Modelling Energy Consumption of IoT Devices in DISSECT-CF-Fog

Andras Markus and Attila Kertesz

Department of Software Engineering, University of Szeged, Szeged, Hungary

Keywords:

Energy Consumption, Fog Computing, Internet of Things, Simulation.

Abstract:

The continuously evolving information technology carries requirements to foster cost, resource and energy-

aware systems. The Internet of Things is considered as one of the most trending technology, which is often

coupled with Cloud or Fog Computing resources that manage the possibly big data generated by smart devices

in an effective way. To reduce the carbon footprint of such IoT-Fog-Cloud infrastructures, planning and

optimisation of their energy consumption is necessary to realise sustainable solutions. It is also inevitable to

use simulation in the design phase of such complex systems, hence it would by hardly feasible and rather

costly to evaluate numerous settings effecting the energy use. In this paper, we design an IoT energy model

based on real world measurements, and propose an extension of the energy model of the DISSECT-CF-Fog

simulator to enable the energy usage monitoring of complex, IoT-Fog-Cloud infrastructures. We also present

a validation of the extension with a weather forecasting use case to exemplify its conﬁguration possibilities

that meet the design requirements of the energy sector.

1 INTRODUCTION

The latest complex distributed systems involving

thousands of IoT devices promote widely usable ser-

vices by leveraging the computing and storing capac-

ities of cloud datacenters. To enhance the elasticity

of a concrete service, cloud resources are often aided

by resource-constrained fog nodes to improve the re-

sponse time of the IoT application and to disperse the

various types and unforeseen amount of data (Mah-

mud et al., 2018).

Besides scalability, latency and resource manage-

ment issues, energy consumption of a fog environ-

ment and the corresponding smart devices is also a

great challenge as stated in (Atlam et al., 2018), there-

fore it should be considered as one of the key fac-

tors of the development of Fog Computing solutions.

Energy-efﬁcient solutions also have a signiﬁcant im-

pact on carbon footprint and on climate change. In

order to avoid wasting energy, smart decisions could

take into account IoT device motion or correspond-

ing environmental parameters, in order to handle opti-

mally the related equipment – and such analytic eval-

uation is usually done in the cloud (Motlagh et al.,

2020).

In parallel with the spreading of Cloud Comput-

ing, the need for energy-aware systems have been in-

creased, which led to the appearance of Green Com-

puting (Garg and Buyya, 2011). The main features to

handle requests in an energy-efﬁcient way are the fol-

lowing: (i) dynamic provisioning, (ii) multi-tenancy,

(iii) server utilisation and lastly (iv) datacenter efﬁ-

ciency (requiring the usage of energy-efﬁcient tech-

nologies). An IoT system generally utilises different

types of devices, such as smart phones or microcon-

trollers, which are responsible for sensing the envi-

ronment and behave as a gateway between the hetero-

geneous system components. Microcontrollers typi-

cally have low capacity of processing unit and mem-

ory. They connect to the Internet using wireless net-

work protocols, such as Bluetooth or WiFi. It is not

negligible that the energy consumption of these de-

vices are signiﬁcantly less compared to the energy us-

age of a cloud or fog node (Samie et al., 2016).

Analysing complex IoT-Fog-Cloud systems in a

simulation environment is a common practise among

researchers, because investigating various large-scale

network topologies with thousands of IoT devices

barely feasible in real world. Though the require-

ments of such a simulator are straightforward, i.e.

to ensure detailed, realistic and ﬁne-grained model

of all entities (e.g. datacenters, fog nodes and IoT

devices), the realisation of corresponding simulators

require great efforts. The survey paper by (Markus

and Kertesz, 2020) overviews the current approaches,

revealing that they are far not complete, and energy

modelling of these systems are rarely studied.

The main contributions of this paper are: (i) de-

320

Markus, A. and Kertesz, A.

Modelling Energy Consumption of IoT Devices in DISSECT-CF-Fog.

DOI: 10.5220/0010500003200327

In Proceedings of the 11th International Conference on Cloud Computing and Services Science (CLOSER 2021), pages 320-327

ISBN: 978-989-758-510-4

signing an IoT energy model based on real world

experiments, (ii) proposing a novel extension of the

DISSECT-CF-Fog simulator (Markus et al., 2020)

for energy usage monitoring of IoT devices with this

model, and (iii) validating of the proposal with a

weather forecasting use case to exemplify the conﬁg-

uration possibilities and the use of the extended sim-

ulator.

The rest of this paper is organised as follows: in

Section 2 we brieﬂy summarise the related works, in

Section 3 we introduce our simulator extension and

the proposed energy model of microcontrollers. Sec-

tion 4 presents the evaluation with various IoT use

cases, and ﬁnally, Section 5 concludes our work.

2 RELATED WORK

The literature has numerous studies covering differ-

ent aspects of energy-aware approaches related to the

interoperation of Cloud and Fog Computing and IoT.

Since our goal is to propose a detailed energy model

for these complex systems, including both computa-

tional resources and sensor devices, we restrict our-

selves to focus on existing simulation tools.

Nevertheless the monitoring of energy consump-

tion entails signiﬁcant challenges for IoT-Fog-Cloud

systems. A research paper by (Sun et al., 2020) aims

to resolve the task ofﬂoading and resource allocation

of an IoT-Fog-Cloud architecture as a minimisation

problem of energy and cost. The authors consider

different energy consumption for a given task sep-

arately executed on IoT device, fog or cloud node.

A slightly detailed approach was proposed in (Oma

et al., 2018), where the authors differentiate computa-

tion, storing and sending processes of the data, how-

ever they applied a simple power consumption model,

which means that the energy consumption is set to the

maximum, if at least one process is executed, other-

wise the computing resource consumes minimal en-

ergy. (Ahvar et al., 2019) shows an elaborated model,

in which static and dynamic consumption compo-

nents are deﬁned, thus each of them takes into ac-

count the computational and network (i.e. routers and

switches) parts of the system considering idle power,

cores power, which is proportional to the CPU us-

age of a resource, and lastly, energy consumption of a

switch, when storing or transmitting data.

Such formal models are sufﬁcient for the analy-

sis of speciﬁc IoT and fog related problems, how-

ever considering only the extreme values of energy

consumption, which may lead to a distortion of the

results. Nevertheless, more sophisticated approaches

take into consideration the diverse types of constants

and factors. For example, the energy consumption of

a computational node is represented by the power us-

age effectiveness (PUE) coefﬁcient, which makes the

model more punctual. In general, a simulation-based

approach may conduct to a more conﬁgurable and re-

alistic model, for instance the energy consumption of

various entities (i.e. CPU, network or storage) con-

nected to state transition of those leads a more man-

ageable and extensible realisation.

Numerous studies offer and compare IoT and fog

simulations based on different properties and features

- without claiming completeness -, such as code qual-

ity, architecture, VM management, resource and en-

ergy consumption, addressed in (Perez Abreu et al.,

2020) and (Markus and Kertesz, 2020).

The CloudSim-based solutions, such as iFogSim

(Gupta et al., 2017) or EdgeCloudSim (Sonmez et al.,

2018) inherit the energy model, which deﬁnes the

consumption as follows: a static constant power

(e.g. a switched on machine) as the idle power, and

dynamic components (using linear function) as the

busy/max power are summed. IoT devices and fog

nodes are handled equally in this model.

FogNetSim++ associates the energy consumption

to certain tasks as the combination of device and fog

node consumption (Qayyum et al., 2018). The device

energy is calculated based on the transmission power

weighted by the ratio of the task size and the band-

width, whilst the fog node energy takes in consider-

ation the idle power with weight of the task size and

the computing power. This approach also considers

the residual energy of mobile nodes.

YAFS also ensures a minimal model for monitor-

ing the energy consumption (Lera et al., 2019), how-

ever it is represented by only a static consumption

value of computing and IoT entities. IoTSim-Edge

targets to model smart devices using low energy pro-

tocols (Jha et al., 2020). The authors also consider

the simulation of battery power by using a predeﬁned

drainage rate.

DISSECT-CF-Fog can be used to simulate cloud,

fog and IoT infrastructures, and their combined, hy-

brid solutions. As a result, three types of elements

can appear in it simulations: cloud datacenters, fog

nodes and IoT devices. The initial energy model of

DISSECT-CF published in 2015 (Kecskemeti, 2015)

covered cloud datacenters, by introducing resource

consumption modelling for CPU, disk and network

energy utilisation. This approach consider minimum,

idle and maximum power values as well during the

calculation of energy consumption based on linear

interpolation. As a summary, our current exten-

sion of DISSECT-CF-Fog has the most detailed en-

ergy model using dynamic consumption values for the

Modelling Energy Consumption of IoT Devices in DISSECT-CF-Fog

321

Figure 1: The utilisation of Raspberry Pi (left) and ESP32 (middle) microcontrollers and KCX-017 meter (right).

widest variety of resources (cloud, fog and IoT).

3 MODELLING ENERGY

CONSUMPTION IN

DISSECT-CF-FOG

In this paper we build on the original model of

DISSECT-CF and use its earlier proposed extension

for fog and IoT systems. To ensure the required level

of system granularity, the simulator mimics the be-

haviour of infrastructure clouds by predeﬁned states

of physical machines (PM), virtual machines (VM),

storage and disks. For instance, a PM can be in the

following states: turned off, switching on, running

and switching off. As a result, the basic concept of

energy saving can be easily realised by turning off

the unused machine. Besides, this reﬁned model sup-

ports the mapping of certain energy consumption val-

ues to the predeﬁned states, which ensures the ﬁne-

granularity of the simulator.

As we discussed earlier, the energy model takes

into account: the minimum (min) power (e.g. the ma-

chine/device is turned off, but still plugged into the

energy source), the maximum (max) power (e.g. if the

CPU is fully utilised), and the idle power (e.g. when

the PM is running without executing computational

tasks). At this moment the simulator has two power

models: (i) dynamic power draining behaviour ap-

plies linear interpolation between idle and max power

values, whilst (ii) constant power draining behaviour

can consider any power value (e.g. min). By default,

the dynamic model is applied in case of states with

high energy consumption (e.g. running state of a PM),

and it handles the idle power with min power values,

and the consumption range can be get by subtracting

the idle power value from the max power value.

In the simulator, the PhysicalMachineEner-

gyMeter class can be utilised to monitor the physi-

cal cloud resources concerning energy consumption

(introduced in DISSECT-CF-Fog in 2015), and we

proposed a fog extension (with DISSECT-CF-Fog in

2020) by introducing the ComputingAppliance class

to simulate fog nodes (possibly having additional pa-

rameters). With this extension we arrived to a uniﬁed

energy model for fogs and clouds, since the Physi-

calMachineEnergyMeter class is encapsulated in the

ComputingAppliance class, and can be used to simu-

late both fog and cloud nodes.

In this work we take a step forward, and cover IoT

devices with our proposed extended model, to enable

complex energy utilisation analysis of IoT-Fog-Cloud

systems. First, we started to analyse real power con-

sumption of microcontrollers, which is detailed in the

next subsection.

3.1 Analysis of Real Microcontrollers

In order to determine a ﬁne-grained energy model for

microcontrollers, we measured and collected energy

consumption values of real devices. In our exper-

iments we chose ESP32

(WROOM-32) and Rasp-

berry Pi

(1 Model A+) microcontrollers for further

analysis. Both devices were equipped with DTH22

temperature and humidity sensors, and a KCX-017

meter was applied to display the voltage and the cur-

rent of the connected through USB port. The assem-

bly of the used gadgets can be seen in Figure 1.

To measure the general power consumption of IoT

applications, we developed a typical and simple pro-

gram written in Python/MicroPython covering the fol-

lowing functionalities: sensor data reading (tempera-

ture and humidity values in our current case), message

creation and sending as an IoT client device by using

the MQTT protocol. We scheduled sensor value sam-

pling every minute by default, and connected the de-

vices to thee Internet via WiFi. The data application

running on the microcontrollers forwarded the sensor

The ofﬁcial website of ESP32 is available at:

https://www.espressif.com/en/products/socs/esp32/

The ofﬁcial website of Raspberry Pi (RPi) is available

at: https://www.raspberrypi.org/

CLOSER 2021 - 11th International Conference on Cloud Computing and Services Science

322

Table 1: Uniform sampling of microcontrollers.

Microcontroller ESP32 Raspberry Pi (RPi)

Sampling (min.) V I P V I P

1 5.13 0.07 0.3591 5.12 0.28 1.4336

2 5.17 0.02 0.1034 5.12 0.26 1.3312

3 5.19 0.02 0.1038 5.12 0.31 1.5872

4 5.17 0.02 0.1034 5.13 0.26 1.3338

5 5.19 0.02 0.1038 5.13 0.28 1.4364

6 5.17 0.02 0.1034 5.12 0.28 1.4336

7 5.16 0.05 0.258 5.13 0.28 1.4364

8 5.19 0.02 0.1038 5.13 0.26 1.3338

9 5.21 0.02 0.1042 5.12 0.28 1.4336

10 5.17 0.02 0.1034 5.13 0.28 1.4364

11 5.18 0.02 0.1036 5.13 0.31 1.5903

12 5.17 0.02 0.1034 5.13 0.28 1.4364

13 5.20 0.05 0.26 5.12 0.28 1.4336

14 5.21 0.02 0.1042 5.13 0.28 1.4364

15 5.18 0.02 0.1036 5.13 0.26 1.3338

Table 2: Mapping the benchmark and measured values to the model power values in DISSECT-CF-Fog.

Data Source Research Papers Websites Our Experiments IoT Energy Model

Microcontroller ESP32 RPi ESP32 RPi ESP32 RPi ESP32 RPi

Power

Cons. (W)

min N.A. N.A. 0.01 0.1 N.A. N.A. 0.01 0.1

idle 0.17 0.94 0.04 1.1 0.1 1.33 0.1 1.1

max 0.28 1.57 0.42 2.1 0.36 1.59 0.35 1.75

data to an IoT analytics platform called Thingspeak

where it could be visualised.

To determine the electric power (P measured in

watts) in the SI system, we multiplied the metered

voltage (V measured in volts) with the metered elec-

tric current (I measured in amps) values:

P = V ∗ I, e.g.1W = 1V ∗ 1A (1)

Finally, we can determine the energy usage (J mea-

sured in joules/watt-second/kilowatt-hour) by:

J = P ∗t, e.g.1J = 1W ∗ 1s (2)

3.2 The Energy Model for IoT Devices

Finally, in order to utilise monitored data of real IoT

devices in DISSECT-CF-Fog, we executed our sam-

pling application ﬁve times on both microcontrollers

for 15 minutes, while measuring the power consump-

tion each millisecond.

Table 1 presents the average values of the uniform

sampling of the metering device for each one minute

periods. Based on the monitored values, we calcu-

lated the electric power. The results show that our

The ofﬁcial website of Thingspeak is available at:

https://thingspeak.com/

typical IoT monitoring application consumed 0.1 to

0.36 W per minute in average with ESP32, and 1.3 to

1.59 W with RPi. In the next subsection we show how

we applied these measured values to our proposed IoT

energy model.

Concerning power consumption of IoT resources,

we had to build up the energy model from scratch.

In this work, we had to extend the Device class of

DISSECT-CF-Fog, which represents any smart ob-

jects, and responsible for power consumption meter-

ing for IoT devices during simulations.

In the previous version of DISSECT-CF-Fog, only

the elements of the IoT layer lacked energy meter-

ing functions. To resolve this issue, we decided to

create the Microcontroller class for implementing our

energy model of microcontrollers. Such realisation

keeps the already existing functionalities (e.g. data

sensing of IoT sensors, temporary data storing and

data forwarding to fog or cloud nodes), and introduces

predeﬁned states for microcontrollers, which allow to

map a certain power consumption to a certain state.

Besides our real measurements of a typical use

of a microcontroller, we gathered information from

the following works. (Maier et al., 2017) and (Kaup

et al., 2014) focus on the comparative analysis and

the monitoring of ESP32 and Raspberry Pi devices,

Modelling Energy Consumption of IoT Devices in DISSECT-CF-Fog

323

while detailed online benchmark results for their en-

ergy consumption can also be found on websites

4 5

After studying these sources, we The collected and

measured numbers are shown and compared in Table

2. It also shows the predeﬁned values (for min, idle

and max) we chose to be the base for our IoT En-

ergy Model. We arrived to these values by counting

the median for the concrete values gathered from the

research papers, websites and by our measurements.

Our ﬁndings and experiments revealed that the

power consumption values of microcontrollers are

highly dependant on their actual behaviour and their

use cases. Typical modifying circumstances may be

the usage of wired connection instead of wireless,

and/or different types of power supply cable or con-

verter. As we mentioned it earlier, during our exper-

iments we used an online service for retrieving and

storing the generated data by the DTH22 sensor. Ta-

ble 1 also shows that in a few cases (i.e. after ev-

ery ﬁfth sampling), the consumption doubled in case

of ESP32. To handle such extreme cases and to be

able to simulate uncertainty, we introduce three dif-

ferent states of a microcontroller in our model (that

have been implemented in the Microcontroller class

of our simulator).

The state OFF indicates a fully turned off device

with static minimal energy consumption using the

min power preset value. The RUNNING state repre-

sents a high energy consumption state, where the ac-

tual power consumption can change dynamically wrt.

the actual CPU utilisation. The minimal and maximal

consumption values in this state are set by the prede-

ﬁned idle and max power values. To simulate speciﬁc

events when high power spikes appear (caused by e.g.

activating a previously unused port of the device), we

introduce the ACTIVE state. It also represents a high

energy consumption state allowing dynamic changes,

but its minimal value should be higher than in the

RUNNING state; by default it is set to the double of

the idle power value.

According to our observation, we experienced

such behaviour in 20% of the sampling process, there-

fore we decided that the ACTIVE state will be set dur-

ing the sensing process of IoT sensors until the data

is saved into the local storage of the IoT device. In

this way, each simulated IoT device enters the OFF

state when it is created, the RUNNING state, when

it is started, and it is periodically switches between

Raspberry Pi benchmark values are available at:

https://lastminuteengineers.com/esp32-sleep-modes-

power-consumption/

ESP32 benckmark values are available at:

https://lastminuteengineers.com/esp32-sleep-modes-

power-consumption/

the ACTIVE (performing sensor data generation) and

RUNNING states till it is stopped (OFF state) or ter-

minated.

We would also note that DISSECT-CF-Fog pro-

vides a transparent and easily usable interface to cre-

ate additional, new states, and hence multiple energy

models, and it is up to the researcher, where to use

such new states during the simulation.

In the next section we continue with the evalua-

tion of the proposed, extended energy model on two

typical and widespread IoT use cases.

Figure 2: Cumulative energy consumption of cloud, fog

nodes and IoT devices.

Figure 3: Energy consumption percentage of cloud, fog

nodes and IoT devices.

4 EVALUATION

In this section we illustrate the use of the extended,

uniﬁed energy model for IoT-Fog-Cloud architectures

in DISSECT-CF-Fog. For this purpose, we model

one of the typical IoT use cases, which represents a

weather forecasting scenario with numerous weather

stations (run by IoT microcontrollers with special sen-

sors). These devices can communicate with a fog

layer directly, which contains three different nodes

with the equal amount of resources, utilising 40 CPU

cores and 40 GB of memory in total. On the top of

the fog topology, there is one cloud datacenter having

CLOSER 2021 - 11th International Conference on Cloud Computing and Services Science

324

Table 3: Comparison of the ﬁnal results of the simulated scenarios.

MicroController ESP32 Raspberry Pi

Number of Devices 100 1 000 10 000 100 1 000 10 000

Number of VMs 7 7 29 7 7 29

Cloud/Fog Cost ($) 0.84 0.85 2.79 0.84 0.84 2.79

IoT Cost ($) 0.009 0.90 83.2 0.009 0.90 83.2

Delay (min.) 1.03 1.25 3.50 1.03 1.25 3.50

Runtime (sec.) 0 4 95 0 4 118

Energy

Consumption (kWh)

cloud1

Consumption

0.067 0.068 0.068 0.067 0.034 0.068

fog1 Consumption 0.456 0.456 0.456 0.456 0.456 0.456

fog1 Device

Consumption

0.006 0.053 0.525 0.053 0.500 5.000

fog2 Consumption 0.436 0.431 0.531 0.433 0.456 0.492

fog2 Device

Consumption

0.011 0.106 1.050 0.105 1.003 10.000

fog3 Consumption 0.611 0.611 0.578 0.611 0.611 0.611

fog3

Device Consumption

0.016 0.158 1.575 0.150 1.500 14.972

Total Consumption by Nodes 1.570 1.569 1.636 1.569 1.558 1.628

Total Consumption by Devices 0.032 0.316 3.150 0.307 3.003 29.972

56 CPU cores and 40 GB of memory, furthermore,

the devices are not allowed to send messages (unpro-

cessed sensor data) directly to the cloud (they are con-

nected only to the fog). We considered two types of

virtual machine images simulating existing Amazon

Cloud (AWS) instances. The cloud1 node can utilise

VMs with 8 CPU and 4 GB of memory, their hourly

prices were set to 0.202$, while the fog1, fog2 and

fog3 nodes can deploy VMs with 4 CPU, 2 GB of

memory with 0.101$ hourly price. We also set the

IoT side pricing by applying the IBM Cloud pricing

schema, which charges the consumer after the amount

of data exchanged (in MB).

In our simulation, the microcontrollers can use

either ESP32 or Raspberry Pi energy models, and

they are equipped with a temperature-humidity sen-

sor (similarly to our real world measurements). In our

weather forecasting use case, we deﬁned three differ-

ent scenarios by scaling up the number of operating

devices. In the ﬁrst case, we utilised 100 IoT devices,

then we increased the number of devices to 1000, ﬁ-

nally in the last case the maximum device number

was 10000, operated for 60 minutes within the exper-

iments. The microcontrollers measured the environ-

mental parameters every 60 seconds, similarly to the

real device evaluation (shown in Section 3.1), hence

our goal was to map the real monitoring execution in

the DISSECT-CF-Fog simulation environment.

The evaluation process is the following in each

scenarios: (i) the IoT microcontrollers monitor the en-

vironment based on their sampling frequency, (ii) the

generated data are forwarded to the less loaded fog

node (using the default scheduling algorithm), (iii) a

node allocates a task (i.e. collection of 256 Kilobytes

of data) to a VM to be processed, or requests a new

one, if there is no free VM available, in case the cur-

rent resource capacity allows it. Otherwise, the unal-

located task will be moved to a less loaded node (in

the fog or to the cloud layer).

During the evaluation, we modelled a European-

wide scenario, where the cloud was located in Frank-

furt, whilst the the three fog nodes were positioned in

London, Budapest and Vienna. The latency between

them was determined based on online ping statistics

The delay between a device and a fog node was set

to an average 50 ms weighted with the actual phys-

ical distance, and the positions of the devices were

randomly generated across Europe.

In order to highlight the energy consumption of

the nodes, the number of VMs were scaled up and

down dynamically according to the actual load caused

by the tasks. To be as realistic as we can, each com-

putational resource dealt with different energy mod-

els based on the resource schema of LPDS Cloud of

MTA SZTAKI

. The exact values we used to set the

energy model parameters are summarised in Table 2

and Table 4.

WonderNetwork website is available at:

https://wondernetwork.com/pings/

LPDS Cloud of MTA SZTAKI website is available at:

https://www.sztaki.hu/en/science/departments/lpds/

Modelling Energy Consumption of IoT Devices in DISSECT-CF-Fog

325

Table 4: The chosen values of the energy model for nodes

and microcontrollers.

Resource Type Min Power Idle Power Max Power

cloud1 20 398 533

fog1 20 296 493

fog2 20 296 533

fog3 20 398 493

The comparison of the results can be seen in Table

3. We listed the number of VMs utilised by all nodes

(Number of VMs), and the cost of both the cloud/fog

and IoT sides (Cloud/Fog cost, IoT cost). The Delay

value reﬂects to the makespan of the IoT application,

whilst the Runtime indicates the elapsed time in the

execution environment required by the actual simu-

lation. The Power Consumption ensures consump-

tion information detailed for each computational node

(e.g. cloud1 Consumption denotes the consumed en-

ergy by the cloud1 node), and we also counted the

summed consumption values of IoT devices related to

an actual node (e.g. fog1 - IoT Device Consumption

denotes the total consumed energy by all simulated

microcontrollers connected to fog1). Lastly, total en-

ergy usage of both nodes and devices are presented by

Total Consumption of Nodes and Total Consumption

of Devices.

As we can see from Table 3, the cloud resource

utilisation is basically the same in all six simulation

cases, because they had to deal with around the same

amount of unprocessed data/tasks (coming from the

fog layer). Nevertheless, it also shows that in case

of 1 000 devices, seven VMs could easily handle the

scheduled amount of tasks for both microcontrollers.

The more data a task contains, the more time it takes

for the task to be processed, and additional incoming

tasks may trigger new VMs to be deployed (depend-

ing on the applied task scheduling policy threshold).

In the third cases having 10 000 devices, the num-

ber of VMs is dramatically increased to 29, for both

device types.

Since the IBM Cloud pricing is independent of the

actual device type, only the transmitted data counts,

and the cost of the computational nodes is propor-

tional to the number of utilised VMs, therefore the

corresponding costs are the same in case of ESP32

and RPi. It can also be observed that the timeout de-

lay (i.e. application makespan minus the set operation

interval of the IoT devices (60 minutes in these sce-

narios)) is less than 90 seconds for 100 and 10 000

devices. As we can see for the third, 10 000 devices

cases, the throughput of the system decreased, hence

the delay increased to 3.5 minutes. The execution

time (runtime) of the simulations for all cases remains

within two minutes for all cases, which points out that

DISSECT-CF-Fog can manage thousands of entities

on a single PC (for the evaluation we used a PC with

Windows 10 OS, i5-4460 CPU and 8 GB memory).

Figure 2 and Figure 3 highlight the results by com-

paring the energy consumption ratio of the utilised

cloud node, fog nodes and IoT devices (i.e. micro-

controllers). Figure 2 depicts the total energy used in

kW h for each categories, while Figure 3 depicts their

ratio in percentage. As we can see from the diagrams,

the cloud consumption takes only a small part of the

total energy consumption in all six scenarios. The fog

nodes are mostly capable of handling the vast amount

of data with their own resources generated by the IoT

layer, and there is no need to involve cloud resources

drastically. Nevertheless, when we scale up the num-

ber of microcontrollers in the IoT layer, our results

show a signiﬁcant increase in the total energy con-

sumption, caused by only exclusively the operation of

the IoT devices. For the case of using 10 000 RPi de-

vices, we can see that the energy consumed by the IoT

layer takes up almost 95% of the total consumption,

as shown in Figure 3. For smaller scales, we can ob-

serve that 100 ESP32 devices caused only 2% of the

total energy consumption. This ratio goes up to about

16%, in case of 1 000 ESP32 and 100 RPi devices,

and we experienced around the same ratio in case of

10 000 ESP32 and 1 000 RPi devices (with

66%).

5 CONCLUSIONS

In this paper we presented a novel extension of the en-

ergy model of the DISSECT-CF-Fog simulator to en-

able the energy monitoring of its simulated IoT com-

ponents. In this way, we realised a uniﬁed energy

model capable of analysing the power consumption of

complex, IoT-Fog-Cloud infrastructures. We evalu-

ated the extension and exempliﬁed its utilisation with

a weather forecasting use case. Our results showed

that detailed energy consumption values can be gath-

ered by the extended DISSECT-CF-Fog, and the pro-

posed solution enables detailed conﬁguration for eval-

uating various power settings for all system elements.

Our future work will address battery draining sim-

ulation of smart devices, and sophisticated energy-

aware scheduling algorithm development for task of-

ﬂoading for fog nodes.

The IoT energy model extension of DISSECT-CF-

Fog is available online on GitHub:

https://github.com/andrasmarkus/dissect-cf/tree/energy/

CLOSER 2021 - 11th International Conference on Cloud Computing and Services Science

326

ACKNOWLEDGEMENTS

The research leading to these results was supported

by the Hungarian Government and the European

Regional Development Fund under the grant num-

ber GINOP-2.3.2-15-2016-00037 (”Internet of Liv-

ing Things”), and by the Ministry for Innovation

and Technology, Hungary under the grant number

NKFIH-1279-2/2020.

REFERENCES

Ahvar, E., Orgerie, A., and L

ebre, A. (2019). Es-

timating energy consumption of cloud, fog and

edge computing infrastructures. IEEE Transac-

tions on Sustainable Computing, EA:1–12, DOI:

10.1109/TSUSC.2019.2905900.

Atlam, H. F., Walters, R. J., and Wills, G. B. (2018).

Fog computing and the internet of things: A review.

Big Data and Cognitive Computing, 2(2):10, DOI:

10.3390/bdcc202001.

Garg, S. and Buyya, R. (2011). Green cloud comput-

ing and environmental sustainability. Harnessing

Green It: Principles and Practices, 16:315–339, DOI:

10.1002/9781118305393.ch16.

Gupta, H., Dastjerdi, A., Ghosh, S., and Buyya, R. (2017).

ifogsim: A toolkit for modeling and simulation of

resource management techniques in the internet of

things, edge and fog computing environments. Soft-

ware: Practice and Experience, 47:Pages 1275–1296,

DOI: 10.1002/spe.2509.

Jha, D. N., Alwasel, K., Alshoshan, A., Huang, X., Naha,

R., Battula, S., Garg, S., Puthal, D., James, P.,

Zomaya, A., Dustdar, S., and Ranjan, R. (2020).

Iotsim-edge: A simulation framework for modeling

the behavior of internet of things and edge comput-

ing environments. Software: Practice and Experience,

50:844–867, DOI: 10.1002/spe.2787.

Kaup, F., Gottschling, P., and Hausheer, D. (2014). Pow-

erpi: Measuring and modeling the power consump-

tion of the raspberry pi. Annual IEEE Conference

on Local Computer Networks, pages 236–243, DOI:

10.1109/LCN.2014.6925777.

Kecskemeti, G. (2015). Dissect-cf: A simulator to fos-

ter energy-aware scheduling in infrastructure clouds.

Simulation Modelling Practice and Theory, 58:188–

218, DOI: 10.1016/j.simpat.2015.05.009.

Lera, I., Guerrero, C., and Juiz, C. (2019). Yafs: A

simulator for iot scenarios in fog computing. IEEE

Access, 7:91745–91758, DOI: DOI: 10.1109/AC-

CESS.2019.2927895.

Mahmud, R., Kotagiri, R., and Buyya, R. (2018). Fog

computing: A taxonomy, survey and future directions.

Internet of Everything: Algorithms, Methodologies,

Technologies and Perspectives, pages 103–130, DOI:

10.1007/978–981–10–5861–5 5.

Maier, A., Sharp, A., and Vagapov, Y. (2017). Com-

parative analysis and practical implementation of the

esp32 microcontroller module for the internet of

things. International Conference on Internet Tech-

nologies and Applications, pages 143–148, DOI:

10.1109/ITECHA.2017.8101926.

Markus, A., Gacsi, P., and Kertesz, A. (2020). Develop

or dissipate fogs? evaluating an iot application in fog

and cloud simulations. International Conference on

Cloud Computing and Services Science, pages 193–

203, DOI: 10.5220/0009590401930203.

Markus, A. and Kertesz, A. (2020). A survey and taxon-

omy of simulation environments modelling fog com-

puting. Simulation Modelling Practice and Theory,

101:102042, DOI: 10.1016/j.simpat.2019.102042.

Motlagh, N. H., Mohammadrezaei, M., Hunt, J., and

Zakeri, B. (2020). Internet of things (iot) and

the energy sector. Energies, 13(2):494, DOI:

10.3390/en13020494.

Oma, R., Nakamura, S., Duolikun, D., Enokido, T.,

and Takizawa, M. (2018). An energy-efﬁcient

model for fog computing in the internet of things

(iot). Internet of Things, 1-2:14–26, DOI:

10.1016/j.iot.2018.08.003.

Perez Abreu, D., Velasquez, K., Curado, M., and Mon-

teiro, E. (2020). A comparative analysis of simu-

lators for the cloud to fog continuum. Simulation

Modelling Practice and Theory, 101:102029, DOI:

10.1016/j.simpat.2019.102029.

Qayyum, T., Malik, A., Khan, M., Khalid, O., and Khan,

S. (2018). Fognetsim++: A toolkit for model-

ing and simulation of distributed fog environment.

IEEE Access, 6:63570–63583, DOI: 10.1109/AC-

CESS.2018.2877696.

Samie, F., Bauer, L., and Henkel, J. (2016). Iot

technologies for embedded computing: A sur-

vey. International Conference on Hardware/Software

Codesign and System Synthesis, pages 1–10, DOI:

10.1145/2968456.2974004.

Sonmez, C., Ozgovde, A., and Ersoy, C. (2018). Edge-

cloudsim: An environment for performance eval-

uation of edge computing systems. Transactions

on Emerging Telecommunications Technologies, 29,

DOI: 10.1002/ett.3493.

Sun, H., Yu, H., Fan, G., and Chen, L. (2020). Energy

and time efﬁcient task ofﬂoading and resource alloca-

tion on the generic iot-fog-cloud architecture. Peer-to-

Peer Networking and Applications, 13:548–563, DOI:

10.1007/s12083–019–00783–7.

Modelling Energy Consumption of IoT Devices in DISSECT-CF-Fog

327