [85] Y. Wang et al. An Energy Efficient and Balance Hierarchical Unequal
wireless sensor networks. Computer Networks, 43(4):459–480, 2003. Clustering Algorithm for Large Scale Sensor Networks, 2009.
[86] G. Chen et al. An unequal cluster-based routing protocol in wireless
[62] S Bandyopadhyay et al. Energy Efficient Hierarchical Clustering
sensor networks. Wireless Networks, 15:193–207, 2009.
Algorithm for Wireless Sensor Networks. In Proc. IEEE INFOCOM,
pages 1713–1723. IEEE, 2003. [87] J. Lloret et al. A cluster-based architecture to structure the topology of
parallel wireless sensor networks. Sensors, 9(12):10513–10544, 2009.
[63] H. Chan et al. Ace: An emergent algorithm for highly uniform cluster
[88] M. Youssef et al. Overlapping multihop clustering for wireless sensor
formation. In European workshop on wireless sensor networks, pages
networks. IEEE Transactions on Parallel and Distributed Systems,
1–16. springer, 2004.
20(12):1844–1856, 2009.
[64] G. Smaragdakis et al. SEP: A Stable Election Protocol for clustered
[89] Li-Qing Guo et al. Improvement on LEACH by combining adaptive
heterogeneous wireless sensor networks . in proc. of the 2nd Int.
cluster head election and two-hop transmission. In Int. Conf. on
workshop on SANPA, pages 1–11, 2004.
Machine Learning and Cybernetics (ICMLC),, pages 1678–1683. IEEE,
[65] P. Ding et al. Distributed energy-efficient hierarchical clustering for 2010.
wireless sensor networks. In International conference on distributed
[90] Samer A B Awwad et al. Cluster based routing protocol for mobile
computing in sensor systems, pages 322–339. Springer, 2005.
nodes in wireless sensor network. In Int. Symposium on Collaborative
[66] Wang Ke et al. Attribute-based clustering for information dissemination Technologies and Systems, pages 233–241. IEEE, 2009.
in wireless sensor networks. In 2005 Sec. Annual IEEE Commu- [91] D. Koutsonikolas et al. Hierarchical geographic multicast routing for
nications Society Conf. on Sensor and AdHoc Communications and wireless sensor networks. Wireless Networks, 16:449–466, 2010.
Networks, SECON 2005, volume 2005, pages 498–509. IEEE, 2005.
[92] L. Buttyán et al. Position-Based Aggregator Node Election in Wireless
[67] S. Soro and W.B. Heinzelman. Prolonging the lifetime of wireless Sensor Networks. International Journal of Distributed Sensor Net-
sensor networks via unequal clustering. In 19th IEEE Int. Parallel and works, 6(1):1–15, 2010.
Distributed Processing Symposium, pages 1–8. IEEE, 2005. [93] Ning Xu et al. Coverage and connectivity guaranteed topology
[68] H. Luo et al. Ttdd: Two-tier data dissemination in large-scale wireless control algorithm for cluster-based wireless sensor networks. Wireless
sensor networks. Wireless networks, 11:161–175, 2005. Communications and Mobile Computing, 12(1):23–32, 2012.
[69] S. D Muruganathan et al. A Centralized Energy-Effiecient Routing Pro- [94] H. Zhou et al. A novel stable selection and reliable transmission pro-
tocol for Wireless Sensor Networks. IEEE Communications Magazine, tocol for clustered heterogeneous wireless sensor networks. Computer
43(3):s8–s13, 2005. Communications, 33(15):1843–1849, 2010.
[70] Mao Ye et al. An energy efficient clustering scheme in wireless sensor [95] Y. Liu et al. Multi-layer clustering routing algorithm for wireless
networks. Ad-Hoc and Sensor Wireless Networks, 3(2-3):99–119, 2007. vehicular sensor networks. IET Communications, 4(7):810–816, 2010.
[71] Adel Youssef et al. Distributed formation of overlapping multi-hop [96] H. Bagci et al. An energy aware fuzzy unequal clustering algorithm
clusters in wireless sensor networks. In IEEE GLOBECOM, pages for wireless sensor networks. In International Conference on Fuzzy
1–6. IEEE, 2006. Systems, pages 1–8. IEEE, 2010.
[72] B. Aoun et al. Clustering in WSN with latency and energy consumption [97] Anshu Khare and Krishna M Sivalingam. On recovery of lost targets
constraints. Journal of Network and Systems Management, 14(3):415– in a cluster-based wireless sensor network. In IEEE International
439, 2006. Conference on Pervasive Computing and Communications Workshops
[73] W. Choi et al. A distributed weighted clustering algorithm for mobile ad (PERCOM Workshops), pages 208–213. IEEE, 2011.
hoc networks. In Advanced Int’l Conf. on Telecommunications and Int’l [98] Yajie Ma et al. Distributed clustering-based aggregation algorithm for
Conf. on Internet and Web Applications and Services (AICT-ICIW’06), spatial correlated sensor networks. IEEE Sensors Journal, 11(3):641–
pages 1–6. IEEE, 2006. 648, 2011.
[74] Sung Min Jung et al. The concentric clustering scheme for efficient [99] Sungryoul Lee et al. LUCA: An energy-efficient unequal clustering
energy consumption in the PEGASIS. In The 9th Int. Conf. on algorithm using location information for wireless sensor networks.
Advanced Communication Technology, pages 260–265. IEEE, 2007. Wireless Personal Communications, 56(4):715–731, 2011.
[75] M. Yu et al. A dynamic clustering and energy efficient routing [100] S. Chaurasiya et al. An enhanced energy-efficient protocol with static
technique for sensor networks. IEEE Transactions on Wireless Com- clustering for WSN. In Int. Conf. on Information Networking ,, pages
munications, 6(8):3069–3079, 2007. 58–63. IEEE, 2011.
[76] Y. Tian et al. Cluster–based information processing in wireless sensor [101] Jiguo Yu et al. An energy-aware distributed unequal clustering protocol
networks: an energy–aware approach. Wireless Communications and for wireless sensor networks. International Journal of Distributed
Mobile Computing, 7(7):893–907, 2007. Sensor Networks, 7(1):1–8, 2011.

[102] Jiguo Yu et al. An energy-driven unequal clustering protocol for [128] Md Azharuddin et al. Energy efficient fault tolerant clustering and
heterogeneous wireless sensor networks. Journal of Control Theory routing algorithms for wireless sensor networks. Computers and
and Applications, 9(1):133–139, 2011. Electrical Engineering, 41:177–190, 2015.
[103] W K Lai et al. Arranging cluster sizes and transmission ranges for [129] De gan Zhang et al. A new clustering routing method based on
wireless sensor networks. Information Sciences, 183(1):117–131, 2012. PECE for WSN. Eurasip Journal on Wireless Communications and
[104] Ghalib A Shah et al. Cluster based coordination and routing framework Networking, 2015(1):1–13, 2015.
for wireless sensor and actor networks. Wireless Communications and [130] R. R. Sahoo et al. A bio inspired and trust based approach for clustering
Mobile Computing, 11(8):1140–1154, 2011. in WSN. Natural Computing, 15:423–434, 2016.
[105] S. Deng et al. Mobility-based clustering protocol for wireless sensor [131] Dan Liu et al. Cluster-based energy-efficient transmission using a new
networks with mobile nodes. IET Wireless Sensor Systems, 1(1):39 – hybrid compressed sensing in wsn. In IEEE conference on computer
47, 2011. communications workshops (INFOCOM WKSHPS), pages 372–376.
[106] T. Liu et al. An energy-balancing clustering approach for gradient- IEEE, 2016.
based routing in wireless sensor networks. Computer Communications, [132] Yanqiu Huang et al. Pkf-st: A communication cost reduction scheme
35(17):2150 –2161, 2012. using spatial and temporal correlation for wireless sensor networks. In
[107] A. Wang et al. A clustering algorithm based on energy information EWSN, pages 47–52, 2016.
and cluster heads expectation for wireless sensor networks. Computers [133] Vrinda Gupta and Rajoo Pandey. An improved energy aware distributed
& Electrical Engineering, 38(3):662–671, 2012. unequal clustering protocol for heterogeneous wireless sensor net-
[108] Hoda Taheri et al. An energy-aware distributed clustering protocol works. Engineering Science and Technology, an International Journal,
in wireless sensor networks using fuzzy logic. Ad Hoc Networks, 19(2):1050–1058, 2016.
10(7):1469–1481, 2012. [134] R. Logambigai et al. Fuzzy logic based unequal clustering for wireless
[109] J. S. Lee et al. Fuzzy-logic-based clustering approach for wireless sensor networks. Wireless Networks, 22(3):945957, 2016.
sensor networks using energy predication. IEEE Sensors Journal, [135] B. Baranidharan et al. DUCF: Distributed load balancing Unequal
12(9):2891 – 2897, 2012. Clustering in wireless sensor networks using Fuzzy approach. Applied
[110] Zhu Yong and Qing Pei. A energy-efficient clustering routing algorithm Soft Computing Journal, 40:495–506, 2016.
based on distance and residual energy for Wireless Sensor Networks. [136] Srikanth Jannu and Prasanta K Jana. A grid based clustering and
Procedia Engineering, 29:1882–1888, 2012. routing algorithm for solving hot spot problem in wireless sensor
[111] A. U. R Khan et al. Clustering-based power-controlled routing for mo- networks. Wireless Networks, 22(6):1901–1916, 2016.
bile wireless sensor networks. International journal of communication
[137] P. C. Srinivasa Rao and Haider Banka. Novel chemical reaction opti-
systems, 25(4):529–542, 2012.
mization based unequal clustering and routing algorithms for wireless
[112] J. Xu et al. Improvement of leach protocol for wsn. In 9th Int. Conf. sensor networks. Wireless Networks, 23(3):759778, 2017.
on Fuzzy Systems and Knowledge Discovery, pages 2174–2177. IEEE,
[138] Nabil Sabor et al. An Unequal Multi-hop Balanced Immune Clustering
2012.
protocol for wireless sensor networks. Applied Soft Computing Journal,
[113] Song Mao et al. An improved fuzzy unequal clustering algorithm 43:372–389, 2016.
for wireless sensor network. Mobile Networks and Applications,
[139] A. Guiloufi et al. An Energy-Efficient Unequal Clustering Algorithm
18(2):206214, 2013.
Using Śierpinski Trianglef́or WSNs. Wireless Personal Communica-
[114] Huan Li et al. COCA: Constructing optimal clustering architecture
tions, 88:449465, 2016.
to maximize sensor network lifetime. Computer Communications,
36(3):256–268, 2013. [140] Hui Xia et al. Energy-Efficient Routing Algorithm Based on Unequal
Clustering and Connected Graph in Wireless Sensor Networks. In-
[115] Y. Chen et al. Distributed Clustering With Directional Antennas for
ternational Journal of Wireless Information Networks, 23(2):141150,
Wireless Sensor Networks. IEEE Sensors Journal, 13(6):2166–2180,
2016.
2013.
[116] S. Sheik Faritha Begum et al. Modified load-balanced clustering algo- [141] D S Mantri et al. Mobility and heterogeneity aware cluster-based
rithm with distributed self-organization for wireless sensor networks. data aggregation for wireless sensor network. Wireless Personal
International Journal of Pharmacy and Technology, 8(4):23349–23356, Communications, 86:975–993, 2016.
2016. [142] P. Nayak et al. A Fuzzy Logic-Based Clustering Algorithm for WSN
[117] S. S. Wang et al. LCM: A link-aware clustering mechanism for energy- to Extend the Network Lifetime. IEEE Sensors Journal, 16(1):137 –
efficient routing in wireless sensor networks. IEEE Sensors Journal, 144, 2016.
13(2):728 – 736, 2013. [143] R. Zhang et al. NDCMC: A Hybrid Data Collection Approach for
[118] D. Mahmood et al. MODLEACH: A variant of LEACH for WSNs. Large-Scale WSNs Using Mobile Element and Hierarchical Clustering.
In 8th Int. Conf. on Broadband, Wireless Computing, Communication IEEE Internet of Things Journal, 3(4):533–543, 2016.
and Applications, BWCCA, pages 158–163. IEEE, 2013. [144] Nabajyoti Mazumdar and Hari Om. Distributed fuzzy logic based
[119] B. Manzoor et al. Q-LEACH: A new routing protocol for WSNs. energy-aware and coverage preserving unequal clustering algorithm
Procedia Computer Science, 19:926–931, 2013. for wireless sensor networks. International Journal of Communication
[120] Nadeem Javaid et al. EDDEEC: Enhanced developed distributed Systems, 30(13):e3283, 2017.
energy-efficient clustering for heterogeneous wireless sensor networks. [145] PC S. Rao et al. A particle swarm optimization based energy efficient
Procedia Computer Science, 19:914–919, 2013. cluster head selection algorithm for wireless sensor networks. Wireless
[121] M. A. Jan et al. PASCCC: Priority-based application-specific conges- networks, 23:2005–2020, 2017.
tion control clustering protocol. Computer Networks, 74:92–102, 2014. [146] D. Lin et al. A game theory based energy efficient clustering routing
[122] Suraj Sharma and Sanjay Kumar Jena. Cluster based multipath routing protocol for WSNs. Wireless Networks, 23:1101–1111, 2017.
protocol for wireless sensor networks. ACM SIGCOMM Computer [147] M. Bala Krishna and M. N. Doja. Deterministic K-means secure
Communication Review, 45(2):14–20, 2015. coverage clustering with periodic authentication for wireless sensor
[123] P. Kuila et al. A novel differential evolution based clustering algorithm networks. International Journal of Communication Systems, 30(4):1–
for wireless sensor networks. Applied Soft Computing, 25:414–425, 16, 2017.
2014. [148] A. Shahraki et al. Hierarchical distributed management clustering
[124] Sachin Gajjar et al. FAMACRO: Fuzzy and ant colony optimization protocol for wireless sensor networks. Telecommunication Systems,
based MAC/routing cross-layer protocol for wireless sensor networks. 65:193–214, 2017.
Procedia Computer Science, 46:1014–1021, 2015. [149] Xiaohui Yuan et al. A Genetic Algorithm-Based, Dynamic Clustering
[125] J RejinaParvin et al. Particle swarm optimization-based clustering by Method Towards Improved WSN Longevity. Journal of Network and
preventing residual nodes in wireless sensor networks. IEEE sensors Systems Management, 25(1):21–46, 2017.
journal, 15(8):4264–4274, 2015. [150] M. Shokouhifar et al. Optimized sugeno fuzzy clustering algorithm
[126] Davood Izadi et al. An alternative clustering scheme in WSN. IEEE for wireless sensor networks. Engineering Applications of Artificial
sensors journal, 15(7):4148–4155, 2015. Intelligence, 60:16–25, 2017.
[127] Chuan Zhu et al. A Tree-Cluster-Based Data-Gathering Algorithm for [151] Jin Wang et al. Particle swarm optimization based clustering algorithm
Industrial WSNs With a Mobile Sink. IEEE Access, 3(1):381–396, with mobile sink for WSNs. Future Generation Computer Systems,
2015. 76:452–457, 2017.

[152] K. A Darabkh et al. EA-CRP: A novel energy-aware clustering and [178] Fang Liu and Tong Li. A clustering k-anonymity privacy-preserving
routing protocol in wireless sensor networks. Computers & Electrical method for wearable iot devices. Security and Communication Net-
Engineering, 72:702–718, 2017. works, 2018:1–8, 2018.
[153] PC Srinivasa Rao and Haider Banka. Energy efficient clustering [179] S M Ghaleb et al. Mobility management for IoT: a survey. EURASIP
algorithms for wireless sensor networks: novel chemical reaction Journal on Wireless Communications and Networking, 2016(1):1–25,
optimization approach. Wireless Networks, 23(2):433–452, 2017. 2016.
[154] Jin Wang et al. Energy-efficient cluster-based dynamic routes adjust- [180] Abida Sharif et al. Internet of things enabled vehicular and ad
ment approach for wireless sensor networks with mobile sinks. The hoc networks for smart city traffic monitoring and controlling: a re-
Journal of Supercomputing, 73(7):32773290, 2017. view. International Journal of Advanced Networking and Applications,
[155] A. Abuarqoub et al. Dynamic clustering and management of mobile 10(3):3833–3842, 2018.
wireless sensor networks. Computer Networks, 117:62–75, 2017. [181] Abhijit Sarma et al. Deciding handover points based on context-aware
[156] S. Singh et al. HSCA: a novel harmony search based efficient clustering load balancing in a wifi-wimax heterogeneous network environment.
in heterogeneous WSNs. Telecommunication Systems, 67:651667, IEEE Transactions on Vehicular Technology, 65(1):348–357, 2015.
2018. [182] Stefan Feirer and Thilo Sauter. Seamless handover in industrial wlan
[157] D. Agrawal. FUCA: Fuzzy-based unequal clustering algorithm to using ieee 802.11 k. In IEEE 26th international symposium on
prolong the lifetime of wireless sensor networks. International Journal industrial electronics (ISIE), pages 1234–1239. IEEE, 2017.
of Communication Systems, 31(2):e3448, 2018. [183] Afonso Oliveira and Teresa Vazão. Low-power and lossy networks
under mobility: A survey. Computer networks, 107:339–352, 2016.
[158] Melika Khabiri and Ali Ghaffari. Energy-Aware Clustering-Based
[184] Ved P Kafle et al. Internet of things standardization in itu and
Routing in Wireless Sensor Networks Using Cuckoo Optimization
prospective networking technologies. IEEE Communications Maga-
Algorithm. Wireless Personal Communications, 98(3):2473–2495,
zine, 54(9):43–49, 2016.
2018.
[185] S. Kumar et al. Using clustering approaches for response time aware
[159] John Anzola et al. A clustering WSN routing protocol based on k-d job scheduling model for internet of things (IoT). International Journal
tree algorithm. Sensors, 18(9):1–26, 2018. of Information Technology, 9:177–195, 2017.
[160] T. Gaber et al. Trust-based secure clustering in WSN-based intelligent [186] Vijay Laxmi Kalyani et al. Iot: machine to machine (m2m), device to
transportation systems. Computer Networks, 146:151–158, 2018. device (d2d) internet of everything (ioe) and human to human (h2h):
[161] K. Muthukumaran et al. An energy efficient clustering scheme future of communication. Journal of Management Engineering and
using multilevel routing for wireless sensor network. Computers and Information Technology, 2(6):17–23, 2015.
Electrical Engineering, 69:642–652, 2018. [187] J. W. Hui et al. IPv6 in low-power wireless networks. Proc. of the
[162] Peyman Neamatollahi et al. Distributed Clustering-Task Scheduling for IEEE, 98(11):1865 – 1878, 2010.
Wireless Sensor Networks Using Dynamic Hyper Round Policy. IEEE [188] Felix Jonathan Oppermann et al. A decade of wireless sensing
Transactions on Mobile Computing, 17(2):334–347, 2018. applications: Survey and taxonomy. In The Art of Wireless Sensor
[163] Niayesh Gharaei et al. Collaborative Mobile Sink Sojourn Time Networks, pages 11–50. Springer, 2014.
Optimization Scheme for Cluster-Based Wireless Sensor Networks. [189] X. Sun et al. EdgeIoT: Mobile edge computing for the Internet of
IEEE Sensors Journal, 18(16):6669 – 6676, 2018. Things. IEEE Communications Magazine, 54(12):22 – 29, 2016.
[164] S. K. Singh et al. An Energy Efficient Protocol to Mitigate Hot [190] Bhaskar P Deosarkar et al. Clusterhead selection in clustering al-
Spot Problem Using Unequal Clustering in WSN. Wireless Personal gorithms for wireless sensor networks: A survey. In Int. Conf. on
Communications, 101:799–827, 2018. Computing, Communication and Networking, pages 1–8. IEEE, 2008.
[165] S K Preeth et al. An adaptive fuzzy rule based energy efficient [191] Qingjian Ni et al. A novel cluster head selection algorithm based
clustering and immune-inspired routing protocol for WSN-assisted IoT on fuzzy clustering and particle swarm optimization. IEEE/ACM
system. Journal of Ambient Intelligence and Humanized Computing, Transactions on Computational Biology and Bioinformatics, 14(1):76–
pages 1–13, 2018. 84, 2015.
[166] J. Lin et al. Efficient fault-tolerant routing in iot wireless sensor [192] K. Jung et al. Improving adaptive cluster head selection of teen protocol
networks based on bipartite-flow graph modeling. IEEE Access, using fuzzy logic for WMSN. Multimedia Tools and Applications,
7:14022–14034, 2019. 76:1817518190, 2017.
[167] Quan Wang et al. An energy-efficient compressive sensing-based clus- [193] S. Sharma et al. Artificial neural network based cluster head selec-
tering routing protocol for wsns. IEEE Sensors Journal, 19(10):3950– tion in wireless sensor network. International Journal of Computer
3960, 2019. Applications, 119(4):34–41, 2015.
[168] Kh. A. Darabkh et al. BPA-CRP: A balanced power-aware clustering [194] K Vijayalakshmi et al. A multi objective Tabu particle swarm
and routing protocol for wireless sensor networks. Ad Hoc Networks, optimization for effective cluster head selection in WSN. Cluster
82:155–171, 2019. Computing, 22:1227512282, 2018.
[169] D. Lin. An Energy-Efficient Clustering Algorithm Combined Game [195] T Shankar et al. Hybrid HSA and PSO algorithm for energy efficient
Theory and Dual-Cluster-Head Mechanism for WSNs. IEEE Access, cluster head selection in wireless sensor networks. Swarm and
7:49894 – 49905, 2019. Evolutionary Computation, 30:1–10, 2016.
[170] S. Randhawa et al. MLBC: Multi-objective Load Balancing Clustering [196] Yifeng Zhao et al. Distributed dynamic cluster-head selection and
technique in Wireless Sensor Networks. Applied Soft Computing, clustering for massive iot access in 5g networks. Applied Sciences,
74:66–89, 2019. 9(1):1–15, 2019.
[197] Qian Li et al. EnergIoT: A solution to improve network lifetime of
[171] Rashmi Sharan Sinha et al. A survey on lpwa technology: Lora and
IoT devices. Pervasive and Mobile Computing, 42:124–133, 2017.
nb-iot. Ict Express, 3(1):14–21, 2017.
[198] Jeong-Won Kim. A cluster head replacement based on threshold in
[172] Sh. Manurkar. Building IoT nodes-a flexible approach. In Proc. of the internet of things. The Journal of the Korea institute of electronic
the 17th ACM/IEEE Int. Conf. on Information Processing in Sensor communication sciences, 9(11):1241–1248, 2014.
Networks, pages 154–155, 2018. [199] Jine Tang et al. An energy efficient hierarchical clustering index tree
[173] Ilhem Souissi et al. A multi-level study of information trust models in for facilitating time-correlated region queries in the Internet of Things.
wsn-assisted iot. Computer Networks, 151:12–30, 2019. Journal of Network and Computer Applications, 40(1):1–11, 2014.
[174] Güngör Yıldırım and Yetkin Tatar. On wsn heterogeneity in iot and [200] Jing Zhang, Xin Feng, et al. A grid-based clustering algorithm via load
cpss. In Int. Conf. on Computer Science and Engineering (UBMK), analysis for industrial internet of things. IEEE Access, 6:13117–13128,
pages 1020–1024. IEEE, 2017. 2018.
[175] R. Tandon et al. Weight based clustering in wireless sensor networks. [201] V. Loscrı̀ et al. A two-levels hierarchy for low-energy adaptive
In National Conf. on Communications, pages 1–5. IEEE, 2013. clustering hierarchy (TL-LEACH). In IEEE 62nd Vehicular Technology
[176] Amir Taherkordi, Frank Eliassen, and Geir Horn. From iot big data Conf., pages 1809–1813. IEEE, 2005.
to iot big services. In Proceedings of the Symposium on Applied [202] A. Bagula et al. Service-aware clustering: An energy-efficient model
Computing, SAC ’17, page 485491, 2017. for the internet-of-things. Sensors, 16:1–24, 2015.
[177] Hitesh Mohapatra and Amiya Kumar Rath. Fault Tolerance Through [203] Nandhakumar Ramachandran et al. Sensor search using clustering
Energy Balanced Cluster Formation (EBCF) in WSN. In Smart technique in a massive iot environment. In Industry Interactive
Innovations in Communication and Computational Sciences, pages Innovations in Science, Engineering and Technology, pages 271–281.
313–321. Springer, 2019. Springer, 2018.

[204] Tao Han et al. A novel cluster head selection technique for edge- [228] Mohammed Zaki Hasan and Fadi Al-Turjman. Optimizing multipath
computing based iomt systems. Computer Networks, 158:114–122, routing with guaranteed fault tolerance in internet of things. IEEE
2019. Sensors Journal, 17(19):6463–6473, 2017.
[205] Arash Bozorgchenani et al. An energy-aware offloading clustering [229] Thaha Muhammed et al. Hcdsr: A hierarchical clustered fault tolerant
approach (eaoca) in fog computing. In Int. Symposium on Wireless routing technique for iot-based smart societies. In Smart Infrastructure
Communication Systems (ISWCS), pages 390–395. IEEE, 2017. and Applications, pages 609–628. Springer, 2020.
[206] Omar Said et al. Iot-rtp and iot-rtcp: Adaptive protocols for multimedia [230] S. Chouikhi et al. A survey on fault tolerance in small and large scale
transmission over internet of things environments. IEEE access, wireless sensor networks. Computer Communications, 69:22–37, 2015.
5:16757–16773, 2017. [231] J. Chudzikiewicz et al. Fault-tolerant techniques for the Internet of
[207] Hyung-Sin Kim et al. Challenging the ipv6 routing protocol for low- Military Things. In 2nd World Forum on Internet of Things (WF-IoT),
power and lossy networks (rpl): A survey. IEEE Communications pages 496–501. IEEE, 2015.
Surveys & Tutorials, 19(4):2502–2525, 2017. [232] Rajat Subhra Chakraborty et al. Security and fault tolerance in internet
[208] Harith Kharrufa et al. Rpl-based routing protocols in iot applications: of things. Springer, 2019.
A review. IEEE Sensors Journal, 19(15):5952–5967, 2019. [233] Hala Hamadeh et al. Area, energy, and time assessment for a distributed
[209] Oana Iova et al. Rpl: The routing standard for the internet of things... tpm for distributed trust in iot clusters. Integration, 58:267–273, 2017.
or is it? IEEE Communications Magazine, 54(12):16–22, 2016. [234] Vitor Barbosa Souza et al. Proactive vs reactive failure recovery
assessment in combined Fog-to-Cloud (F2C) systems. In 22nd Int.
[210] S. Jannu et al. A grid based clustering and routing algorithm for solving
Workshop on Computer Aided Modeling and Design of Communication
hot spot problem in wireless sensor networks. Wireless Networks,
Links and Networks (CAMAD), pages 1–5. IEEE, 2017.
22(6):1901–1916, 2016.
[235] Ling Li, Shancang Li, and Shanshan Zhao. QoS-aware scheduling of
[211] A Sebastian and S Sivagurunathan. A survey on load balancing services-oriented internet of things. IEEE Transactions on Industrial
schemes in rpl based internet of things. International Journal of Informatics, 10(2):1497 – 1505, 2014.
Scientific Research in Network Security and Communication, 6(3):43– [236] Roberto Morabito et al. Evaluating performance of containerized IoT
49, 2018. services for clustered devices at the network edge. IEEE Internet of
[212] J. Li et al. A clustering based routing algorithm in IoT aware Wireless Things Journal, 4(4):1019–1030, 2017.
Mesh Networks. Sustainable Cities and Society, 40:657–666, February [237] T. López et al. Resource management in the internet of things:
2018. Clustering, synchronisation and software agents. In Architecting the
[213] Behrouz Pourghebleh and Vahideh Hayyolalam. A comprehensive and Internet of Things, pages 159–193. Springer, 2011.
systematic review of the load balancing mechanisms in the internet of [238] A. Shahraki et al. Social ethics in Internet of Things: An outline and
things. Cluster Computing, 23:641661, 2019. review. In Industrial Cyber-Physical Systems (ICPS), pages 509–516.
[214] Marjan Farahani and Akbar Ghaffarpour Rahbar. Double leveled IEEE, 2018.
unequal clustering with considering energy efficiency and load bal- [239] S. Din et al. MGR: Multi-parameter Green Reliable communication for
ancing in dense iot networks. Wireless Personal Communications, Internet of Things in 5G network. Journal of Parallel and Distributed
106(3):1183–1207, 2019. Computing, 118:34–45, 2018.
[215] Chinyang Henry Tseng. Multipath load balancing routing for internet [240] Mauro Conti et al. Remi: a reliable and secure multicast routing
of things. Journal of Sensors, 2016:1–8, 2016. protocol for iot networks. In Proceedings of the 12th Int. Conf. on
[216] S Santiago et al. Ealba: energy aware load balancing algorithm for iot Availability, Reliability and Security, pages 1–8, 2017.
networks. In Proc. of the 2018 Int. Conf. on Mechatronic Systems and [241] Liumeng Song et al. Qos-aware energy-efficient cooperative scheme
Robots, pages 46–50, 2018. for cluster-based iot systems. IEEE Systems Journal, 11(3):1447–1455,
[217] Thair A Al-Janabi et al. Optimised clustering algorithm-based cen- 2017.
tralised architecture for load balancing in iot network. In Int. Sym- [242] Nandhakumar Ramachandran et al. Delay-aware heterogeneous cluster-
posium on Wireless Communication Systems (ISWCS), pages 269–274. based data acquisition in internet of things. Computers & Electrical
IEEE, 2017. Engineering, 65:44–58, 2018.
[218] Ahmed Aziz et al. Effective algorithm for optimizing compressive [243] Xin Feng et al. An Unequal Clustering Algorithm Concerned With
sensing in IoT and periodic monitoring applications. Journal of Time-Delay for Internet of Things. IEEE Access, 6:33895 – 33909,
Network and Computer Applications, 126:12–28, 2019. 2018.
[219] Hamid Tahaei et al. The rise of traffic classification in iot networks: A [244] Faouzi Hidoussi et al. PEAL: Power Efficient and Adaptive Latency
survey. Journal of Network and Computer Applications, 154:102538, Hierarchical Routing Protocol for Cluster-Based WSN. Wireless
2020. Personal Communications, 96(4):49294945, 2017.
[220] Fatima Hussain et al. A new block-based reinforcement learning [245] Ricardo Severino et al. Dynamic cluster scheduling for cluster-tree
approach for distributed resource allocation in clustered iot networks. WSNs. SpringerPlus, 3(1):1–17, 2014.
IEEE Transactions on Vehicular Technology, 69(3):2891 – 2904, 2020. [246] M. Aslam et al. CEEC: Centralized energy efficient clustering a new
routing protocol for WSNs. In 9th Annual IEEE Communications
[221] B. H Al-Qarni et al. An efficient networking protocol for internet
Society Conf on Sensor, Mesh and Ad Hoc Communications and
of things to handle multimedia big data. Multimedia Tools and
Networks (SECON), pages 103–105. IEEE, 2012.
Applications, 78:30039–30056, 2018.
[247] Syed Bilal Shah et al. Energy and interoperable aware routing
[222] Shesha Sreenivasamurthy and Katia Obraczka. Clustering for load for throughput optimization in clustered iot-wireless sensor networks.
balancing and energy efficiency in iot applications. In IEEE 26th Int. Future Generation Computer Systems, 81:372–381, 2018.
Symposium on Modeling, Analysis, and Simulation of Computer and [248] Rowayda A Sadek. Hybrid energy aware clustered protocol for iot
Telecommunication Systems (MASCOTS), pages 319–332. IEEE, 2018. heterogeneous network. Future Computing and Informatics Journal,
[223] S. Halder et al. LiMCA: an optimal clustering algorithm for lifetime 3(2):166–177, 2018.
maximization of internet of things. Wireless Networks, 25:4459–4477, [249] Dnyaneshwar Mantri et al. MHBCDA: Mobility and heterogeneity
2018. aware bandwidth efficient cluster based data aggregation for wireless
[224] Feyza Yildirim Okay and Suat Ozdemir. Routing in fog-enabled iot sensor network. In Wireless VITAE, pages 1–5. IEEE, 2013.
platforms: A survey and an sdn-based solution. IEEE Internet of Things [250] Muwonge S Bernard et al. Qos strategies for wireless multimedia
Journal, 5(6):4871–4889, 2018. sensor networks in the context of iot. In Int. Conf. on e-Infrastructure
[225] Chi-Sheng Shih et al. Data alignment for multiple temporal data and e-Services for Developing Countries, pages 228–253. Springer,
streams without synchronized clocks on iot fusion gateway. In IEEE 2018.
Int. Conf. on Data Science and Data Intensive Systems, pages 667–674. [251] Zeyu Sun et al. An optimized clustering communication protocol based
IEEE, 2015. on intelligent computing in information-centric internet of things. IEEE
[226] F. Jalali et al. Cognitive iot gateways: automatic task sharing and Access, 7:28238–28249, 2019.
switching between cloud and edge/fog computing. In Proceedings of [252] Krzysztof Nowicki and Tadeus Uhl. Qos/qoe in the heterogeneous
the SIGCOMM Posters and Demos, pages 121–123. ACM, 2017. internet of things (iot). In Beyond the Internet of Things, pages 165–
[227] K Sneha and BG Prasad. Load balanced efficient routing technique 196. Springer, 2017.
for mobility based rpl. In Int. Conf. on Smart Trends for Information [253] Pongnapat Jutadhamakorn et al. A scalable and low-cost mqtt broker
Technology and Computer Communications, pages 225–235. Springer, clustering system. In 2nd Int. Conf. on Information Technology, pages
2017. 1–5. IEEE, 2017.

[254] Sameh Ben Fredj et al. A scalable iot service search based on [279] M. Hammi et al. BCTrust: A decentralized authentication blockchain-
clustering and aggregation. In IEEE Int. Conf. on Green Computing based mechanism. In IEEE Wireless Communications and Networking
and Communications and IEEE Internet of Things and IEEE Cyber, Conf., pages 1–6, 2018.
Physical and Social Computing, pages 403–410. IEEE, 2013. [280] Wei She et al. Blockchain Trust Model for Malicious Node Detection
[255] J Sathish Kumar et al. Hierarchical clustering for dynamic and in Wireless Sensor Networks. IEEE Access, 7:38947 – 38956, 2019.
heterogeneous internet of things. Procedia Computer Science, 93:276– [281] M. Banerjee et al. Blockchain-Based Security Layer for Identification
282, 2016. and Isolation of Malicious Things in IoT: A Conceptual Design. In
[256] Mohammad Mozaffari et al. Mobile internet of things: Can uavs 27th Int. Conf. on Computer Communication and Networks, pages 1–6.
provide an energy-efficient mobile architecture? In IEEE global IEEE, 2018.
communications conference (GLOBECOM), pages 1–6. IEEE, 2016. [282] V. Sharma. An Energy-Efficient Transaction Model for the Blockchain-
[257] Jiachen Yang et al. Optimization of real-time traffic network assign- enabled Internet of Vehicles (IoV). IEEE Communications Letters,
ment based on iot data using dbn and clustering model in smart city. 23(2):246 – 249, 2018.
Future Generation Computer Systems, 108:976–986, 2017. [283] A. Dorri et al. Blockchain: A distributed solution to automotive security
[258] N. Jabeur et al. Toward a bio-inspired adaptive spatial clustering and privacy. IEEE Communications Magazine, 55(12):119 – 125, 2017.
approach for IoT applications. Future Generation Computer Systems, [284] Ali Dorri et al. Lsb: A lightweight scalable blockchain for iot security
107:736–744, 2016. and privacy. arXiv preprint arXiv:1712.02969, pages 1–17, 2017.
[259] Amir Taherkordi, Reza Mohammadi, and Frank Eliassen. A [285] Ali Dorri et al. Blockchain for IoT security and privacy: The case study
communication-efficient distributed clustering algorithm for sensor of a smart home. In IEEE Int. Conf. on Pervasive Computing and
networks. AINAW ’08, page 634638, 2008. Communications Workshops (PerCom Workshops), pages 1–6. IEEE,
2017.
[260] Haejoon Jung and In-Ho Lee. Performance analysis of three-
[286] Ali Dorri et al. Towards an optimized blockchain for IoT. In second
dimensional clustered device-to-device networks for internet of things.
Int. conf. on Internet-of-Things design and implementation, pages 173–
Wireless Communications and Mobile Computing, 2017:1–10, 2017.
178. IEEE, 2017.
[261] Muhammad Junaid Farooq et al. A multi-layer feedback system [287] Z. Hong et al. A secure routing protocol with regional partitioned
approach to resilient connectivity of remotely deployed mobile internet clustering and Beta trust management in smart home. Wireless
of things. IEEE Transactions on Cognitive Communications and Networks, 25:3805–3823, 2018.
Networking, 4(2):422–432, 2018. [288] Mohammad Javad Amiri et al. Sharper: Sharding permissioned
[262] Mona Jaber et al. 5g backhaul challenges and emerging research blockchains over network clusters. arXiv preprint arXiv:1910.00765,
directions: A survey. IEEE access, 4:1743 – 1766, 2016. pages 1–25, 2019.
[263] J. Zhang et al. A Grid-Based Clustering Algorithm via Load Analysis [289] Elli Androulaki et al. Hyperledger fabric: a distributed operating system
for Industrial Internet of Things. IEEE Access, 6:13117–13128, 2018. for permissioned blockchains. In Proceedings of the thirteenth EuroSys
[264] Chittabrata Ghosh et al. Method and apparatus for iot device clustering, conference, pages 1–15, 2018.
June 29 2017. US Patent App. 14/757,628. [290] MA Rashid and Houshyar Honar Pajooh. A security framework for iot
[265] Al-Sakib Khan Pathan. Security of self-organizing networks: MANET, authentication and authorization based on blockchain technology. In
WSN, WMN, VANET. CRC press, 2016. 18th IEEE International Conference On Trust, Security And Privacy In
[266] Paulo Gaona-Garcı́a et al. Analysis of security mechanisms based Computing And Communications/13th IEEE International Conference
on clusters iot environments. International Journal of Interactive On Big Data Science And Engineering (TrustCom/BigDataSE), pages
Multimedia & Artificial Intelligence, 4(3):55–60, 2017. 264–271. IEEE, 2019.
[267] T. Gaber et al. Trust-based secure clustering in WSN-based intelligent [291] Bacem Mbarek et al. Mbs: multilevel blockchain system for iot.
transportation systems. Computer Networks, 146:151–158, 2018. Personal and Ubiquitous Computing, pages 1–8, 2019.
[268] N. Abhishek et al. An intrusion detection system for detecting [292] Yogachandran Rahulamathavan et al. Privacy-preserving blockchain
compromised gateways in clustered iot networks. In 2018 IEEE Inter- based IoT ecosystem using attribute-based encryption. In IEEE Int.
national Workshop Technical Committee on Communications Quality Conf. on Advanced Networks and Telecommunications Systems (ANTS),
and Reliability (CQR), pages 1–6. IEEE, 2018. pages 1–6. IEEE, 2017.
[269] Chang-Sik Choi et al. Densification leveraging mobility: An iot [293] Sidra Malik et al. ProductChain: Scalable Blockchain Framework to
architecture based on mesh networking and vehicles. In Proceedings Support Provenance in Supply Chains. In IEEE 17th Int. Symposium
of the Eighteenth ACM International Symposium on Mobile Ad Hoc on Network Computing and Applications (NCA), pages 1–10. IEEE,
Networking and Computing, pages 71–80, 2018. 2018.
[270] D. Mantri et al. Two tier cluster based data aggregation (TTCDA) [294] Malavika Nair et al. The blockchain and increasing cooperative
for efficient bandwidth utilization in wireless sensor network. Wireless efficacy. Independent Review, 22(4):529–550, 2018.
Personal Communications, 75:2589–2606, 2014. [295] Loi Luu et al. Smartpool: Practical decentralized pooled mining. In
[271] Liumeng Song et al. Energy efficiency cooperative scheme for cluster- 26th {USENIX} Security Symposium, pages 1409–1426, 2017.
based capillary networks in internet of things systems. In IEEE [296] Ana Reyna et al. On blockchain and its integration with IoT. Challenges
27th Annual Int. Symposium on Personal, Indoor, and Mobile Radio and opportunities. Future Generation Computer Systems, 88:173–190,
Communications (PIMRC), pages 1–6. IEEE, 2016. 2018.
[297] Xavier Chesterman. THE P2POOL MINING POOL, 2018.
[272] Junyu Liu et al. Network densification in 5g: From the short-
[298] Open Networking Fundation. Software-defined networking: The new
range communications perspective. IEEE Communications Magazine,
norm for networks. ONF White Paper, 2, 2012.
55(12):96–102, 2017.
[299] N. McKeown et al. OpenFlow: enabling innovation in campus net-
[273] Ghulam Shabbir et al. Network performance enhancement of multi- works. ACM SIGCOMM Computer Communication Review, 38(2):69–
sink enabled low power lossy networks in sdn based internet of things. 74, 2008.
International Journal of Parallel Programming, 48:367–398, 2018. [300] NFVISG ETSI. Network functions virtualization, white paper, 2012.
[274] Tie Qiu et al. A novel shortcut addition algorithm with particle swarm [301] T. A. Al-Janabi et al. Efficient whale optimisation algorithm-based
for multisink internet of things. IEEE Transactions on Industrial SDN clustering for IoT focused on node density. In 16th Annual
Informatics, 16(5):3566 – 3577, 2019. Mediterranean Ad Hoc Networking Workshop (Med-Hoc-Net), pages
[275] Yoonyoung Sung et al. A multi-hop clustering mechanism for scalable 1–6. IEEE, 2017.
iot networks. Sensors, 18(4):1–15, 2018. [302] Ahmed Abdelaziz et al. Distributed controller clustering in software
[276] Nor Aida Mahiddin et al. An internet access solution: Manet routing defined networks. PloS one, 12(4):1–19, 2017.
and a gateway selection approach for disaster scenarios. The Review [303] Hadar Sufiev et al. Dynamic SDN Controller Load Balancing. Future
of Socionetwork Strategies, 11(1):47–64, 2017. Internet, 11(3):1–21, 2019.
[277] D. Zhaoyang et al. Blockchain: a secure, decentralized, trusted cyber [304] Fatma Al Shuhaimi et al. Software defined network as solution to
infrastructure solution for future energy systems. Journal of Modern overcome security challenges in IoT. In 5th Int. Conf. on Reliability,
Power Systems and Clean Energy, 6(5):958 – 967, 2018. Infocom Technologies and Optimization (Trends and Future Direc-
[278] Amin Shahraki, Mahmoud Abbasi, and Øystein Haugen. Boosting tions)(ICRITO), pages 491–496. IEEE, 2016.
algorithms for network intrusion detection: A comparative evaluation [305] F. Olivier, G. Carlos, and N. Florent. Sdn based architecture for
of real adaboost, gentle adaboost and modest adaboost. Engineering clustered wsn. In 2015 9th Int. Conf. on Innovative Mobile and Internet
Applications of Artificial Intelligence, 94:103770, 2020. Services in Ubiquitous Computing, pages 342–347. IEEE, 2015.

[306] Y. Nam et al. Clustered NFV service chaining optimization in mobile [333] X. Wu et al. FlashLinQ: A synchronous distributed scheduler for peer-
edge clouds. IEEE Communications Letters, 21(2):350 – 353, 2017. to-peer ad hoc networks. IEEE/ACM Transactions on Networking,
[307] Jun Wu et al. NLES: A novel lifetime extension scheme for safety- 21(4):1215 – 1228, 2013.
critical cyber-physical systems using SDN and NFV. IEEE Internet of [334] A. Kunz et al. Machine type communications in 3GPP: from release
Things Journal, 6(2):2463 – 2475, 2018. 10 to release 12. In IEEE Globecom Workshops, pages 1747–1752.
[308] A. Taherkordi and F. Eliassen. Scalable modeling of cloud-based iot IEEE, 2012.
services for smart cities. In 2016 IEEE International Conference [335] Shao-Yu Lien et al. Enhanced LTE device-to-device proximity services.
on Pervasive Computing and Communication Workshops (PerCom IEEE Communications Magazine, 54(12):174 – 182, 2016.
Workshops), pages 1–6, 2016. [336] Anas Irfan and Aqeel Khalique. A Comparative Demonstration and
[309] Rustem Dautov et al. Data agility through clustered edge computing Analysis of File Sharing Applications on Android Mobile Devices.
and stream processing. Concurrency and Computation: Practice and International Journal of Computer Applications, 156(12):52–56, 2016.
Experience, page e5093, 2018. [337] J. Harding et al. Vehicle-to-vehicle communications: readiness of v2v
[310] J. Oueis et al. The fog balancing: Load distribution for small cell cloud technology for application. Technical report, United States. National
computing. In IEEE 81st Vehicular Technology Conf., pages 1–6. IEEE, Highway Traffic Safety Administration, 2014.
2015. [338] Vincenzo Mancuso et al. Tie-breaking can maximize fairness without
[311] M. Elbamby et al. Proactive edge computing in latency-constrained fog sacrificing throughput in D2D-assisted networks. In IEEE 17th Int.
networks. In European conference on networks and communications Symposium on A World of Wireless, Mobile and Multimedia Networks
(EuCNC), pages 1–6. IEEE, 2017. (WoWMoM), pages 1–9. IEEE, 2016.
[312] Xiaoqiang Shao et al. Dynamic IoT Device Clustering and Energy [339] Selcuk Bassoy et al. Load aware self-organising user-centric dynamic
Management with Hybrid NOMA Systems. IEEE Transactions on CoMP clustering for 5G networks. IEEE Access, 4:2895 – 2906, 2016.
Industrial Informatics, 14(10):4622 – 4630, 2018. [340] B. Romanous et al. Network densification: Challenges and opportuni-
[313] R. Morabito et al. Evaluating Performance of Containerized IoT ties in enabling 5g. In IEEE 20th International Workshop on Computer
Services for Clustered Devices at the Network Edge. IEEE Internet of Aided Modelling and Design of Communication Links and Networks
Things Journal, 4(4):1019–1030, 2017. (CAMAD), pages 129–134. IEEE, 2015.
[314] Claus Pahl et al. A container-based edge cloud paas architecture based [341] Yingshuan Song et al. Intelligent access scheme for internet of things
on raspberry pi clusters. In IEEE 4th Int. Conf. on Future Internet supported by 5G wireless network. Lecture Notes of the Institute
of Things and Cloud Workshops (FiCloudW), pages 117–124. IEEE, for Computer Sciences, Social-Informatics and Telecommunications
2016. Engineering, LNICST, 237:341–351, 2018.
[315] Shancang Li et al. A distributed consensus algorithm for decision [342] L. Xu et al. A smart and balanced energy-efficient multihop clustering
making in service-oriented internet of things. IEEE Transactions on algorithm (smart-BEEM) for MIMO IoT systems in future networks.
Industrial Informatics, 10(2):1461 – 1468, 2014. Sensors, 17(7):1–23, 2017.
[316] Leila Shooshtarian, Dapeng Lan, and Amir Taherkordi. A clustering- [343] A. Asadi et al. Network-Assisted Outband D2D-Clustering in 5G
based approach to efficient resource allocation in fog computing. In Cellular Networks: Theory and Practice. IEEE Transactions on Mobile
Pervasive Systems, Algorithms and Networks, pages 207–224. Springer, Computing, 16(8):2246 – 2259, 2017.
2019. [344] X. Lin et al. The interplay between massive MIMO and underlaid
[317] G. Akpakwu et al. A survey on 5G networks for the Internet of Things: D2D networking. IEEE Transactions on Wireless Communications,
Communication technologies and challenges. IEEE Access, 6:3619 – 14(6):3337 – 3351, 2015.
3647, 2018. [345] Mahmoud Kamel et al. Ultra-dense networks: A survey. IEEE
Communications Surveys & Tutorials, 18(4):2522 – 2545, 2016.
[318] A. Gupta and R. K. Jha. A Survey of 5G Network: Architecture and
Emerging Technologies. IEEE Access, 3:1206 – 1232, 2015.
[319] Shancang Li et al. 5G Internet of Things: A survey. Journal of
Industrial Information Integration, 10(February):1–9, 2018.
[320] Waleed Ejaz et al. Internet of Things (IoT) in 5G wireless communi-
cations. IEEE Access, 4:10310 – 10314, 2016.
[321] S K Goudos et al. A survey of IoT key enabling and future
technologies: 5G, mobile IoT, sematic web and applications. Wireless
Personal Communications, 97(2):1645–1675, 2017.
[322] Fadi Al-Turjman. Agile Medium Access in Smart-Cities Vehicular-IoT.
In Smart Things and Femtocells. Taylor and Francis, 2018.
[323] Zhiguo Ding et al. A survey on non-orthogonal multiple access for
5G networks: Research challenges and future trends. IEEE Journal on
Selected Areas in Communications, 35(10):2181 – 2195, 2017.
[324] S. Talwar et al. Enabling technologies and architectures for 5G wireless.
In IEEE MTT-S Int. Microwave Symposium, pages 1–4. IEEE, 2014.
[325] A. Morgado et al. A survey of 5G technologies: Regulatory, stan-
dardization and industrial perspectives. Digital Communications and
Networks, 4(2):87–97, 2018.
[326] I Chih-Lin et al. Toward green and soft: A 5G perspective. IEEE
Communications Magazine, 52(2):66–73, 2014.
[327] C. Wang et al. Cellular architecture and key technologies for 5G
wireless communication networks. IEEE communications magazine,
52(2):122 – 130, 2014.
[328] S. A. Busari et al. Millimeter-wave massive MIMO communication
for future wireless systems: A survey. IEEE Communications Surveys
& Tutorials, 20(2):836 – 869, 2018.
[329] Irfan Ahmed et al. A survey on hybrid beamforming techniques in 5G:
Architecture and system model perspectives. IEEE Communications
Surveys & Tutorials, 20(4):3060 – 3097, 2018.
[330] X. Xia et al. A 5G-enabling technology: benefits, feasibility, and
limitations of in-band full-duplex mMIMO. IEEE Vehicular Technology
Magazine, 13(3):81 – 90, 2018.
[331] Rafay Iqbal Ansari et al. 5g d2d networks: Techniques, challenges,
and future prospects. IEEE Systems Journal, 12(4):3970–3984, 2017.
[332] V. Jungnickel et al. The role of small cells, coordinated multipoint,
and massive MIMO in 5G. IEEE Communications Magazine, 52(5):44
– 51, 2014.