internet of things architectures: a comparative study
TRANSCRIPT
Internet of Things Architectures: A Comparative Study
Marcela G. dos Santos1a, Darine Ameyed2b, Fabio Petrillo1c, Fehmi Jaafar3d , Mohamed Cheriet2e
1Dpartement de Informatique et mathmatique, Universit du Qubec Chicoutimi, Chicoutimi, Qubec, Canada2cole de Technologie Suprieure, University of Quebec, Montreal, Qubec, Canada
3Computer Research Institute of Montral, Montreal, Qubec, [email protected], [email protected], c,edarine.ameyed.1, [email protected], d [email protected]
Keywords: Internet of Things, IoT, Architectures, Layers-Model, Providers
Abstract: Over the past two decades, the Internet of Things (IoT) has become an underlying concept to a variety ofsolutions and technologies that it is now hardly possible to enumerate and describe all of them. The conceptbehind the Internet of Things is as powerful as it is complex, and for the components in the IoT solution tomesh together perfectly, they all have to be part of a well-thought-out structure. That is where understandingthe IoT architecture becomes paramount. Because of the vast domain of IoT, there is no single consensuson IoT architecture. Different researchers and organizations proposed different architectures under a varietyof classifications, mainly: conceptual, standard and, industrial or commercial adoption. It is indispensableto make a systematic analysis of IoT architecture to be able to compare the industrial proposals and identifytheir similarities and their differences. In this work, we summarize information about seven IoT industrialarchitectures in order to propose an approach that make possible a comparative analysis between different IoTarchitectures. This work presents two main contributions: (i) an approach for analyzing and comparing IoTarchitectures using Layer-Model; (ii) a comparative study of seven industrial IoT architectures.
1 INTRODUCTION
The Internet of Things (IoT) marketplace is growingspectacularly in the last few years. Indeed, Huaweiexpectation for the number of devices connected is100 billion by 2025. Besides that, the impact of In-ternet of Things Market on the global economy isenormous. McKinsey Global Institute expected thisimpact to be around 10 trillion US dollars by 2025(Ahmed et al., 2019).
Nevertheless, there are many challenges in devel-oping IoT applications: the lack of general guidelinesor frameworks that handle low level communicationand simplify high level implementation, the using ofmultiple programming languages to implement IoTapplications, the diversity of the communication pro-tocols, and the high complexity of distributed com-puting (Ammar et al., 2018).
Several researchers pointed out the using of IoTArchitecture as a tool to clarify the complexity of theIoT solutions and to provide a better comprehensionof the issues that may threaten them (Alshohoumiet al., 2019).
Concretely, Alshohoumi et al. (Alshohoumi et al.,2019) analyzed 148 studies and identified sixteen dif-
ferent IoT architectures. Moreover, Pratap Singh etal. (Pratap Singh et al., 2020) affirmed that there is ingeneral three different ways to classify the IoT archi-tectures: domain-specific architectures, layer-specificarchitectures, and industrial or commercial defined ar-chitectures.
In the industrial context, we noticed that there isa variety of IoT architecture used and presented (Fig-ure 1). Thus, it is indispensable to make a system-atic analysis of IoT architecture based on a referencemodel to be able to compare the industrial proposalsand identify their similarities and their differences.
Thus, in this paper, we selected and analyzedseven industrial architectures in accordance of oneof the reference model of IoT architectures, i.e. theLayer-based architecture model.This study presentstwo main contributions: (i) an approach for com-paring IoT architectures in according with a referenceIoT architecture and (ii) a comparative study of sevenof the main industrial IoT architectures (Intel, Mi-crosoft, Cisco, Google,and IBM, Ericsson, Amazon).
The audience of this study are (i) researchers in-terested on modeling IoT architectures, and (ii) prac-titioners interested on comparing industrial IoT archi-tectures.
arX
iv:2
004.
1293
6v1
[cs
.SE
] 2
7 A
pr 2
020
(a) Cisco Architecture (Cisco, 2014)
(b) Microsoft Architecture (Microsoft, 2018)
Figure 1: Examples of IoT Architectures
The remaining paper is organized as follows: Sec-tion 2 provides an overview about the IoT Architec-tures. Section 3 presents the related work. Section 4is the kernel of the work, which provides a approachto analyze and model IoT architectures using Layers-Model. Section 5 presents the modeling of seven in-dustrial architectures using our approach. Section 6performing the analyze of seven industrial architec-tures and the lessons learned in our study. Section 7explains the threats to validity. Finally, Section 8 syn-thesizes the final remarks and future work.
2 IoT LAYER MODEL
While acknowledging the existence of other IoTarchitectures classification as conceptual, domain-based and industrial. In our actual study, as mention,we focus on the layered-IoT architectures and indus-trial architectures. Firstly, each layer address pointsseparated before it is integrated and perform as a sys-tem. This methodology helps manage the complexityof the system. IoT scenarios have a high level of com-plexity because of the integration of various kinds oftechnologies, devices, objects, and services.
The primary studies designed their layered archi-tecture ranged from 3 to 7 layers which are composedby the main building blocks in the IoT platforms,
varying from the basic to the end-to-end solutions.
The early IoT model was a three-layer architec-ture presented by Gubbi et al. (Gubbi et al., 2013).Basically, it consists of perception as a ground layerincluding sensors and actuators as things, cloud as aninformation processing layer, and application layerthat allows users’ interaction as the third layer. Ex-tended by adding a business layer to provide the four-layer model (Muccini and T. Moghaddam, 2018).
Furthermore, A. Al-Fuqaha et al. (Al-Fuqahaet al., 2015) gives the definition of IoT architectureas middleware layer-based and five-layer model. Themiddleware layer-based, include service composition,service management and object abstraction. Besides,there is the six-layer model adding fog layer or a gate-way layer to the five-based model including also edgeand hybrid edge-cloud (Pan and McElhannon, 2018).
Finally, the recent proposal for the IoT layeredarchitecture is delivered by Cisco as a seven-layermodel. The previous architecture was changed byadding a user and process layer and edge computinglayer (Cisco, 2014).
2.1 IoT Layers-Model: 3,5 and 7-LayersModel
There are many model Layers Model Architecturesdescribed by the literature. For example, in the sys-tematic review about the Internet of Things architec-ture, Alshohoumi et al. (Alshohoumi et al., 2019) no-ticed, after analyzing 148 studies with 16 mappingarchitectures, that the most of architectures can beclassified as a three, four or five layers. Another sur-vey about IoT architecture is the study conducted byPratap Singh (Pratap Singh et al., 2020) noticed thatthe model layer classification applied is three, four,five, six, seven layers.
In our current study, we have focus on the 7-Layers architecture model because it is considered thecomplete model, regarding the complexity of the IoTsystems nowadays and the trend in this area.In ourcurrent study, we choose to set the focus on the 3, 5,and 7-Layers architecture models for several reasons.Firstly, the 3-Layers model is the most abstract andstraightforward model where the majority of researchin IoT architecture starts ((Khan et al., 2012) and (Wuet al., 2010)). Secondly, the 7-Layers is consideredthe complete model, regarding the complexity of theIoT systems nowadays and the trend in this area. Fi-nally, the 5-Layer model is a model between the othertwo reference models (3 and 7-Layers model).
2.1.1 3-Layers Model
According to Ray (Ray, 2018) and Lin et al. (Linet al., 2017) the trivial architecture for IoT systemsis called 3-Layers as it has three layers: perception,network, and application layers (Figure 2a).
On the Perception Layer are the things. The thingshave sensors that are responsible for taking informa-tion, and actuators to interact with the environment.
The Networking Layer is in charge of connectingthe thing with another things, network devices, andservers.
The last layer on the architecture is the Applica-tion Layer that addresses the delivery of the servicesfor the final user, and it is on this layer that is theclouds and servers.
The IoT 3-Layers as an accepted structure but istrivial modeling of the IoT ecosystem (Alshohoumiet al., 2019).One point important for 3-Layer archi-tecture is the fact that there is not a layer for Business.
2.1.2 5-Layers Model
Researchers proposed a new architecture to solvethe issues summarized in the 3-Layers architecture.The 5-Layers (Figure 2b) is an extension of 3-Layerswith the introduction of Processing and Business Lay-ers((Wu et al., 2010) and (Sethi and Sarangi, 2017)).
The characteristics and goals of the Perceptionand Application Layers are the same as the 3-Layersarchitecture. The Transport Layer is responsible fortransferring the data from things to the ProcessingLayer in both ways.
The Processing Layer works as a middleware inthe 5-Layers architecture; the role of this layer is tostore, analyze, and process the information of objectsreceived from the transport layer. With the ProcessingLayer, various technologies can be applied, for exam-ple, database, cloud computing, and big data process-ing modules.
Finally, the Business Layer is the layer that threat-ens all IoT system management; it includes the ap-plication, business, and profit models and user pri-vacy. In the 5-Layers, it is considered data storageand processing, but neither security and privacy arediscussed.
2.1.3 7-Layers Model
The 7-Layers (Figure 2c) used in our current studywas proposed in the Internet of Things World Forum(IoTWF) (Pisching et al., 2018). Following the de-scription for each layer.
• The first layer is where a variety of devices, sen-sors, and controllers that enable their interconnec-tion are situated.
• The second layer is responsible for making allconnections and data transfers in the IoT system.Thus, this layer specifies the communication pro-tocols.
• The third layer, the Edge/Fog Computing layer, iswhere the data analysis and data transformation,is performed (Tzafestas, 2018).
• The fourth layer, the Data Accumulation, leadswith the storage data and guarantee that the datais moving correctly.
• The fifth is where the data is prepared to be ana-lyzed using the data mining techniques or the dataimplementation of machine learning (Tzafestas,2018). The last two layers are Application andCollaboration & Process.
• The Application Layer is where the users canuse the information about the environment that istaken by the things.
• Finally, the seventh layer represents the actors thatuse the data to make a decision based on the dataextracted on the IoT ecosystem (Pisching et al.,2018).
Network Layer
Perception Layer
Application Layer
3-Layers
(a)
Processing Layer
Network Layer
Application Layer
Perception Layer
Business Layer
5-Layers
(b)
Data Abstraction
Data Accumulation
Application
Edge(Fog)Computing
Collaboration & Processes
7-Layers
Connectivity
Physical Devices &Controllers
(c)Figure 2: IoT Layers Architectures
3 RELATED WORK
A systematic review of existing IoT architectures wasperformed by Alshohoumi et al. (Alshohoumi et al.,2019) on a set of 144 studies from 2008 to 2018. Itis a review of IoT architectures in terms of architec-ture classification (the number of layers), limitations
in each architecture, and considerations of differentaspects or features in each layer. Our study differsfrom the review perform by Alshohoumi et al. (Am-mar et al., 2018) in the following terms: (i) we in-vestigate the architecture proposed by IoT providers,(ii) we apply an approach to model and analyze thearchitecture following for IoT providers.
Ammar et al. (Ammar et al., 2018) presents a sur-vey on security of IoT frameworks; a total of 8 frame-works are considered. Their study had as a goal: clari-fying the state of the art IoT platforms and identifyingthe trends of current designs of such platforms, pro-viding a high-level comparison between the differentarchitectures of the various frameworks. Our studydiffers from their because, in our research, we modelthe IoT providers using our approach and before per-forming a comparison.
Pratap Singh et al. (Pratap Singh et al., 2020)presents an analysis of Layer-Specific, Domain-Specific, and Industry defined IoT architectures. Themain contribution is a summarizing of state of the artof IoT with a comparison of the industry-defined ar-chitectures. The work by Pratap Singh et al. and ourstudy can be considered as complementary each ofthem analyzes IoT architectures. Still, our study pro-poses an approach for modeling IoT architectures andalso compares seven industrial IoT architectures (In-tel, Microsoft, Cisco, Google, and IBM).
4 APPROACH FOR MODELINGAND ANALYZING OF IoTARCHITECTURES
In this section, we discuss our approach for model-ing and analyzing IoT architectures. The evaluationapproach which we propose is built upon the referen-tial architecture described in the section 2. First, wediscuss the construction, and then we summarize thesteps that composed our approach.
We start analyzing the documentation that IoTproviders make available publicly. These data areavailable on websites or in white papers ((Intel, 2015),(Microsoft, 2018), (Google, 2019), (IBM, 2019),(Cisco, 2014), (Amazon, 2016), (Ericsson, 2017)).
To analyze this information systematically, wesummarize essential aspects for each layer in the IoTarchitecture by the provider. We read the documenta-tion and conduct an extraction of some elements foreach layer: input, output, activities performed, andthe principal objective. It is important to note thatsome aspects have different terminology or function-alities in each layer for each provider.
Based on the extracted data for each layer in eachprovider (input, output, activities, and main objective)for the 7-Layers Model, we classified the layers inthe industrial architecture according the layer in thereferential model. For example, for the architecturefollowing by Microsoft, we compare each layer withthe reference model using the data extraction for Mi-crosoft IoT white paper (Microsoft, 2018), and wecould classified each layer in the Microsoft proposalin most similar layer in the 7-Layers model.
It is important to highlight there is some layer thatare not described neither in the documentation nor inthe figure represented the architecture. In this cases,we decided to represent the layer using dotted line todemonstrated that this layer does not have descriptionin the available documentation,
In our approach for each layer in the IoT architec-ture in question, we extracted data to answer the fol-lowing set of questions using the information madeavailable for the IoT provider:
• What is(are) the input(s) for the layer?
• What is(are) the output(s) for the layer?
• What does the layer perform the activities?
• What is the principal objective of the layer?
5 MODELING SEVEN IOTPROVIDERS USING OURAPPROACH
In this section, we apply the approach described inSection 4 for seven leading industrial actors: Intel,Microsoft, Cisco, Google, IBM, Ericsson, and Ama-zon.
Our inclusion criteria for the IoT architectureswere selecting architectures that were studied at leasttwice by Ammar et al. (Ammar et al., 2018), Lueth(Lueth, 2015) and Asemani et al. (Asemani et al.,2019). We used these studies as a guideline to se-lect the IoT architectures because of the followingreasons: (i) these researchers perform a study simi-lar than our study, (ii) these studies analyzed deeplythe IoT architecture and (iii) these studies analyzedthe leading IoT companies around the world.
5.1 Intel Architecture
Intel IoT Platform Reference Architecture is a systemarchitecture specification (SAS) to connect any prod-ucts and services to the cloud (Intel, 2015). Thereare two versions of this IoT Architecture. Firstly, ver-sion 1.0 is an architecture to connect the unconnected
things, and the version is the Intel SAS version 2.0that brings a specification about the integration of avariety of devices with intelligence and connectivityintegrated (Breivold, 2017).
In our study, we classified the Version 2.0 becauseof it being a reference architecture that facilitated theconvergence of operational technology and informa-tion technology. Additionally, it is a future-lookingreference architecture.
According to the (Intel, 2015), the three maincomponents are: things, network and cloud (Figure3). The Things component is not represent in the Lay-ered Architecture but as we notice in the documenta-tion available, Intel defined a end-to-end solution forconnecting nearly any type of device to the cloud.
In Figure 4 more specifically in the figure morein the right and dotted, we have the architecture pro-poses by Intel (Intel, 2015). The first group of layers(in dark blue) have layers that are the major run-timelayers. The Communication and Connectivity layer issituated on the bottom of the architecture, and are re-sponsible for enabling multi-protocol data communi-cation between not only devices and the edge but alsobetween endpoint devices/gateways, the network, andthe data center (Intel, 2015).
The second layer is the Data layer with Analyt-ics, whose role is providing customer value. It isachieved using valuable insights generated by dataanalytics and improved closed-loop control systems.On the Intel IoT reference architecture, this needis addressed by allowing analytics to be distributedacross the cloud, gateways, and smart endpoint de-vices (Breivold, 2017).
The Management layer is the next layer in whichthe primordial role is for realizing automated discov-ery and provisioning of endpoint devices. Intel rec-ommends Device Cloud product to perform the man-ageability functions (Breivold, 2017). And to com-plete the dark blue group, the Control layer that pro-vides a way to separate the management layer into amanagement plane and control plane, with policy andcontrol objects and APIs.
Now, the second group is composed of user lay-ers. The first is the Application layer that is used bythe Business Layer to access the other layers in theIntel IoT architecture. Besides these layers, there is avertical security layer. The role of the security layeris to address protection and security across all tiers.
We could summarize some aspect about Intel ar-chitecture, applying our approach. The first aspectimportant in the Intel architecture is that there is aBusiness Layer, what shows that for Intel is impor-tant to have clearly the IoT economic impact. And, inIntel architecture, the Data Abstraction layer (Figure
4) is not described neither in the documentation norin the layer architecture
Figure 3: Intel End-to-End IoT Solution form Things toNetwork to Cloud. (Intel, 2015)
5.2 Microsoft Architecture
The design of the Microsoft IoT architecture is basedin three layers: Thing, Insight, and Actions. Thereare a set of subsystems in all layers, and it is essen-tial to highlight that Edge Devices, Data Transforma-tion, Machine Learning, and User Management areoptional subsystems (Microsoft, 2019).
The architecture proposes by Microsoft (Figure1b) the first group of subsystems are in the layer“Things” and are IoT Devices, IoT Edge Devices, andDevice Provisioning.
The IoT devices are devices that need to registerin a way secure with the cloud to send and receivedata. The second group is on the layer “Insights” andconsists of Cloud Gateway, Data Transformation, UIand Reporting Tools, Stream Processing, Warm PathStore, Cold Path Store.
In the case of the devices or field gateways thatare not able to use the standard protocols used by IoTHub, adaptation is necessary, and Azure IoT protocolgateway can be used (Microsoft, 2019). After that, wehave in the Microsoft IoT architecture the Stream Pro-cessing that is responsible for consuming the massivestreams of data records and evaluate rules for thosestreams.The third layer is the “Action” layer. On thislayer, we have the Business integration, which essen-tial role is performing actions based on the teleme-try data during stream processing. Also, in the “Ac-tion” layer, we have the User Management subsystemwhose goal is restricting the user or user groups ac-tion on the devices. And finally, we have the MachineLearning subsystem that is responsible for perform-ing predict algorithms using the telemetry data. Thisprediction can be applied in predictive maintenance,for example.
We apply our approach and could summarize sim-ilarities and divergences between the layers in the Mi-crosoft architecture and the layers in the 7-Layers. Wecould find similarities between the layers IoT EdgeDevices and Cloud Gateway and the Transport layer.
2
3
4
7
6
Intel Architecture
Communications andConnectivity Layer
Application Layer Control Layer
Not described
Business Layer
Data Layer (Analytics)
Management Layer
3
2
6
7
4
Things1
5
6
Figure 4: Analyze of Intel IoT Architecture.
The Cloud Gateway and Data Transformationmodules in the architecture follows by Microsoft havesimilarities with the layer Edge(Fog)computing in thereferential model 7-Layers. The Data Accumulationis represented by the Warm Path Store and Cold PathStore. For the Data Abstraction in the referentialmodel, we have the Stream Processing in the Mi-crosoft architecture.
For the Application layer in the 7-Layers model,we could find similarities with the modules UI andReport Tools, User Management and Machine Learn-ing. Finally, the activities that composed the Businesslayer are performed by the module Business Integra-tion.
5.3 Cisco Architecture
Cisco proposal(Figure 6)is a multi-level reference thatis composed of seven layers in the dotted area shows(Cisco, 2014).
In level 1 is the Things on the Internet of Things.As an example, we can have endpoint devices thatmay send and receive information. Level 2 is focusedon communication and connectivity. The connectivityincludes the data transmission between devices andthe network, across networks and between the net-work and low-level information processing. The ac-tivities in level 3 are focused on reorganizing the datathat is produced by the things in the level 1 into in-formation that is convenient for level 4 (storage andhigher processing). The priority of level 4 is guar-anteeing that the data is moving precisely. In other
words, the data is moving at the rate and organizationdefined by the devices that generated the data in level1. Data Accumulation level will make network dataturn into data usable for the application. The idea isto convert data-in-motion to data-at-rest, convertingdata network packets to relational database tables.
After the data accumulation is necessary to ren-der the data and store this information in a way tofacilitate the development of application more simpleand with higher performance, this is performing bythe level 5.Level 6 is the application level, where allthe data that is generated is interpreted using varioustypes of an application. The collaboration & processlevel execute the applications with the specific needs.With the apps (level 6), people have access to the rightdata at the right time so that they can do the right thing(Cisco, 2014).
We applying our approach and find just similari-ties between the layers in the Cisco architecture andthe layers in the 7-Layers. For us, the analysis ofCisco was proof that our approach is a valuable toolto be used to model and analyze IoT architectures.
5.4 Google Architecture
Google Cloud Platform (GCP) is a complete set oftools to connect, process, store, and analyze data bothat the edge and in the cloud (Google, 2019). To GCP,the IoT solution consists of three essential compo-nents, the device, gateway, and cloud. The elementthat is directly related to the world is the device. Forthis reference architecture, the device can be hardware
IoT Devices
IoT Edge Devices and Cloud Gateway
UI and Report Tools,Machine Learning and User
Management
Stream Processing
Business Integration
Cloud Gateway and DataTransformation
Warm Path Store and Cold Path Store
1
2
3
4
5
6
7
Microsoft Architecture
1
2
2 34
4
5
3
6
6
7
6
Figure 5: Analyze of Microsoft IoT Architecture
Physical Devices andControllers
Connectivity
Application
Data Abstraction
Collaboration and Process
Edge(Fog) Computing
Data Accumulation
1
2
3
4
5
6
7
1
2
3
4
5
6
7
Cisco Architecture
Figure 6: Analyze of Cisco IoT Architecture.
and software and might be directly or indirectly con-nected to the Internet. Besides that, devices can com-municate with each other via a network.
The gateway is responsible for guaranteeing thedevices that are not directly connected to the Internetto reach cloud services. Another aspect of the gate-way is that it processes data on behalf of a group orcluster of devices. The devices collect data; the gate-way sends this data to Cloud Platform, the third partof the architecture. On the cloud platform, the dataare processed and combined with other data that weresent for different devices.
We analyzed the similarities and divergence be-tween Google architecture and our referential model(Figure 7). In our analysis, we concluded that there isa high level of linkage between the internal modulesin the Data Analytics in the Cloud. For example, there
are three modules in the Data Analytics in the Cloudthat can be classified as Data Accumulation in the7-Layer model Although Google uses only four ele-ments to present the IoT architecture following in itssolutions, it was necessary to ”open” the modules tounderstand deeply what the activities performed andhow we could classify using our approach.
5.5 IBM Architecture
The IBM (Figure 8) proposal for IoT architecture iscomposed by the following layers User Layers, Prox-imity Network, Public Network, Provider Cloud, andEnterprise Network. The main objective of IBM IoTarchitecture is to perform a connection to IoT de-vices and quickly build scalable apps and visualiza-tion dashboards to gain insights from IoT data. IBMIoT architecture uses IBM Cloud IoT, data, and AI
Cloud Machine Learning5
IoT Devices
Edge Gateway
Data Studio and CloudDataLab
Insights
Cloud IoT Core, Functions
Cloud(Pub/Sub,Bigtable,DataFlow)
1
2
3
4
6
7
6
75
3
4
4
4
Data Analytics in the CloudEdge gateway Data Usage
Control
Data
Data
Serving
Training
Update deviceconfig
Direct cloudconnectivitythroughCloud IoTCore
Control
Powered by CloudIoT Device SDK
IoT Devices
Control
Data
Data
Security librariesConnectivity
librariesCloud IoT
Providiong agentRTOS Kernel
HW Driver
Update deviceconfig
Updatedeviceconfig &deploy MLmodel andcontainer
Figure 7: Analyze of Google IoT Architecture.
services to achieve the main goal (IBM, 2019).There are five layers called the User Layer, Prox-
imity Network, Provider Cloud, and Enterprise Net-work. We could find similarities between the layersin the IBM architecture and the 7-Layers model withthe application of our approach. But it is essential tohighlight that we made the analysis using the inter-nal modules insight the layers: User Layer, ProximityNetwork, Provider Cloud, and Enterprise Network.
The sensors (water leak detection, water flow,temperature, etc.) and actuators (automatic watershutoff valves) are in the Proximity Network; thislayer can be at home for example, if we talk aboutIoT platforms for Smart Homes. They are attached tothe device maker’s cloud service.
In this scenario, we can see an IoT application, theSmart homes. The connected devices let insurancecompanies improve service for their policyholders.Besides that, with data collected is possible to provideinsight about risks that can happen in the home. Forexample, leak-detection sensors can monitor for waterleaks. With a correct analyze, valves can be trigged,and the IoT platform can help protect the home fromresulting damage.
The Device performs activities comparable withthe Perception Layer. The IoT Gateway is the Trans-port Layer in the IBM architecture.
In the layer numbering with three, we have morethan one module in the industrial architecture (IBM)
performing as one layer in the referential model.Edge Service, IoT Transformation & Connectivity,and Transformation & Connectivity compos the thirdlayer in the IBM architecture as is comparable withthe Edge(Fog) Computing in the 7-Layers model.
The modules Data Store and Enterprise Data num-bering perform activities similarities with the DataAccumulation. Application logic (IBM architecture)is the Data Abstraction(7-Layers model).
All device modules (registry, management, andidentity service), API Management, User Directory,Process Management, Analytics, and Visualizationcompos the sixth layer in the IBM architecture andis comparable with the Application Layer in the 7-Layers model.
Finally, IoT User and Enterprise Application arethe seventh layer in the IBM architecture and are com-parable with the Collaboration and Processes in the7-Layers.
5.6 Ericsson Architecture
Ericsson’s IoT architecture proposal (Figure 9) is a se-curity solution that provides continuous monitoring ofthreats, vulnerabilities, risks, and compliance, alongwith automated remediation (Ericsson, 2017).
We analyze the architecture proposal by Ericssonusing our methodology. We classified the IoT devicesas the Perception layer, the IoT Gateway, Access, and
1
2
3 4
6
7
5 7
Device
IoT Gateway
Application, Device (Registry, Management IdentityService), API Management, User Directory, Process
Management, Analytics, Visualization
Application Logic
IoT User, Entereprise Application
Edge Service, (IoT) Transformation & Connectivity
Data Store, Enterprise Data
1
2
3
4
5
6
7
6 6
6
6
66 6 46
3
3
IBM Architecture
Figure 8: Analyze of IBM IoT Architecture.
network connectivity as a Network layer. In the layerIoT app, platform & cloud, we could find similaritiesbetween the activities performed in the module CloudInfrastructure with Edge(Fog)Computing, and Appli-cation with Data Abstraction, we do not find the de-scription about the Data Accumulation. IoT User asApplication Layer.We do not find represent the Busi-ness Layer.
The Business Layer and Data Accumulation in theEricsson architecture is not represented but all theother layers have similarities with the layer in the 7-Layers model.
5.7 Amazon Architecture
The main objective of the AWS IoT is providing se-cure, bi-directional communication between Internet-connected devices such as sensors, actuators, embed-ded micro-controllers, or smart appliances, and theAWS Cloud (Amazon, 2016).
The elements that compose the architecture areAlexa voice service integration for AWS IoT, customauthentication service, device gateway, device provi-sioning service, device shadow, device shadow ser-vice, group registry, jobs service, message broker,registry, Rules engine and security, and integrity ser-vice (Amazon, 2016).
We analyze the architecture proposal by Ama-zon(Figure 10) using our approach and classified thecomponent Devices as a Perception Layer. The Ama-zon documentation describes the Gateway compo-nent, but it does not show in the architecture. Thecore of the AWS solution (Guth et al., 2016) is com-posed of the components numbering by 3, and theyhave similarities with the Processing in the 7-Layers
model. As well as in the Ericsson architecture (Sub-Section 5.6), there is no description of a BusinessLayer in the Amazon architecture.
6 ANALYZING SEVEN IOTPROVIDERS USING OURAPPROACH
This section presents a comparative analysis of sevenindustrial architecture modeled in Layer-Model usingour approach (Figure 11).
Although the fact the seven industrial architec-tures have terminology different, it was possible tocompare the layers providers to the layer-model us-ing the methodology that we proposed. We noticed asfar as that the abstraction level concerning the physi-cal layer increases, the terminology became more di-verse.
We could classified all the layers in the seven ar-chitectures studied into seven layers. In other words,we use 7-layers model to identify the layers in the in-dustrial architecture that can be considered similar.
For the network layer in the 7-Layers model, wenotice that five of the seven industrial have used theterm Gateway or similarities to express the layer re-sponsible for performing the communication with andbetween the perception layer. It is important to clar-ify the gateway has a specific function in networking,it is also used to describe a class of device that pro-cesses data on behalf of a group or cluster of devices(Google, 2019).
Five industrial architectures defined, in the docu-mentation, the Business layer, and three of them use
1 2 6
3
Ericsson Architecture
2
IoT Gateway, Access and network connectivity
IoT user (Application)
IoT app (Application)
Not described
Cloud Infrastrucutre
Not described
IoT Device1
3
5
4
7
2
5
6
Figure 9: Analyze of Ericsson Architecture
the term Business or similar to denoted the last levelof them architectures. We conclude that for the ma-jority of the companies is important to represent theeconomic impact of the IoT solutions. Another aspectin the last layer for the industrial architectures and be-come a recommendation for Ericsson and Amazon ismaking explicit the Business layer in them architec-tures.
Our study could show in a comparative approachthe fact that is known in the IoT community: thereis no single consensus on IoT architecture. The com-plexity of IoT architecture is one of the reasons formany proposal architectures. With decreasing com-plexity, using a referential model, we could evaluatedifferent architectures.
We learned during the IoT architecture study somemajor lessons:
• The use of layers to represent architecture is a waymore clear to understand the architectures.
• There are some layers in the industrial IoT archi-tectures analyzed that perform the same activitiesbut with different nomenclature.
• There are modules in industrial architecture thatperform activities inherent to more than one layerin the layer-model architecture.
• It is necessary a standard to provide a better com-prehension of the issues and threats in IoT.
7 THREATS TO VALIDITY
During our study, we have counted some threats thatwe need to address. We work carefully to mitigateeventual issues that could compromise the validity ofour results or conclusions. In this section we highlightsome of those threats and what mechanism that weapplied to address it.
First, the main limitation of this work is the anal-ysis of the IoT architecture by the providers was per-forming using only the information available in sitesand white paper. However, we opted to performthe analysis only with the data made available forthe companies because it is this information that ouraudience also has to analyze and compare the IoTproviders. Besides, we could extract the data, pro-poses an approach, and compare the architectures.
The second threat to validity is the bias created bythe fact that we used one type of referential model,and we chose three models. However, as this, ourmapping is preliminary work; consequently, we de-cide to start the study only with these models and af-ter as a future work performing the same analysis withother models.
8 CONCLUSION
IoT market is continually growing; as a consequence,the number of IoT architectures has increased too.There is no unique or consensus about the IoT archi-tecture; each architecture has the approach and the in-
33
3
3
5
5
5
4
5
5
1
6
Devices
Device Gateway
IoT Applications
Amazon (DynamoDB, Kinesis, Lambda, SNS, SQS)
Not described
Gateway, Message broker, Device Shadows, Rules Engine
Amazon S3
1
3
4
6
7
2
5
Amazon Architecture
Figure 10: Analyze of Amazon Architecture: the Business Layer is not clearly represented in the architecture documenta-tion((Amazon, 2016)) but all the other layers have similarities with the 5-Layers model.
Google Architecture
IoT Devices
Data Studio and Cloud DataLab
Cloud Machine Learning
Insights
Cloud IoT Core, Functions
Cloud(Pub/Sub, Bigtable,DataFlow)
1
3
4
5
6
7
Edge Gateway2
IoT Devices and IoT Edge Devices Physical Devices and Controllers
Connectivity
Application
Data Abstraction
Collaboration and Process
Edge(Fog) Computing
Data Accumulation
1
2
3
4
5
6
7
UI and Report Tools,MachineLearning and User Management
Stream Processing
Business Integration
Cloud Gateway and Data Transformation
Warm Path Store and Cold Path Store
Microsoft Architecture
1
3
4
5
6
7
Device
IoT Gateway
Application, Device (Registry, Management, IdentityService), API and ProcessManagement, User Directory,
Analytics and Visualization
Applicaiton Logic
IoT User, Enterprise Application
Edge Service, (IoT) Transformatiomn &Connectivity
Data Store and Enterprise Data
1
2
3
4
5
6
7
Cloud Gateway2
IoT Gateway, Access and network connectivity
Not described
Google Architecture
IoT Devices
Data Studio and Cloud DataLab
Cloud Machine Learning
Insights
Cloud IoT Core, Functions
Cloud(Pub/Sub, Bigtable,DataFlow)
1
3
4
5
6
7
Edge Gateway2
IoT Devices and IoT Edge Devices Physical Devices and Controllers
Connectivity
Application
Data Abstraction
Collaboration and Process
Edge(Fog) Computing
Data Accumulation
1
2
3
4
5
6
7
UI and Report Tools,MachineLearning and User Management
Stream Processing
Business Integration
Cloud Gateway and Data Transformation
Warm Path Store and Cold Path Store
Microsoft Architecture
1
3
4
5
6
7
Device
IoT Gateway
Application, Device (Registry, Management, IdentityService), API and ProcessManagement, User Directory,
Analytics and Visualization
Applicaiton Logic
IoT User, Enterprise Application
Edge Service, (IoT) Transformatiomn &Connectivity
Data Store and Enterprise Data
1
2
3
4
5
6
7
Cloud Gateway2
7
IoT user (Application)6
Cloud Infrastrucutre3
IoT Device1
Ericsson Architecture
Not described7
IoT Applications6
Gateway, Message broker, Device Shadows,Rules Engine 3
Device Gateway
Devices1
Amazon Architecture
Amazon (DynamoDB, Kinesis, Lambda, SNS, SQS)
Amazon S3
5
4
222
Business Layer7
Intel ArchitectureIntel Architecture
Cisco Architecture IBM ArchitectureCisco Architecture IBM Architecture
Communications and Connectivity Layer
Application Layer Control Layer
Not described
Data Layer (Analytics)
Management Layer
3
2
6
4
Things1
5
4
IoT app (Application)
Not described
5
Figure 11: Industrial providers classified as a Layers-Model
terpretation of IoT for the provider. In this study, we summarize information about
seven IoT industrial architectures. And, we providean approach that makes possible the modeling of IoTindustrial architectures based on Layers-Model, aswell as the comparison between IoT architectures.
With our analysis, we could concluded that eventhough the industrial architectures have representedIoT solutions in a different way with a different termi-nology, architectures perform the same activities. Be-sides that, it was clear that some architectures follow-ing the seven-layers model, for example, Microsoftand IBM) but others are used the model differently,for example, Intel and Google.
We could conclude not only the need for a stan-dard for IoT architecture and but also the advantagesto represent the architectures using layers. With thelayers, it was easier to compare and analyze better theindustrial IoT architectures.
We aim that our study benefit researchers that per-forming analysis of IoT architectures and practition-ers that need to choose IoT providers. Future work in-cludes (i) a quantitative analysis of IoT architecturesstudied, (ii) a systematic literature review of Internetof Things architectures, and (iii) systematic analysisabout the products that are available by the providers.
ACKNOWLEDGEMENTS
To the members of the SmArtSE Research Team,Synchromedia Laboratory and Computer ResearchInstitute of Montral (CRIM) for their supportand knowledge sharing. This work was financedby the Canadian program MITACS and LabVI(http://quartierinnovationmontreal.com/en/open-sky-laboratory-smart-life).
REFERENCES
Ahmed, B. S., Bures, M., Frajtak, K., and Cerny, T. (2019).Aspects of quality in internet of things (iot) solutions:A systematic mapping study. IEEE Access, 7:13758–13780.
Al-Fuqaha, A., Guizani, M., Mohammadi, M., Aledhari,M., and Ayyash, M. (2015). Internet of things: Asurvey on enabling technologies, protocols, and ap-plications. IEEE Communications Surveys Tutorials,17(4):2347–2376.
Alshohoumi, F., Sarrab, M., Al-Hamdani, A., and Al-Abri,D. (2019). Systematic review of existing iot architec-tures security and privacy issues and concerns. Inter-national Journal of Advanced Computer Science andApplications, 10.
Amazon (2016). AWS IoT Documentation. https://docs.aws.amazon.com/iot/.[Online; accessed 04-March-2020].
Ammar, M., Russello, G., and Crispo, B. (2018). Inter-net of things: A survey on the security of iot frame-works. Journal of Information Security and Applica-tions, 38:8 – 27.
Asemani, M., Abdollahei, F., and Jabbari, F. (2019). Un-derstanding iot platforms : Towards a comprehen-sive definition and main characteristic description. In2019 5th International Conference on Web Research(ICWR), pages 172–177.
Breivold, H. P. (2017). A survey and analysis of referencearchitectures for the internet-of-things. ICSEA 2017,page 143.
Cisco (2014). The Internet of Things reference model.http://cdn.iotwf.com/resources/71/IoT ReferenceModel White Paper June 4 2014.pdf.[Online; ac-cessed 06-January-2020].
Ericsson (2017). End-to-end security manage-ment for the IoT. https://www.ericsson.com/4ab42a/assets/local/reports-papers/ericsson-technology-review/docs/2017/managing-iot-security-end-to-end-etr-10-2017.pdf.[Online; accessed 20-December-2019].
Google (2019). Overview of internet of things. https://cloud.google.com/solutions/iot-overview.[Online; ac-cessed 11-December-2019].
Gubbi, J., Buyya, R., Marusic, S., and Palaniswami, M.(2013). Internet of things (iot): A vision, architec-tural elements, and future directions. Future Gen-eration Computer Systems, 29(7):1645 – 1660. In-cluding Special sections: Cyber-enabled DistributedComputing for Ubiquitous Cloud and Network Ser-vices & Cloud Computing and Scientific ApplicationsBig Data, Scalable Analytics, and Beyond.
Guth, J., Breitenbcher, U., Falkenthal, M., Leymann, F., andReinfurt, L. (2016). Comparison of iot platform archi-tectures: A field study based on a reference architec-ture. In 2016 Cloudification of the Internet of Things(CIoT), pages 1–6.
IBM (2019). Internet of things for insights from con-nected devices. https://www.ibm.com/cloud/garage/architectures/iotArchitecture.[Online; accessed 06-December-2019].
Intel (2015). The Intel IoT Platform. https://www.intel.com/content/www/us/en/internet-of-things/white-papers/iot-platform-reference-architecture-paper.html.[Online; accessed 17-December-2019].
Khan, R., Khan, S. U., Zaheer, R., and Khan, S. (2012). Fu-ture internet: The internet of things architecture, pos-sible applications and key challenges. In 2012 10thInternational Conference on Frontiers of InformationTechnology, pages 257–260.
Lin, J., Yu, W., Zhang, N., Yang, X., Zhang, H., and Zhao,W. (2017). A survey on internet of things: Archi-tecture, enabling technologies, security and privacy,and applications. IEEE Internet of Things Journal,4(5):1125–1142.
Lueth, K. L. (2015). The top 20 Internet ofThings companies right now: Intel over-takes Google. https://iot-analytics.com/20-internet-of-things-companies/.[Online;accessed18-February-2020].
Microsoft (2018). Azure iot reference architecture.https://aka.ms/iotrefarchitecture.[Online;accessed 16-December-2019].
Microsoft (2019). Supoort additional protocols foriot hub. https://docs.microsoft.com/en-us/azure/iot-hub/iot-hub-protocol-gateway.[Online; accessed16-December-2019].
Muccini, H. and T. Moghaddam, M. (2018). IoT Architec-tural Styles, pages 68–85. 12th European Conferenceon Software Architecture, ECSA 2018.
Pan, J. and McElhannon, J. (2018). Future edge cloud andedge computing for internet of things applications.IEEE Internet of Things Journal, 5(1):439–449.
Pisching, M., Pessoa, M., Junqueira, F., Santos Filho, D.,and Miyagi, P. (2018). An architecture based on rami4.0 to discover equipment to process operations re-quired by products. Computers & Industrial Engi-neering.
Pratap Singh, S., Kumar, V., Kumar Singh, A., and Singh,S. (2020). A survey on internet of things (iot): Layerspecific vs. domain specific architecture. In Smys,S., Senjyu, T., and Lafata, P., editors, Second In-ternational Conference on Computer Networks andCommunication Technologies, pages 333–341, Cham.Springer International Publishing.
Ray, P. (2018). A survey on internet of things architectures.Journal of King Saud University - Computer and In-formation Sciences, 30(3):291 – 319.
Sethi, P. and Sarangi, S. (2017). Internet of things: Archi-tectures, protocols, and applications. Journal of Elec-trical and Computer Engineering, 2017:1–25.
Tzafestas, S. G. (2018). The internet of things: A concep-tual guided tour. European Journal of Advances inEngineering and Technology, 5(10):745–767.
Wu, M., Lu, T.-J., Ling, F.-Y., Sun, J., and Du, H.-Y.(2010). Research on the architecture of internet ofthings. In 2010 3rd International Conference on Ad-vanced Computer Theory and Engineering (ICACTE),volume 5, pages V5–484. IEEE.