Latency Minimization in Wireless IoT Using

Prioritized Channel Access and Data Aggregation

Sabin Bhandari, Shree Krishna Sharma, Xianbin Wang
Department of Electrical and Computer Engineering, Western University, London, Ontario, Canada
Email: {sbhanda7, sshar323, xianbin.wang}@uwo.ca

Abstract— Future Internet of Things (IoT) networks are ex- network latency has been considered as one of the most critical
pected to support a massive number of heterogeneous de- issues in industrial automation and control sub-systems. The
vices/sensors in diverse applications ranging from eHealthcare to main network parameters that affect the system delay are
industrial control systems. In highly-dense deployment scenarios
such as industrial IoT systems, providing reliable communication node density, data rate, and energy per node, processing
links with low-latency becomes challenging due to the involved power, routing protocol and Medium Access Control (MAC)
system delay including data acquisition and processing latencies protocol [6]. To deal with the latency issue, an IoT network
at the edge-side of IoT networks. In this regard, this paper must be designed to meet the real-time requirements of the
proposes a priority-based channel access and data aggregation aforementioned application scenarios [5].
scheme at the Cluster Head (CH) to reduce channel access and
queuing delays in a clustered industrial IoT network. First, a One of the potential approaches to reduce system delay
prioritized channel access mechanism is developed by assigning in dense wireless IoT networks is to devise a suitable MAC
different Medium Access Control (MAC) layer attributes to the protocol, which can effectively regulate the access of limited
packets coming from two types of IoT nodes, namely, high- channel resources. At the MAC layer, several factors such
priority and low-priority nodes, based on the application-specific as overhearing, over-emitting, collisions, and control packets
information provided from the cloud-center. Subsequently, a
preemptive M/G/1 queuing model is employed by using separate overhead affect the overall system delay. These factors are
low-priority and high-priority queues before sending aggregated generally related to the radio operating mode, the medium
data to the Cloud. Our results show that the proposed priority- access technique and the service time. In this context, a
based method significantly improves the system latency and number of MAC protocols have been proposed based on the
reliability as compared to the non-prioritized scheme. IEEE 802.15.4 standard [7] to address the latency issues. A
Index Terms— Internet of Things (IoT), Latency minimization,
Cloud-center, Data aggregation. MAC protocol based backoff time decision rule has been
presented in [8] for a hierarchical M2M network having
different clustered nodes. Besides, a mathematical model has
I. I NTRODUCTION
been introduced in [9] for superframe and access latency of
Internet of Things (IoT) has been emerged as a new the MAC protocol for an industrial IoT environment based
paradigm that interconnects various objects and processes for on the queuing theory. Furthermore, an extended channel
distributed real-time information collection and utilization in access mechanism namely, Explicit Prioritized Channel Access
several applications [1]. A typical IoT architecture mainly Protocol (EPCAP) [10] has been studied based on the IEEE
consists of four interconnected sub-systems, including con- 802.15.4 standard. The EPCAP proposed in [10] incorporates
nected intelligent objects/things through a sensor network, different traffic priority levels to handle critical events and
routers/gateways at the edge, backbone communication in- utilizes M/G/c based multi-server queuing network system.
frastructure, and the clouds [2]. Today’s developments in the The level of network latency can be further reduced by
Wireless Sensor Networks (WSNs), Device-to-Device (D2D), dividing incoming data packets into different queues, and sub-
Internet, Machine-to-Machine (M2M), and mobile computing sequently by employing a suitable data aggregation scheme.
technologies have a significant impact to extend the sensory The data aggregation process helps to eliminate the data
capabilities of IoT networks [3]. However, due to large-scale redundancy, to minimize the communication load, and hence
and highly-dense nature of many IoT applications, performing to reduce the overall network latency. In this regard, the
timely acquisition and analysis of IoT related data is crucial authors in [11] proposed to employ a data aggregation scheme
to support low-latency applications. to reduce the network signaling load. In addition, a tunnel
Among many potential applications, industrial IoT is con- based data aggregation method has been proposed in [12], in
sidered as a key enabler for industrial automation, intelligent which an aggregator merges the M2M data packets, appends
transportation, logistics and control systems [4]. Various appli- with its own packet, and forwards the aggregated data to the
cation requirements have brought many challenges to design gateway/base station. Besides, the authors in [13] proposed
more efficient and reliable industrial IoT networks. The main a priority based data aggregation scheme for M2M commu-
challenges in industrial IoT networks include low latency, nication over the cellular network to maintain the trade-off
low per node energy consumption, reliability, and secure data between delay requirements and power constraints by using
transmissions to the application servers [5]. Out of these, IoT a preemptive M/G/1 queuing model. However, the existing
works did not consider the joint impact of priority based
Cloud
channel access and prioritized queuing in heterogeneous IoT
networks. In addition, the potential benefit of involving the
cloud in latency reduction at the IoT edge network has not
been considered.
In this paper, a cloud-assisted priority-based channel ac- Gateway
HP: High priority
cess and data aggregation scheme is proposed for irregularly LP: Low priority
deployed sensor nodes to minimize the network latency and
HP queue HP queue
to enhance the system reliability of IoT networks. The cloud LP queue LP queue

center is equipped with massive processing power, and storage

capabilities [14], however, it does not support low-latency
applications [15, 16]. In the considered framework, the Cluster
Head (CH) extends the cloud’s functions to the edge of
: Cluster head : High priority node : Normal node
the network by prioritizing and aggregating the incoming
data packets, and the cloud-center provides various levels of
information such as priority levels and locations of the IoT Fig. 1. System model for the hierarchical IoT network.
nodes to the CH. A priority based channel access scheme is sensor nodes belonging to the same CH contend to access the
employed at the CH to reduce the channel access latency by channel to the corresponding parent node of the link. Data
assigning different MAC layer attributes to the incoming data generated from terminal nodes are transmitted to the gateway
packets. Subsequently, the prioritized data packets are sent to after data aggregation at the CH for subsequent transmission
the separate queues according to their priority levels and are to the application server. The gateway and CH are considered
aggregated before sending to the cloud via a gateway. Finally, to be positioned at the specific locations and usually have
the performance of the proposed joint prioritized channel the higher energy and computational power as compared to
access and data aggregation is analyzed using the preemptive the sensor nodes. The CH can get the application-specific
M/G/1 queuing model and compared with the conventional information such as priority levels and locations from the cloud
non-prioritized scheme. application server. In the considered system setup, the queuing
The rest of this paper is organized as follows. In Section delay for each priority class depends on the scheduling policy
II, the overall system model of a hierarchical IoT network adopted at the CH.
is presented. In Section III, the proposed prioritized channel The M/G/1 queuing method can be used to model the
access and data aggregation scheme is described in detail. stochastic behavior of a device to measure the network per-
In Section IV, the performance of the proposed method is formances such as throughput, latency, packet loss probability,
evaluated via simulations. Finally, the paper is concluded in and energy consumption [17]. The M/G/1 queuing system with
Section V. priorities can be further classified into non-preemptive and
preemptive queuing models. In the case of non-preemptive
II. S YSTEM M ODEL
priority packet scheduling, when a low priority packet 𝑃𝑙 starts
In this paper, we consider an industrial IoT scenario, com- execution, the current processing task continues even if the
posed of 𝑁 number of heterogeneous sensor nodes deployed higher priority packet 𝑃ℎ arrives at the queue. In addition, 𝑃ℎ
over an area of 𝑙 × 𝑙 𝑚2 (i.e., rectangular industrial sub-unit) packets have to wait in the queue until the task of 𝑃𝑙 packets
as shown in Fig. 1. In each industrial unit, we classify the becomes complete. However, in the preemptive priority packet
data gathered by 𝑁 sensors into two classes, i.e., normal data scheduling, the higher priority packets 𝑃ℎ are processed first
(ND) and event driven (ED) data traffic. The ND packets and can preempt the lower priority packets by saving their
are regularly generated by low priority 𝑃𝑙 nodes during some contexts, if they are already executing the task [13]. In the
process-related measurements, while the sporadic ED packets considered system, we propose to employ the preemptive
are triggered by high priority 𝑃ℎ nodes when a physical M/G/1 priority queuing model at the CH, as detailed later
quantity detected by a sensor crosses its threshold. We assume in Section III.
that each node supports only one type of data, i.e., either ND or
ED. Also, 𝑀 out of 𝑁 nodes transmit high priority packets, A. System Delay
i.e., 𝑃ℎ packets and the remaining nodes transmit only low In an IoT network, the overall system latency depends on
priority packets, i.e., 𝑃𝑙 packets. In addition, the proposed a set of parameters such as distance and the number of hops
network topology is considered to be static over the time. We towards the destination node, data rate, node density, MAC and
assume that the gateway and the cloud-center are connected via routing protocols, and the available energy and computational
high-speed wireless links with negligible latency and packet resources at the nodes. All the above-mentioned parameters
loss. may lead to unpredictable and high end-to-end latency. Out of
All the deployed sensor nodes are associated with the these, the employed MAC layer protocol determines the one-
CH/aggregator. Also, we consider that the nodes including CH hop delay and the network layer is responsible for controlling
and the gateway have the child-parent relationships. All the the multi-hop delay. The one-hop delay 𝜏ℎ𝑑 resulted at the
 Basic CSMA/CA CSMA/CA with prioritization
Channel status ...
. HP HP
.
. LP LP
High priority node ...
LP LP LP
.
.
. HP HP LP HP

Low priority node ... LP LP LP HP

.
.
.
macMinBE ℎ ℎ
High priority node ... macMaxBE
ℎ ℎ
CW
Time period ℎ ℎ

: Short backoff period : Long backoff period : Data Transmission

CSMA CSMA CSMA
: CCA detection : Busy channel : Idle channel

Fig. 2. Prioritized channel access mechanism. FIFO scheduling FIFO scheduling Priority scheduling

Fig. 3. Illustrations of the basic CSMA/CA as First-In-First-Out (FIFO)

MAC layer can be expressed in terms of different delay scheduling and the prioritized CSMA/CA.
components as follows [18]
The behavior of the CSMA/CA is affected by different
𝜏ℎ𝑑 = 𝜏𝑝𝑟𝑑 + 𝜏𝑞𝑑 + 𝜏𝑐𝑑 + 𝜏𝑡𝑑 + 𝜏𝑝𝑔𝑑 + 𝜏𝑟𝑑 , (1) MAC parameters such as the minimum backoff exponent (
macMinBE), the maximum backoff exponent (macMaxBE),
where 𝜏𝑡𝑑 , 𝜏𝑟𝑑 , and 𝜏𝑝𝑔𝑑 denote transmission, reception, and the intial value of the contention window (CW), and the max-
propagation delays, respectively, and are hardware dependent. imum number of backoffs (macMaxCSMABackoffs). Different
Similarly, 𝜏𝑝𝑟𝑑 , 𝜏𝑐𝑑 , and 𝜏𝑞𝑑 are the processing, channel access, values of these MAC parameters have a great impact on the
and queuing delays, respectively, and higher latency may result performance of an IoT network. Instead of having the same
due to the queuing of the packets and the time required to value of CSMA/CA parameters for both traffic (i.e., low prior-
access a channel. The critical MAC layer challenge for IoT ity and high priority), we can assign its own attributes for each
networks is to facilitate the channel access to an extremely class. Let us define [𝑚𝑎𝑐𝑀 𝑖𝑛𝐵𝐸ℎ , 𝑚𝑎𝑐𝑀 𝑎𝑥𝐵𝐸ℎ ] and 𝐶𝑊ℎ
large number of devices with unique traffic characteristic and as the backkoff interval and contention window values for
diverse service requirements. In this regard, this paper focuses high priority nodes, and [𝑚𝑎𝑐𝑀 𝑖𝑛𝐵𝐸𝑙 , 𝑚𝑎𝑐𝑀 𝑎𝑥𝐵𝐸𝑙 ] and
on improving queuing strategies and channel access techniques 𝐶𝑊𝑙 as the corresponding values for the low priority nodes.
to ensure Quality of Service (QoS) requirements in terms of Moreover, by specifying different CSMA/CA parameters, the
latency and reliability. priority based scheduling can be implemented to reduce the
channel access latency of the high priority packets as depicted
III. P ROPOSED P RIORITIZED C HANNEL ACCESS AND in Fig. 3 [19].
DATA AGGREGATION S CHEME
B. Data Aggregation without Prioritization
A. Prioritized Channel Access Mechanism In case of data aggregation without prioritization, the data
The data prioritization and delay modeling are performed by packets from sensor nodes arrive at the CH and are placed
the application layer by considering the MAC layer parameters on the queue. The individual packets at the CH are served
according to the requirements of industrial applications and the in different time lengths. In the considered M/G/1 queuing
network conditions. The IEEE 802.15.4 standard uses Carrier model, the data arrival pattern follows the Poisson distribution
Sense Multiple Access with Collision Avoidance (CSMA/CA) with a packet arrival rate 𝜆, and the utilization rate of the
to access the radio channel. However, CSMA/CA is not packet at the CH is given by
suitable for the delay critical industrial applications since it
𝜌 = 𝜆𝐸[𝑆], (2)
does not include the prioritization and delay responsiveness
properties [19]. In the industrial IoT systems, flow control, where 𝐸[𝑆] is the expected service time of the aggregated data
process monitoring, and fault detection sub-system must have without priority. The expected waiting time 𝐸[𝑊 ] of the non-
priority and delay aware medium access mechanisms. priority agrregated data before being served and the expected
Figure 2 shows the timing diagram of different nodes con- system delay 𝐸[𝐷𝑠𝑦𝑠 ], i.e., the total time that the packet should
tending the channel access according to their priority levels. be in the queue until being transmitted as an aggregated data
In this scenario, any packets in the low priority queue will not can be expressed as [11]
be served until the high priority queue becomes empty. The 𝜌𝐸[𝑆 2 ]
𝑃ℎ nodes always have the fixed short backoff period, more 𝐸[𝑊 ] = , (3)
2(1 − 𝜌)
frequent Common Channel Access (CCA) detection, and high
𝐸[𝐷𝑠𝑦𝑠 ] = 𝐸[𝑆] + 𝐸[𝑊 ], (4)
number of backoffs. However, the 𝑃𝑙 nodes use longer random
backoff period, less frequent detection, and lower number of where 𝐸[𝑆 2 ] is the second order moment of the service time,
backoffs. In addition, CCA detection time of 𝑃𝑙 nodes is and can be computed as follows
considered to be longer than the sum of CCA detection time 4
and the backoff periods of 𝑃ℎ nodes. 𝐸[𝑆 2 ] = 𝐸[𝑆]2 . (5)
3
 In this paper, we adapt the general mathematical model of residual service time for the packets currently being served
CSMA/CA procedure of IEEE 802.15.4 presented in [17, 20]. and the service time of the CH are denoted by 𝑅𝑖 and
Using this model, the expected service time can be expressed 𝑆𝑖 , respectively. The overall system delay is given by the
as [17] summation of the waiting time and the service time of the
packets. By using the Little’s law, the expected waiting time
𝐸[𝑆] = 𝐸[𝐷] + 𝑇𝑇 𝑥 + 2𝑇𝑡𝑢𝑟𝑛 + 𝑇𝐴𝐶𝐾 , (6)
of the 𝑖th priority packet is given by
where 𝐸[𝐷] denotes the time duration from the epoch that the ∑𝑖
data packet just arrives at the head of queue to the epoch just 𝑗=1 𝜌𝑗 𝐸[𝑅𝑗 ]
𝐸[𝑊𝑖 ] = ,
before packet transmission or discarded. The 𝑇𝑇 𝑥 and 𝑇𝐴𝐶𝐾 (1 − (𝜌1 + . . . + 𝜌𝑖 ))(1 − (𝜌1 + . . . + 𝜌𝑖−1 ))
are the transmission time of data and acknowledgment packet 𝑖 ∈ {1, 2, . . . , 𝑃 }, (10)
respectively, and 𝑇𝑡𝑢𝑟𝑛 is the turnaround time. The parameter
𝐸[𝐷] depends on the CSMA/CA procedure and is affected where 𝜌𝑖 = 𝜆𝑖 𝐸[𝑆𝑖 ], 𝐸[𝑆𝑖 ] is the expected service time, and
by different MAC parametes such as 𝐶𝑊 , 𝑚𝑎𝑐𝑀 𝑎𝑥𝐵𝐸, 𝐸[𝑅𝑖 ] represents the expected residual time. Let 𝐸[𝑆ˆ𝑖 ] and
𝑚𝑎𝑐𝑀 𝑖𝑛𝐵𝐸, and 𝑚𝑎𝑐𝑀 𝑎𝑥𝐶𝑆𝑀 𝐴𝐵𝑎𝑐𝑘𝑜𝑓 𝑓 𝑠. The ex- 𝐸[𝐷ˆ 𝑠𝑦𝑠 ] are the expected service time of 𝑖th priority packet by
𝑖
pected value 𝐸[𝐷] can be expressed as [17] considering the interruptions of higher priority packet and the
{ 𝑣 } expected system delay in the 𝑖th priority queue respectively,
∑𝑚 ∑ 𝐶𝑊𝑖 − 1 and are calculated by
𝐸[𝐷] = 𝛼𝑣 (1 − 𝛼) 𝜎 + (𝑣 + 1)𝑇𝐶𝐶𝐴
𝑣=0 𝑖=0
2 𝐸[𝑆𝑖 ]
{𝑚 } 𝐸[𝑆ˆ𝑖 ] = , (11)
∑ 𝐶𝑊𝑖 − 1 (1 − (𝜌1 + . . . + 𝜌𝑖−1 ))
𝑚+1
+𝛼 𝜎 + (𝑚 + 1)𝑇𝐶𝐶𝐴 , (7)
𝑖=0
2
ˆ 𝑠𝑦𝑠 ] = 𝐸[𝑆ˆ𝑖 ] + 𝐸[𝑊𝑖 ].
𝐸[𝐷 (12)
𝑖
where 𝑇𝐶𝐶𝐴 is the time interval for performing CCA, 𝛼 is
the busy channel probability, and 𝜎 is the length of backoff In addition, the service time 𝐸[𝑆𝑖 ] of the CH, the expected
slot. The contention{ window size for the 𝑖th retry
} is given residual time 𝐸[𝑅𝑖 ], and the second-order moment of the
by; 𝐶𝑊𝑖 = 𝑚𝑖𝑛 2𝑖 𝑚𝑎𝑐𝑀 𝑖𝑛𝐵𝐸, 𝑚𝑎𝑐𝑀 𝑎𝑥𝐵𝐸 . The de- service time 𝐸[𝑆𝑖2 ] for the priority-based data aggregation can
fault values of 𝑚𝑎𝑐𝑀 𝑖𝑛𝐵𝐸 and 𝑚𝑎𝑐𝑀 𝑎𝑥𝐵𝐸 are 3 and 5, be expressed as [13]
respectively [17]. The data packets are discarded or dropped
after 𝑚+1 attempts at CCA, and subsequently the data packet 2
𝐸[𝑅𝑖 ] = 𝜆𝑖 𝐸[𝑆𝑖 ]2 . (13)
loss rate is given by [17] 3
𝑃𝑙𝑜𝑠𝑠 = 𝛼𝑚+1 . (8) 4
𝐸[𝑆𝑖2 ] = 𝐸[𝑆𝑖 ]2 , (14)
Then, the probability of channel being busy 𝛼 can be expressed 3
in term of 𝑃𝑙𝑜𝑠𝑠 as [17] Similarly, the value of 𝐸[𝑆𝑖 ] can be calculated by using (6),
(𝑁 −1)(1−𝑃𝑙𝑜𝑠𝑠 )𝐸[𝑁𝜏 ](𝑇𝐶𝐶𝐴 +𝑇𝑇 𝑥 +2𝑇𝑡𝑢𝑟𝑛 +𝑇𝐴𝐶𝐾 ) (7), (8), and (9) in accordance with the values of 𝐶𝑊𝑖 and
𝛼= 1 , 𝜆𝑖 .
𝜆 +𝐸[𝑁𝜏 ]𝐸[𝐷]
(9)
IV. P ERFORMANCE A NALYSIS
where 𝑁 is the number of sensor nodes associated with the
CH, 𝑁𝜏 is the number of packets served in a busy period In this section, we evaluate and analyze the performance of
1
of the M/G/1 queuing system, and 𝐸[𝑁𝜏 ]= 1−𝜌 . Therefore, the proposed scheme in terms of the expected system delay
by solving the non-linear equations (7), (8), and (9), we can and system reliability. The simulation parameters are listed in
obtain the corresponding values of 𝛼, 𝑃𝑙𝑜𝑠𝑠 , and 𝐸[𝐷]. Table I [21]. We used MATLAB software in order to obtain
the results presented in this paper.
C. Data Aggregation with Prioritization
In the case of data aggregation with prioritization, the TABLE I
prioritized M/G/1 queuing model holds 𝑃 priority class of S IMULATION PARAMENTERS
data. The packets with the 𝑖th priority have arrival rate 𝜆𝑖 , Parameters Value
𝑖 ∈ {1, 2, . . . , 𝑃 }, and follow the Poisson distribution. The Max Backoff Exponent 5
lower value of 𝑖 indicates a high priority packet type. In our Min Backoff Exponent 3
Max CSMA Backoff 4
system model, we implement a preemptive priority rule, i.e., MAC Frame Payload 800 bits
the new arrival of class 𝑖th priority packet will immediately Queue Size 51 frames
preempt lower priority data currently being served and get the Data Rate 19.2 kbps
ACK Size 88 bits
access to the services. The workflow diagram of the proposed MAC Overhead 48 bits
scheme is presented in Fig. 4. 𝜎 0.32 ms
The waiting time 𝑊𝑖 of the 𝑖th priority packet is the time 𝑇𝐴𝐶𝐾 0.352 ms
𝑇𝑇 𝑥 1.12 ms
spent in the queue before being served at the CH. The mean 𝑇𝐶𝐶𝐴 0.25 ms
𝑇𝑡𝑢𝑟𝑛 0.192 ms
 Cluster head gets each node
information such as priority level
and location from cloud server

Wait for
packets

Assign Attributes: Assign Attributes:

Low High
ℎ
priority Check priority
priority ℎ
level
ℎ

Perform priority channel

access mechanism as per
Section III. A

Successful
Low High
Yes Check priority Send high and low priority priority Check if low Yes
Wait until service whether high Pause low priority
packets to corresponding priority packet
completed priority packet is
is in service
packet service
in service queues

No No

Send packet to low priority Send packet to high priority

queue for aggregation queue for aggregation

Send aggregated data

to cloud via gateway

Fig. 4. Workflow diagram of the proposed scheme.

115 130

125
110 120
Expected system delay (ms)

Expected system delay (ms)

115
105
110

105
100
100

95
95
Ph with λ = 0.4 90 Priority scheme with λ = 0.4
Ph with λ = 0.6 85
Non-priority scheme with λ = 0.4
90
Pl with λ = 1 Priority scheme with λ = 0.6
Pl with λ = 1.5 80
Non-priority scheme with λ = 0.6
85 75
10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100
Number of nodes Number of nodes
Fig. 5. Performance evaluation of the proposed priority approach in terms Fig. 6. Performance comparison of the proposed priority approach with non-
of the expected system delay. priority scheme in terms of the expected system delay.

A. Expected System Delay channel access mechanism and preemptive priority rule, the
Figure 5 presents the expected system delay of packets with high priority packets do not get any interruptions from the
different priority levels versus the number of sensor nodes. low priority packets and hence, the expected system delay is
The expected system delay of both high and low priority reduced.
packets increases as the number of node increases because
aggregation of higher number of data packets yields the longer B. Reliability
service time. The low priority packets have the longer delay The proposed scheme is modeled as the preemptive M/G/1
as compared to that of the high priority packets because priority queue with the system size 𝐾 and each queue receives
the service time must accommodate the interruptions of all data frames by following the Poisson arrival process with the
packets with the higher priority. Similarly, Fig. 6 shows the rate of 𝜆 data packets per second. The steady state probability
performance comparison of the proposed priority scheme with that 𝑖 data packets are present in the queue is given by [21]
the non-priority scheme. The non-priority scheme has a similar
characteristic curve; however, the delay is higher than the 𝜌𝑖
𝑝 𝑖 = ∑𝐾 . (15)
priority scheduling approach. Moreover, due to the prioritized 𝑗=0 𝜌𝑗
 1
the prioritized channel access and data aggregation scheme
Ph with λ = 0.4 provides substantial improvements in terms of latency and
0.9
Ph with λ = 0.6 system reliability as compared to the non-prioritized scheme.
0.8 Pl with λ = 1 In our future work, we plan to use network simulator tools
Pl with λ = 1.5
0.7 to analyze the performance of the proposed scheme in real-
Network reliability

0.6
world IoT applications such as eHealthcare and industrial
automation.
0.5

0.4 R EFERENCES
0.3
[1] L. Atzori, A. Iera, and G. Morabito, “The Internet of Things: A survey,”
Comput. Net., vol. 54, no. 15, pp. 2787–2805, 2010.
0.2 [2] M. Jutila, “An adaptive edge router enabling Internet of Things,” IEEE
Internet Things J., vol. 3, no. 6, pp. 1061–1069, Dec 2016.
0.1
[3] S. Bhandari, S. K. Sharma, and X. Wang, “Cloud-assisted device
0 clustering for lifetime prolongation in wireless IoT networks,” in Proc.
10 20 30 40 50 60 70 80 90 100 IEEE CCECE, April 2017, pp. 1–4.
Number of nodes [4] L. D. Xu, W. He, and S. Li, “Internet of Things in industries: A survey,”
Fig. 7. Performance evaluation of the proposed priority approach in terms IEEE Trans. Ind. Informat., vol. 10, no. 4, pp. 2233–2243, Nov 2014.
of network reliability. [5] S. K. Sharma, T. E. Bogale, S. Chatzinotas, X. Wang, and L. B. Le,
“Physical layer aspects of wireless IoT,” in Proc. IEEE ISWCS, Sept
The different possibilities that sensor nodes may not be able 2016, pp. 304–308.
[6] N. Nasser, L. Karim, and T. Taleb, “Dynamic multilevel priority packet
to successfully send data packets to the CH include: (i) if scheduling scheme for wireless sensor network,” IEEE Trans. Wireless
the buffer is full, (ii) if nodes fail to find the idle channel, Commun., vol. 12, no. 4, pp. 1448–1459, April 2013.
and (iii) the packets are discarded after exceeding retry limits. [7] IEEE 802.15.4, “Wireless Medium Access Control (MAC) and Physical
Layer (PHY) specifications for Low Rate Wireless Personal Area
By considering these aspects, the system reliability 𝜂 can be Networks (WPANs),” IEEE Std 802.15.4-2006, pp. 1–320, Sept 2006.
calculated as [21] [8] I. Park, D. Kim, and D. Har, “MAC achieving low latency and energy
efficiency in hierarchical M2M networks with clustered nodes,” IEEE
𝜂 = (1 − 𝑝𝑘 )(1 − 𝑃𝑐𝑓 )(1 − 𝑃𝑐𝑟 ), (16) Sensors J., vol. 15, no. 3, pp. 1657–1661, March 2015.
[9] H. Yan, Y. Zhang, Z. Pang, and L. D. Xu, “Superframe planning and
where 𝑝𝑘 is the probability of having full buffer with 𝑘 frames access latency of slotted MAC for industrial WSN in IoT environment,”
and is given by (15), 𝑃𝑐𝑓 is the probability that the packet IEEE Trans. Ind. Informat, vol. 10, no. 2, pp. 1242–1251, May 2014.
[10] S. Jardosh and P. Ranjan, “EPCAP: Explicit Prioritized Channel Access
is dropped due to the channel access failure, and 𝑃𝑐𝑟 is the Protocol for IEEE 802.15.4 based wireless sensor networks,” in Proc.
probability of packet discarded due to the retry limit. IEEE WCNC, April 2013, pp. 1–6.
Figure 7 depicts the overall system reliability versus the [11] N. Kouzayha, M. Jaber, and Z. Dawy, “M2M data aggregation over
cellular networks: signaling-delay trade-offs,” in Proc. IEEE Globecom
number of nodes. It is clearly observed that the network relia- Wkshps, Dec 2014, pp. 1155–1160.
bility decreases as the number of nodes increases. Each node in [12] G. Rigazzi, N. K. Pratas, P. Popovski, and R. Fantacci, “Aggregation
the queue begins to experience the congestion problems due to and trunking of M2M traffic via D2D connections,” in Proc. IEEE ICC,
June 2015, pp. 2973–2978.
a large number of nodes; collisions become more frequent, and [13] S. A. AlQahtani, “Analysis and modelling of power consumption-aware
the packet re-transmissions are more recurrent. Subsequently, priority-based scheduling for M2M data aggregation over long-term-
the delays get longer as the queues become busier. The prob- evolution networks,” IET Commun., vol. 11, pp. 177–184(7), Jan 2017.
[14] M. M. Hassan, H. S. Albakr, and H. Al-Dossari, “A cloud-assisted
ability of frame loss also increases because of the collisions, internet of things framework for pervasive healthcare in smart city
the retry limits, and the link constraints. Moreover, due to environment,” in Proc. EMASC. ACM, 2014, pp. 9–13.
the employed priority-based channel scheduling mechanism [15] S. K. Sharma and X. Wang, “Live data analytics with collaborative edge
and cloud processing in wireless IoT networks,” IEEE Access, vol. 5,
and queuing policy, the network reliability of the high priority pp. 4621–4635, 2017.
nodes is noted to be higher than that of the low priority nodes. [16] S. Mubeen, P. Nikolaidis, A. Didic, H. Pei-Breivold, K. Sandstrom,
and M. Behnam, “Delay mitigation in offloaded cloud controllers in
industrial IoT,” IEEE Access, vol. 5, pp. 4418–4430, 2017.
V. C ONCLUSIONS [17] T. O. Kim, J. S. Park, H. J. Chong, K. J. Kim, and B. D. Choi,
“Performance analysis of IEEE 802.15.4 non-beacon mode with the
IoT networks consist of a large number of sensor nodes unslotted CSMA/CA,” IEEE Commun. Lett., vol. 12, no. 4, pp. 238–
for different sensing and monitoring purposes. The resource- 240, April 2008.
[18] M. Doudou, D. Djenouri, and N. Badache, “Survey on latency issues
constrained IoT devices operating in highly dense networks of asynchronous MAC protocols in delay-sensitive wireless sensor
may be affected by the data collisions, packet loss, packet networks,” IEEE Commun. Surveys Tuts., vol. 15, no. 2, pp. 528–550,
delays and low network throughput. These IoT devices usually Second 2013.
[19] A. Koubaa, M. Alves, B. Nefzi, and Y.-Q. Song, “Improving the IEEE
have diverse data traffic with different latency and system 802.15.4 Slotted CSMA/CA MAC for Time-Critical Events in Wireless
reliability requirements. In this paper, we proposed a cloud- Sensor Networks,” in Proc. RTN, ECRTS Wkshps , July 2006.
assisted latency minimization scheme by using prioritized [20] P. Park, P. D. Marco, P. Soldati, C. Fischione, and K. H. Johansson, “A
generalized markov chain model for effective analysis of slotted IEEE
channel access and data aggregation at the CH. In addition, 802.15.4,” in Proc. IEEE MASS, Oct 2009, pp. 130–139.
we considered the joint impact of packet scheduling and [21] M. H. Zayani, V. Gauthier, and D. Zeghlache, “A joint model for IEEE
aggregation by using the preemptive M/G/1 queuing model. 802.15.4 physical and medium access control layers,” in Proc. IWCMC,
July 2011, pp. 814–819.
With the help of numerical results, it has been shown that

View publication stats