A Dynamic Service Placement in Fog Infrastructure

Mayssa Trabelsi

, Nadjib Mohamed Mehdi Bendaoud

and Samir Ben Ahmed

LIPSIC Laboratory, Faculty of Sciences of Tunis, University of Tunis El Manar, Tunis 2092, Tunisia

Keywords:

Internet of Things (IoT), Fog Computing, Service Placement, Real-Time Application, Quality-of-Service

(QoS), iFogSim.

Abstract:

The Internet of Things (IoT) is a key technology that improves the connectivity between applications and

devices over different geographical locations. However, IoT devices, particularly those used for monitoring,

have stringent timing requirements that Cloud Computing might not be able to satisfy. Fog computing, which

uses fog nodes close to IoT devices, can solve this problem. In this paper, we propose a Dynamic Service

Placement (DSP) algorithm for Fog infrastructures. DSP’s objective is to dynamically place the services

emitted by applications one at a time and in real time on Fog nodes. The algorithm chooses the fog node

with the least response time over the infrastructure and dynamically places the incoming service in it. The

algorithm is implemented in the iFogSim simulator, and its performances were evaluated and compared to

other algorithms. DSP showed very encouraging results, as it proceeded to minimize the average response

times and the application placement rate, thus lowering the infrastructure usage and energy consumption.

1 INTRODUCTION

The Internet of Things (IoT) has become a highly im-

portant aspect in peoples daily lives and businesses.

IoT is the network of smart devices such as sensors,

actuators, vehicles, smartphones, and cameras. These

devices are interconnected and communicated in or-

der to generate massive amounts of data that need

to be stored, processed, analyzed, and represented

to obtain valuable information meeting users’ needs

(Tavousi et al., 2022). However, most IoT devices

have limitations in terms of computing power, storage

capacity, battery power, and bandwidth. Hence, more

powerful devices are required to process and store the

services requested by an IoT device and manage the

enormous amounts of generated data, such as cloud

computing and fog computing (Tavousi et al., 2022)

(Azizi et al., 2019).

The Cloud of Things (CoT) refers to integrating

IoT with Cloud Computing, which enhances network

performance by transferring application services to

the cloud data centers to execute and store. Never-

theless, IoT applications’ geographically distributed

nature differs from the centralized nature of cloud

data centers. As IoT devices are usually far away

https://orcid.org/0000-0002-1723-9486

https://orcid.org/0000-0002-1442-6323

https://orcid.org/0000-0002-4642-2108

from the cloud data centers, high communication la-

tency and high network bandwidth consumption are

inevitable. Hence, cloud computing is not feasible

for delay-sensitive IoT applications requiring Real-

Time (RT) service. To overcome these limitations

(network bandwidth consumption, geographic distri-

bution, high service delay, and privacy-sensitive ap-

plications), Cisco proposed a new paradigm called fog

computing (Bonomi et al., 2012). The primary pur-

pose of the Fog is to extend the Cloud resources and

services at the edge of the network.

Fog computing provides computational and stor-

age resources at the network edge devices closer to the

data generation sources; thus, it provides the services

for RT IoT applications (Natesha and Guddeti, 2021).

It can support latency-critical IoT applications requir-

ing short response times. In fact, using Fog comput-

ing can allow for a drastic reduction of the overall net-

work latency (Naas et al., 2017). Furthermore, it im-

proves the quality of service (QoS), such as latency

and response time, of IoT applications, particularly

for delay-sensitive ones (Khosroabadi et al., 2021).

Any device capable of computing, storing, and con-

necting to the network can be considered a fog node

(FN), such as smart gateways, personal computers,

switches, routers, and local servers. Fog computing

faces new challenges and difﬁculties that need more

research, and attention (Azizi et al., 2019) (Khos-

444

Trabelsi, M., Bendaoud, N. and Ben Ahmed, S.

A Dynamic Service Placement in Fog Infrastructure.

DOI: 10.5220/0011852400003464

In Proceedings of the 18th International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE 2023), pages 444-452

ISBN: 978-989-758-647-7; ISSN: 2184-4895

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

roabadi et al., 2021). For example, the placement of

services is an optimization problem in fog comput-

ing in which services should be placed in the fog in-

frastructure more efﬁciently. This problem is referred

to as the Service Placement Problem (SPP) (Khos-

roabadi et al., 2021).

This paper proposes a Dynamic Service Place-

ment algorithm (DSP) in a real-time fog infrastruc-

ture. Our algorithm aims to solve the problem of dy-

namic service placement for latency-critical IoT ap-

plications in hierarchical Fog infrastructure in order

to achieve high QoS for IoT users. The algorithm is

implemented in the iFogSim simulator (Gupta et al.,

2017) and compared with random and ﬁrst-ﬁt policies

together with the iFogSim built-in cloud-only policy.

The major contribution of this paper can be sum-

marized as follows.

• A framework for placing services dynamically

and in real-time in a hierarchical fog infrastruc-

ture is developed.

• An Integer Linear Programming (ILP) formula-

tion is presented for the SPP problem subject to

constraints such as application deadlines, require-

ments, and fog nodes characteristics.

• This paper proposes an efﬁcient heuristic algo-

rithm that dynamically assigns a placement for a

service in real-time in the most suitable fog node.

The algorithm selects the best fog device that min-

imizes the response time and meets the deadline

of the processed application.

• The novelty of this work lies in the ability of the

framework to dynamically process tasks emitted

by applications one at a time and in real-time. The

algorithm proceeds to optimize the placement of

the services dynamically in the fog infrastructure

as and when a new task is emitted. The proposed

algorithm offers high performances even when the

complexity of the process increases with a large

number of delay-sensitive applications.

The remainder of the paper is organized as follows.

In Section 2, the related works are reviewed. Section

3 describes the system architecture. In Section 4, we

formulate the problem and describe our solution. We

evaluate the proposed solutions in Section 5 and con-

clude in Section 6.

2 RELATED WORK

In this section, we review and discuss some recent re-

lated works that have been proposed for solving the

problem of placing services (SPP) in the fog environ-

ment, focusing on their objective functions and solu-

tions to the SPP.

In (Khosroabadi et al., 2021), authors proposed a

heuristic algorithm, dubbed as ”a clustering of fog de-

vices and requirement-sensitive services ﬁrst” (SCAT-

TER), based on fog node clustering to solve the SPP.

This algorithm, which has promising results, is based

on QoS metrics in terms of application response time,

network usage, average application loop delays, and

energy consumption. In (Farzin et al., 2022), au-

thors proposed a ﬂexible and scalable platform called

FLEX for the SPP in multi-Fog and multi-Cloud en-

vironments. The service placement problem is for-

mulated as an optimization problem and solved by

the heuristic algorithm with the aim of delay and cost

minimization. In (Tran et al., 2019), Tran et al. pro-

posed a novel approach to task placement on fog com-

puting made efﬁcient for IoT applications that can en-

hance the performance of IoT services in terms of

response time, cost, and energy. In (Tavousi et al.,

2022), authors developed a fuzzy approach to classify

IoT applications based on their characteristics. Fur-

thermore, they proposed a heuristic algorithm to place

applications on the virtualized computing resources.

Cost and resource usage are the performance metrics

for evaluating the proposed approach. The problem

is formulated using Mixed Integer Linear Programing

(MILP). In (Azizi et al., 2019), authors introduced

an efﬁcient heuristic algorithm, called Most Delay-

sensitive Application First (MDAF), to solve the SPP

in fog-cloud computing environments. The proposed

algorithm placed the most delay-sensitive application

services closer to the IoT devices. This later work was

extended in (Hassan et al., 2020) where the authors

proposed a service placement policy, called MinRE,

to provide high QoS for IoT services and low energy

consumption for fog service providers. In addition,

they proposed two heuristic-based algorithms to solve

the problem efﬁciently. The ﬁrst one tried to provide

high QoS for critical services in terms of response

time, while the second focused on the fog environ-

ment’s energy efﬁciency. In (Nezami et al., 2021),

authors studied two optimization objectives for IoT

service placement. They formulated a decentralized

global and local load-balancing problem to minimize

the cost of deadline violation, service deployment,

and unhosted services. In (Natesha and Guddeti,

2021), authors developed a docker and containers-

based two-level fog infrastructure to provide the re-

sources. Furthermore, they formulated the service

placement problem as a multiobjective optimization

problem for minimizing service time, cost, and energy

consumption. The multiobjective problem is solved

using the Elitism-based Genetic Algorithm (EGA). In

A Dynamic Service Placement in Fog Infrastructure

445

(Herrera et al., 2021), the authors proposed a frame-

work called Umizatou that combines three optimiza-

tion problems, the Decentralized Computation Dis-

tribution Problem (DCDP), the Fog Node Placement

Problem (FNPP), and the SDN Controller Placement

Problem (CPP) that aimed at minimizing response

time. Umizatou is a holistic deployment optimiza-

tion solution based on Mixed Integer Linear Program-

ming.

In this work, we propose a dynamic service place-

ment algorithm for real-time application in a Fog In-

frastructure. The main goal of this algorithm is to

minimize the response time and meet the deadline

of the processed application. Different from state-

of-the-art approaches, the dynamic algorithm that we

propose is the better solution for service placement in

delay-sensitive IoT applications.

3 SYSTEM ARCHITECTURE

The architecture of the Fog Infrastructure system is

shown in Fig. 1. It presents a hierarchical struc-

ture. We discuss the components and how they in-

teract with each other below.

Figure 1: Fog computing architecture.

3.1 IoT Layer

The IoT devices are distributed in different geograph-

ical locations and include end-point devices such as

sensors and actuators. In IoT devices, the sensors

send their service requests to the fog gateway node,

which sends them to the fog node for further process-

ing and ﬁltering. The actuators provide the results of

service execution for users.

3.2 Fog Layer

The fog layer is the middle layer between IoT de-

vices layer and the cloud layer. Any device with

the capability of hosting application modules in the

fog layer is referred to as fog node, such as routers,

switches, local servers, and base station. These fog

nodes are computing nodes that can host virtual ma-

chines, given their resource capacity. On each virtual

machine, some applications are responsible for run-

ning the services requested by IoT devices. The fog

gateways are the fog devices that connect sensors to

the network, such as Wi-Fi access points, cellular base

stations, and home switches, which are located near

IoT devices (Gupta et al., 2017).

3.3 Cloud Layer

The cloud environment is a set of large-scale data cen-

ters and servers. IoT applications without deadline

requirements are placed and run on servers located at

cloud data centers. Although the cloud offers good

performances on high-demand applications, the com-

munication latency could be very high due to the ge-

ographical distances between the cloud and IoT de-

vices.

4 SERVICE PLACEMENT IN FOG

INFRASTRUCTURE

Our architecture is inspired by (Naas et al., 2017).

This paper formulated a data placement problem and

proposed an exact solution using integer program-

ming. The hierarchical fog architecture used in (Naas

et al., 2017) is suitable for our suggested dynamic ser-

vice placement.

In this section, we describe the proposed architec-

ture and formulate the problem of placing services in

a hierarchical fog infrastructure.

4.1 Proposed Architecture Description

Our proposed architecture is presented in Fig.2. The

sensors are responsible for sending requests to the

gateways, which will send the requests to the other

fog nodes, depending on the physical connections be-

tween the nodes. The choice of the fog node which

will receive and execute the request sent by the gate-

way devices is managed by our proposed framework.

The proposed framework consists of a heuristic algo-

rithm called Dynamic Service Placement (DSP).

From a closer perspective, the sensors collect raw

data from the environment, creating sensing tasks.

ENASE 2023 - 18th International Conference on Evaluation of Novel Approaches to Software Engineering

446

Figure 2: Proposed System Architecture.

The sensor sends the task to the fog node through

their connected gateways, to a fog node at a higher

level, or to the cloud server. The fog node sends the

results back to the actuator speciﬁc to the IoT device

which initially sent the task, as illustrated in Fig 3.

Figure 3: Sensor and Fog Node communication.

The formulation of SPP is discussed in detail

within the following subsections.

4.2 Problem Statement

Suppose the total number of all computing devices

in fog and cloud layer is m, represented by the set

F =

{

, F

, . . . , F

}

. Each Fog device F

has dif-

ferent resources, such as CPU and RAM. Therefore,

cpu

and F

ram

represent the capacity of the CPU,

measured in a million instructions per second (MIPS),

and RAM, measured in Megabytes (MB), of the Fog

device, respectively.

Let A represent a set of n IoT applications that

can be requested by IoT devices. Each applica-

tion A

∈ A consists a set of independent tasks A



i,1

, t

i,2

, . . . , t

i,|A



, where each A

has a different

number of tasks, and t

i,h

represents the h-th task of

application A

. Each task t

i,h

requires the number of

resources, such as CPU (in a million instruction per

second – MIPS) and RAM (in Megabytes – MB), that

is denoted as t

cpu

i,h

and t

ram

i,h

, respectively. Throughout

this paper, the terms service, and task are used inter-

changeably.

In this work, we attempt to place tasks on com-

putational nodes to reduce application response time.

Next, we model the objective functions using ILP.

4.3 Problem Formulation

We consider the binary decision variable x

i,h

to indi-

cate that the task t

i,h

on which fog device F

is located.

i,h











1, if task t

i,h

is placed on node F

( ∀t

i,h

, ∀F

∈ F )

0, Otherwise

(1)

Any latency-sensitive application A

should satisfy

the following Eq. (2), where RT

is the average re-

sponse time of application A

, and D

is the deadline

for A

application users determine that.

≤ D

, ∀A

∈ A (2)

The response time of application A

is deﬁned as the

time interval between the IoT application request by

the respective IoT device and the moment it receives

the result (Apat et al., 2022).

The response time of each task of A

is composed

of three parts: communication delay, processing de-

lay, and queuing delay. Therefore, the response time

i,h

of t

i,h

can be calculated as Eq. (3).

i,h

= TC

i,h

+ T E

i,h

+ T Q

i,h

, ∀t

i,h

(3)

where, TC

i,h

is the sum of the two ways communica-

tion link delays between the IoT device and the com-

putational node where the task of application A

placed. T E

i,h

is the processing execution delay for

i,h

. TQ

i,h

is the queuing delay for t

i,h

The average response time of an IoT application

can be calculated using Eq. (4). Average response

time is obtained by dividing the total response time of

an application over the number of tasks |A

∑

h=1

i,h

(4)

4.4 The Objective Function

Our main goal is to solve the service placement prob-

lem in the Fog infrastructure to minimize the response

A Dynamic Service Placement in Fog Infrastructure

447

time of the tasks, as stated by the following equation.

Min

∑

i=1

∑

h=1

i,h

(5)

Our optimization model involves the following ﬁve

constraints.

∑

i=1

∑

h=1

cpu

i,h

× x

i,h

≤ F

cpu

, ∀F

∈ F (6)

∑

i=1

∑

h=1

ram

i,h

× x

i,h

≤ F

ram

, ∀F

∈ F (7)

∑

j=1

i,h

= 1, ∀t

i,h

(8)

i,h

∈ {0, 1} (9)

≤ D

, ∀A

∈ A (10)

Constraints 6 and 7 indicate that the total CPU and

RAM of all tasks placed on each computational node

should not exceed their capacity. Constraint 8 ensures

that each task is hosted on only one computational

node. Constraint 9 speciﬁes the domain of the deci-

sion variable. Finally, constraint 10 provides that the

deadline for each application must be met.

The proposed model for the dynamic service

placement problem is an Integer Linear Programming

(ILP) problem. As a result, offering an accurate solu-

tion to this problem on a large scale is almost inappli-

cable (Salaht et al., 2020).

In the next section, we propose an efﬁcient heuris-

tic algorithm for solving this problem.

4.5 Proposed Algorithm

In this work, we describe our proposed heuristic al-

gorithm called Dynamic Service Placement (DSP) in

Fog infrastructure. Our framework takes place after

fog gateway nodes. Its purpose is to minimize the

response time for each task sent from the gateway de-

vices. The main idea of DSP is to calculate and place

in real-time the delay-sensitive application services

on fog devices with the lowest response time. This

ensures that the application deadline is met, hence sat-

isfying the global constraint in Eq. (2). Based on the

architecture Fig. 2, each IoT device (which has a sen-

sor and an actuator) is directly connected to a speciﬁc

fog gateway. Therefore, when a sensor is transmitting

a task, it is automatically placed on the corresponding

fog gateway. Then, the DSP framework runs the algo-

rithm to ﬁnd the best host for the task to be placed on.

Once a task is deployed successfully, the node uses

some of its computation power to execute it. Then it

processes the request and sends back the result to the

actuator. The pseudocode of the proposed algorithm

is presented in Algorithm 1.

Algorithm 1: Dynamic Service Placement algorithm.

Input: t

i,h

, FogDevicesList F

Output: t

i,h

−→ F

L1:=

0 , L2:=

for each F

∈ F do

if t

cpu

i,h

≤ F

cpu

&& t

mem

i,h

≤ F

ram

then

L1 := L1 ∪ F

end if

end for

if L1 ̸=

0 then

for each F

∈ L1 do

calculate RT

i,h

using eq.(3);

L2 := L2 ∪ RT

i,h

;

end for

place t

i,h

on F

∈ L2 with the minimum RT

i,h

;

end if

The DSP algorithm is explained in the following para-

graphs:

The algorithm receives as input a task t

i,h

from an

application A

and the list of the available computa-

tional nodes in the architecture F.

The process begins in line 1 by generating two

empty lists, L1 and L2. Then, for each task emitted

by a sensor to a fog gateway, the algorithm will begin

by iterating over all the computational nodes F. If a

computational node F

can satisfy the requirements of

a task t

i,h

, the node F

is added to the list L1 (line 2-6).

Afterward, if the L1 list is not empty, the algo-

rithm will iterate over L1. For each computational

node F

in L1, the algorithm calculates the response

time of task t

i,h

on that node. The response time RT

i,h

is then added to L2 (line 7-11). Finally, the minimum

response time is calculated from list L2, and task t

i,h

will be placed on the corresponding node F

5 EVALUATION

To evaluate our dynamic service placement algorithm,

it is important to examine its performance accord-

ing to the QoS criteria, such as application response

times, energy consumption, network usage, and the

number of service placements at each infrastructure

level.

The proposed DSP algorithm has been imple-

mented on the iFogSim simulator (Gupta et al., 2017).

In addition, the performance of DSP has been com-

ENASE 2023 - 18th International Conference on Evaluation of Novel Approaches to Software Engineering

448

Table 1: Characteristics of Applications.

Application name Number of instructions (MI) Memory requirements (MB) Deadline (ms) Frequency (ms)

App1 60000 256 4000 50

App2 150000 256 4500 60

App3 350000 256 5000 55

App4 180000 256 5000 70

App5 80000 256 4000 100

App6 150000 256 3500 52

App7 100000 256 4000 65

App8 300000 256 4000 150

pared with other benchmark algorithms such as First-

Fit (F.F), Random, and Cloud-Only strategies. The

First-Fit strategy (Tavousi et al., 2022) consists of

placing the service on a computational node of the

ﬁrst level of the infrastructure; this node has to meet

the requirements of the service. Otherwise, the ser-

vice is sent to a higher level until it ﬁnds a computa-

tional node that satisﬁes the service’s requirements. If

no node in the fog infrastructure can host the service,

it will be placed in a data center located in the cloud.

On the other hand, the random strategy, as its name

suggests, will randomly select a computational node

to place a service (Tavousi et al., 2022). Meanwhile,

the cloud-only strategy will have all the application

services running on cloud data centers (Azizi et al.,

2019).

5.1 Experiment Settings

The experiments were performed on a PC Intel Core

i7-2670 CPU 2.20 GHz *8, 3.7 GB RAM, and Ubuntu

20.04.5 LTS.

The simulated infrastructure is modeled accord-

ing to the architecture illustrated in Fig. 2. Eight

IoT applications are considered, and each applica-

tion has one sensor with a distinct periodic frequency

Freq

. All the applications A

have different deadlines

, they emit a task each Freq

. The required re-

sources for each application are presented in Table 1.

Moreover, There are eight fog gateways (Fog Level 0

(FL0)), as each application sensor is connected to a

fog gateway. So then, four fog devices are present in

the ﬁrst level (FL1) of the infrastructure, followed by

two in level 2 (FL2), and ﬁnally, one cloud server.

Furthermore, the computational devices in each

level of the infrastructure have their own character-

istics, RAM and CPU (see Table 2).

Table 2: Characteristics of computational devices.

Level Node Number of node CPU (MIPS) RAM (MB)

0 Fog gateway 8 250 256

1 Fog device 4 2800 1000

2 Fog device 2 4000 4000

3 Cloud server 1 44800 40000

Table 3 shows the latency in (ms) between the

different devices and nodes, such as the cloud, fog

nodes, sensors, and actuators.

Table 3: Nodes communication latencies.

Network link Latency (ms)

Fog node-cloud 3000

Fog node-Fog node 500

Fog node-Fog Gateway 50

Sensor-Fog Gateway 10

Actuator-Fog Gateway 10

5.2 Analysis of the Results

To evaluate the performance of the service place-

ment algorithm, the difference between the applica-

tion deadline and response time (D

− RT

) will be

determined. The results of algorithms performance

based on (D

− RT

) is shown in Table 4. The

First-Fit algorithm violates application A2 deadline

at 5578.68 ms, application A3 deadline at 12044.15

ms, application A4 deadline at 12262.31 ms, applica-

tion A7 deadline at 3220.99 ms, and application A8

at 4642.75 ms.

Meanwhile, the cloud-only algorithm violates

all the application deadlines in 7183.43, 7192.10,

7190.91, 7183.74, 7184.92, 7185.25, and 7123.42

ms. Also, In the random algorithm, the deadlines

for all applications are violated in 10494.98, 9290.20,

9907.91, 9515.52, 10371.82, 9317.8, and 10568.67

ms. However, no application deadline is violated in

DSP; this shows how effective the proposed algorithm

is compared to other algorithms, as it managed to suc-

cessfully meet all the application deadlines.

Fig. 4 offers a better illustration of the average

response time of all the applications and shows the

deadline curve to better measure the performance of

the algorithms.

A Dynamic Service Placement in Fog Infrastructure

449

Table 4: Algorithms performance comparison.

Average Response Time (ms) D

− RT

(ms)

DSP Cloud-Only Random F.F DSP Cloud-Only Random F.F

App1 2930.96 7183.43 10494.98 4722.38 1069.04 -3183.43 -6494.98 -722.38

App2 2899.09 7192.1 9290.20 5578.68 1600.91 -2692.1 -7591.61 -4790.2

App3 3798.41 7190.91 9907.91 12044.15 1201,59 -2190.91 -4907.91 -7044.15

App4 3785.32 7183.74 9515.52 12262.31 1352.4 -2993.12 -7029.42 -7323.56

App5 2264.64 7184.92 10371.83 3340.04 1735.36 -3184.92 -6371.83 659.96

App6 2205.33 7185.25 9317.8 3220.99 1294.67 -3685.25 -5817.8 279.01

App7 2151.98 7183.42 10568.67 4642.75 1848.02 -3183.42 -6568.67 -642.75

App8 2309.94 7203.69 9050.45 4885.57 1690.06 -3203.69 -5050.45 -885.57

Figure 4: Average Response time of applications.

Fig. 5 shows the number of services placed on

each level of the infrastructure. Overall, 1200 tasks

have been sent from sensors in the simulation. All

of the services are placed on the cloud when using

a cloud-only algorithm. Random and ﬁrst-ﬁt algo-

rithms have placed, respectively, 172 and 50 services

in the cloud. This is much higher than DSP, which

only placed one service in the cloud. Moreover, DSP

has placed 704 tasks in fog level 1 (FL1) and 495 tasks

in fog level 2 (FL2), which is higher than both ﬁrst-ﬁt

and random algorithms.

Figure 5: Number of services placed on each level of the

infrastructure (# Apps = 1200).

The observation in the previous paragraph is con-

ﬁrmed in Table 5, where a placement rate for each al-

gorithm in the different infrastructure levels is given.

We can clearly see that DSP (58.66% FL1 and

41.25% FL2) has placed 4.08% more services in the

two ﬁrst levels (FL1 and FL2) than ﬁrst-ﬁt (75% FL1

and 20.83% FL2). DSP has also placed 14.25% more

services than random (56.08% and 29.58%) in FL1

and FL2 combined. It is also clear that DSP has the

least services placed on the cloud (0.08%) compared

to random (14.33%) and ﬁrst-ﬁt (4.16%).

Table 5: Service placement rate per infrastructure level.

Placement (%)

Algorithms FL1 FL2 Cloud

DSP 58.66 41.25 0.08

Cloud-Only 0 0 100

Random 56.08 29.58 14.33

F.F 75 20.83 4.16

Fig. 6 shows clearly that none of the applications

have been placed in the cloud by DSP with the ex-

ception of Application 4, which has placed 64 tasks

in level 1, 77 tasks in level 2 and 1 task in the cloud.

The other applications are predominantly placed by

DSP in the FL1 and FL2, with a higher placement ad-

vantage for FL1 (704 applications placed) over FL2

(495 applications placed).

Figure 6: Number of services placement of each application

on each level.

ENASE 2023 - 18th International Conference on Evaluation of Novel Approaches to Software Engineering

450

Energy consumption is the sum of energy con-

sumed by fog nodes and the cloud for task placement.

Fig. 7 presents the total energy consumption for all

computational resources in DSP, cloud-only, random,

and ﬁrst-ﬁt algorithms. When using the DSP algo-

rithm, less energy is consumed by fog nodes (FL1

= 4.14 MJ and FL2 = 1.90 MJ) compared to the ﬁrst

ﬁt (FL1 = 4.22 MJ and FL2 = 1.90 MJ) and ran-

dom (FL1 = 4.09 MJ and FL2 = 2.01 MJ) algorithms.

Meanwhile, cloud-only consumption is the lowest in

both FL1 and FL2 compared to the other algorithms.

Figure 7: Energy consumption of fog devices for each algo-

rithm.

Finally, Fig. 8 illustrates how low the network us-

age is when using DSP compared to other algorithms.

This shows how efﬁcient DSP is when it comes to net-

work bandwidth and data transfer efﬁciency.

Figure 8: Total Network Usage.

6 CONCLUSION

In this paper, we have proposed a Dynamic Service

Placement (DSP) algorithm in a real-time fog infras-

tructure. We formulated the DSP problem using In-

teger Linear programming (ILP) and solved it using

a heuristic algorithm. DSP process the tasks emit-

ted by applications one at a time, dynamically and in

real-time. The framework selects all the fog nodes

that satisfy the task’s requirements and then chooses

the node that minimizes the response time. The pro-

posed algorithm performance is evaluated according

to the QoS criteria and compared to other placement

policies such as First Fit, Random, and Cloud Only.

The results show that DSP satisﬁed all the application

deadlines compared with the other algorithms; the ap-

plications are generally placed in ﬁrst and second lay-

ers, while the energy consumption and network usage

are lower than other algorithms.

REFERENCES

Apat, H. K., Nayak, R., Sahoo, B., and Mohanty, S.

(2022). Application placement in fog-enabled inter-

net of things (iot) healthcare system using body sensor

networks. In Handbook of Research on Mathematical

Modeling for Smart Healthcare Systems, pages 383–

408. IGI Global.

Azizi, S., Khosroabadi, F., and Shojafar, M. (2019). A

priority-based service placement policy for fog-cloud

computing systems. Computational Methods for Dif-

ferential Equations, 7(4 (Special Issue)):521–534.

Bonomi, F., Milito, R., Zhu, J., and Addepalli, S. (2012).

Fog computing and its role in the internet of things. In

Proceedings of the ﬁrst edition of the MCC workshop

on Mobile cloud computing, pages 13–16.

Farzin, P., Azizi, S., Shojafar, M., Rana, O., and Sing-

hal, M. (2022). Flex: a platform for scalable service

placement in multi-fog and multi-cloud environments.

In Australasian Computer Science Week 2022, pages

106–114.

Gupta, H., Vahid Dastjerdi, A., Ghosh, S. K., and Buyya,

R. (2017). ifogsim: A toolkit for modeling and

simulation of resource management techniques in

the internet of things, edge and fog computing en-

vironments. Software: Practice and Experience,

47(9):1275–1296.

Hassan, H. O., Azizi, S., and Shojafar, M. (2020). Prior-

ity, network and energy-aware placement of iot-based

application services in fog-cloud environments. IET

communications, 14(13):2117–2129.

Herrera, J. L., Gal

an-Jim

enez, J., Bellavista, P., Foschini,

L., Garcia-Alonso, J., Murillo, J. M., and Berrocal, J.

(2021). Optimal deployment of fog nodes, microser-

vices and sdn controllers in time-sensitive iot scenar-

ios. In 2021 IEEE Global Communications Confer-

ence (GLOBECOM), pages 1–6. IEEE.

Khosroabadi, F., Fotouhi-Ghazvini, F., and Fotouhi, H.

(2021). Scatter: Service placement in real-time fog-

assisted iot networks. Journal of Sensor and Actuator

Networks, 10(2):26.

A Dynamic Service Placement in Fog Infrastructure

451

Naas, M. I., Parvedy, P. R., Boukhobza, J., and Lemarchand,

L. (2017). ifogstor: an iot data placement strategy

for fog infrastructure. In 2017 IEEE 1st International

Conference on Fog and Edge Computing (ICFEC),

pages 97–104. IEEE.

Natesha, B. and Guddeti, R. M. R. (2021). Adopting

elitism-based genetic algorithm for minimizing multi-

objective problems of iot service placement in fog

computing environment. Journal of Network and

Computer Applications, 178:102972.

Nezami, Z., Zamanifar, K., Djemame, K., and Pournaras, E.

(2021). Decentralized edge-to-cloud load balancing:

Service placement for the internet of things. IEEE Ac-

cess, 9:64983–65000.

Salaht, F. A., Desprez, F., and Lebre, A. (2020). An

overview of service placement problem in fog and

edge computing. ACM Computing Surveys (CSUR),

53(3):1–35.

Tavousi, F., Azizi, S., and Ghaderzadeh, A. (2022). A fuzzy

approach for optimal placement of iot applications in

fog-cloud computing. Cluster Computing, pages 1–

18.

Tran, M.-Q., Nguyen, D. T., Le, V. A., Nguyen, D. H., and

Pham, T. V. (2019). Task placement on fog computing

made efﬁcient for iot application provision. Wireless

Communications and Mobile Computing, 2019:1–17.

ENASE 2023 - 18th International Conference on Evaluation of Novel Approaches to Software Engineering

452