Survey of Internet-of-Things platforms

Abstract

Internet-of-Things (IoT) platform (often referred to as IoT middleware) is a software that enables connecting the machines and devices and then acquisition, processing, transformation, organization and storing machine and sensor data. The objective of the research behind this paper is to establish a state of the art in the development of IoT platforms. In specific, here we present the main conclusions regarding the functional and design perspective to current IoT platforms and related research. The focus was made on cloud-based IoT platforms with significant user base and successful record of exploitation. We also consider the relevant theoretical foundations of IoT platform research, mostly by taking into account the results of European research. The conclusions and respective discussion are meant to be used in the development of a formal runtime model-driven IoT platform, which conceptual design decisions are shortly presented.

Full text

Survey of Internet-of-Things platforms
Milan Zdravković*, Miroslav Trajanović*, João Sarraipa**, Ricardo Jardim-Gonçalves**, Mario Lezoche***,
Alexis Aubry***, Hervé Panetto***
* Laboratory for Intelligent Production Systems (LIPS),
Faculty of Mechanical Engineering, University of Niš, Niš, Serbia
** CTS, Uninova, Dep. de Eng. Electrotécnica, Faculdade de Ciências e Tecnologia, FCT ,
Universidade Nova de Lisboa , Caparica , Portugal
*** Research Center for Automatic Control of Nancy (CRAN), Université de Lorraine, UMR 7039, Boulevard des
Aiguillettes B.P.70239, 54506 Vandoeuvre-lès-Nancy, France.
milan.zdravkovic@gmail.com, miroslav.trajanovic@masfak.ni.ac.rs
{jfss, rg}@uninova.pt
{mario.lezoche, alexis.aubry, herve.panetto}@univ-lorraine.fr
Abstract – Internet-of-Things (IoT) platform (often referred
to as IoT middleware) is a software that enables connecting
the machines and devices and then acquisition, processing,
transformation, organization and storing machine and
sensor data. The objective of the research behind this paper
is to establish a state of the art in the development of IoT
platforms. In specific, here we present the main conclusions
regarding the functional and design perspective to current
IoT platforms and related research. The focus was made on
cloud-based IoT platforms with significant user base and
successful record of exploitation. We also consider the
relevant theoretical foundations of IoT platform research,
mostly by taking into account the results of European
research. The conclusions and respective discussion are
meant to be used in the development of a formal runtime
model-driven IoT platform, which conceptual design
decisions are shortly presented.
I. INTRODUCTION
Internet of Things (IoT) has attracted attention of major
players in industrial landscape and it is currently one of
the most expected emerging technologies. According to
the 2015’s Gartner’s Hype Cycle for Internet of Things
[1], IoT is currently at the “peak of inflated expectations”.
Although some of its technologies and enablers are getting
closer to the so-called Plateau of Productivity (Wireless
Healthcare Asset Management, ZigBee, RFID for logistics
and transportation), IoT platforms are still at inception
phase of development. The potential market impacts are
considered as quite significant.
The development of IoT platforms is driven by the need
to facilitate machine-to-machine (M2M) connectivity,
which is emerging at unprecedented rate. Machina
Research [2] predicts that M2M connections will rise from
two billion in 2012 to 12 billion in 2020. Cisco [3] values
IoT market to 19 trillion USD. According to the same
source, only 0.6% of the physical objects the potential
candidates for IoT are currently connected. Different
sources refer to estimated 50 billion objects online on
2020.
The previously mentioned Gartner’s study [1] suggests
that the need for IoT platforms is most precisely
articulated in the sector of consumer-centered enterprises,
which are currently classified into early adopters of IoT.
According to Gartner, in 5 years, 2-3 of 10 homes will be
connected homes, with about 500 connectable devices. In
service sector, the increase in overall effectiveness of
employees and workplaces has been sought as the key
expectation from IoT. Intelligent business operations,
machine learning and RFID for logistics and
transportations are foreseen as the earliest facilitators
(within 5 years). Industrial sectors expect a significant
impact of IoT on factory floors; this impact will be driven
in a short term by enterprise manufacturing intelligence
and facilities energy management (within 5 years).
Obviously, all these devices and technologies need a
platform which will act as a command centre in homes,
workplaces and factory floors. Google Nest and Apple
Homekit are some of the examples of “home” aggregators,
IoT platforms capable to implement home automation
functionality. IoT platform (often referred to as IoT
middleware) is thus defined as a software that enables
connecting the machines and devices and then acquisition,
processing, transformation, organization and storing
machine and sensor data.
There exists a strong need for quite diverse sets of
methodologies and tools for effective deployment of IoT
solutions on different scales. In fact, large-scale
deployment of IoT projects under realistic conditions has
been considered only recently [4]. The set of
methodologies and tools deployed in the specific domain,
as the specific IoT solution, is typically referred to as an
integrated IoT platform. IoT platform will be built within
the complex ecosystem of machines, software and people,
dealing with different relevant issues, spanning from
M2M connectivity, to data analytics and visualization, IoT
application development and others. The most important
common features of the different domains are massive
scaling and security.
Main objective of this paper is to define the key
directions in the development of a formal model-based,
generic IoT platform. These directions are presented in the
conclusions of the paper (Section 5). They are made based
on the discussion (Section 4) of the theoretical foundations
for IoT platforms (presented in Section 2); and existing
IoT platforms (Section 3).

II. THEORETICAL FOUNDATIONS FOR IOT PLATFORMS
Stankovic [5] highlighted the following research
directions for IoT:
 massive scaling (addressing, discovery, architectural
models that can support the expected heterogeneity),
 architecture and dependencies (IoT apps,
deployment, resolving interference problem in using
the utility device from different apps by some kind
of multiplexing, dependencies across applications
especially for safety critical apps or when actuators
can cause harm),
 creating knowledge and big data (real-time data
interpretation, knowledge formation, new inference
techniques, trusting data by using confidence levels,
reliable data associations),
 robustness and openness,
 security (detection and diagnosis of attack and
deployment of countermeasures),
 privacy (evaluate requests against policies,
reconciliation of the different policies) and
 humans in the loop (modeling human behaviors,
human use and control).
Obviously, each of the listed directions is highly
relevant for the IoT platform. In fact, the design of one
IoT platform must take into account all these directions
and its development approach must be defined so IoT
platform becomes an enabler of each of the above factors.
As the technologies and standards used for IoT device
manufacturing and communication are still in very early
phase of development and adoption, IoT platforms should
embrace a role of IoT experimentation facilities. Gluhak et
al [4] identified the requirements for a next generation of
experimental research facilities for the IoT:
 scale (supporting thousands of nodes: minimized
human intervention, maximized plug-and-play
configuration, automatic fault management),
 heterogeneity (management of devices, easy
programmability of heterogeneous devices),
 repeatability (across different test beds: agreements
on standards),
 federation (with other test beds, or other
experiments: common framework for authentication,
interoperability),
 concurrency (virtualization of devices, multiple
experiments for one device),
 experimental environment (robustness towards the
environmental conditions),
 mobility (handling system dynamics, movement of
devices) and
 user involvement and impact (multi-modal
mechanisms for user feedback, automated detection
of situations where user behavior influences the data
validity).
III. CLOUD-BASED IOT PLATFORMS AND SERVICES
In this section 16 different cloud-based IoT platforms
and services are presented, with short overview of their
distinguishing architectural designs and functionality.
Initial selection of platforms was done by a Google
search. Then, based on the analysis of the website content,
the selection is filtered to the platforms with a significant
customer base and partnerships with device manufacturers
and system integrators.
Some platforms without M2M connectivity features are
included, considered as IoT support platforms. Domain-
specific platforms are not presented in this paper. The
platforms are listed in alphabetical order.
Arrayent. The IoT platform is composed of four
components. Connect Agent is a firmware, a lightweight
agent deployed in devices (bulk firmware updates
enabled). It exchanges data with the Connect Cloud by
using 128-bit AES encryption. Each of these devices has
its own digital copy in Connect Cloud which hosts the
virtual devices to which mobile apps can connect. Mobile
Framework is used for development of apps which
manage connected devices. It uses also engine for
managing and sending triggered alerts that could also then
trigger response actions in the product which generated
the alert. Finally, Insights provides secure access to data
via dashboards, batch exports, data streaming and data
connectors.
Axeda. Connectivity middleware facilitate connecting
machines and devices to the cloud. Application
enablement platform simplifies development of IoT apps,
with capabilities such as data management, scripting
engine, integration framework, SDKs and web services for
accessing data and apps in the cloud. Connected machine
management applications facilitate remote monitoring,
management, service and control of remote devices.
Capabilities also include software (client, firmware)
distribution and configuration management.
Bugswarm is a lightweight platform that can acquire
data from and control devices using JavaScript or plain
HTTP. It defines a “swarm” – system of resources which
can communicate to other resources within the system,
according to the defined access policy. A resource is
considered as anything that can communicate through
HTTP, not only devices but also web or mobile
applications. Device-specific, client-side applications,
device connectors are available for use, to connect
existing device as a resource to a swarm. When the
specific device is connected, it sends the private message
to all swarm members, with the list of its capabilities or
services that the device can provide (feeds). Other
resources interested in these services could send a feed
request to a device, which then responds with a feed
response (typically, with sensed data).
Carriots. The platform is an aggregator; it enables
connecting any type of device with web connectivity
which can send a stream of data, by using MQTT, cURL,
hURL or Poster, to Carriots REST API. For each of the
protocols, a client installation is needed on the device.
Then, Listener or Trigger component can be developed
and deployed on platform to perform operations with data.
Device control and maintenance is enabled (checking
status, managing configurations, updating firmware). All
development is being done by using Java Carriots SDK,
by putting a code to the specific fields in Carriots Control
Panel web application (Java code
interpretation/execution). Free use is enabled with limited
functionality (up to 10 devices).
Evrythng is natively a digital identity management
platform, often referred to as “Product Relationship
Management” (PRM) platform. Semantic data store is

used to customize dynamic data profiles – digital identities
of the products, so they can exchange data with authorized
applications.
Exosite is cloud-based IoT platform offering M2M
connectivity and data visualization tools and services.
Open API is available for advanced data processing and
integration with enterprise applications.
GrooveStreams is data analytics cloud platform,
allowing data collection from multiple platforms,
including IoT devices. Open API can be used to send data
streams at a fixed (up to 10 second) or random interval or
as a point stream of a fixed value. Data analytics tools are
offered with near real-time performance. Data can be
redistributed as Derived Steams, or visualized with
customizable charts and graphs. The platform is open
access. Premium features are available, related to number
of organizations, users and increased (scalable) data I/O
rate.
Ifttt (If This Then That) is not a native IoT platform; it
is an interoperability-as-a-service platform which allows
users to create chains of simple conditional statements,
called “recipes”, which are executed upon the particular
events recorded from the different services. It is the
platform which enables users to create their own recipes,
which can also include events from the different devices.
Some of the existing examples of IoT related recipes are:
“delay watering your garden if it’s going to rain
tomorrow”, “receive and emergency call if smoke is
detected”, and others. There also exist alternatives to this
service, such as Zapier and Yubnub.
Kaaproject is open-source IoT middleware platform. It
enables management and maintenance (firmware updates
distribution) of device inventory and near real-time
communication between the devices. It is transport-
agnostic and promotes use of structured data. It provides
endpoint SDKs that can be embedded into devices.
Complete solutions already exist for Android, IoS,
Raspberry Pi and other platforms. It is pre-integrated with
existing data processing solutions, such as mongoDB,
Hadoop, Oracle and others.
LinkSmart open source middleware platform is a
framework and a service infrastructure for creation of IoT
applications, originally developed by Hydra EU project.
The project is hosted by Fraunhofer FIT. It includes
Device Connector for integrating devices (with different
implementations for specific devices), Resource Catalog
for managing devices and resources they expose, Service
Catalog (services used to access devices and resources)
and GlobalConnect tunneling service that enables access
to devices beyond the boundaries of a private (routable)
network.
Mbed platform aims at even tighter integration, by
treating all its connected devices as embedded devices.
They all have in-house mbed open-source Operating
System, event-driven single-threaded architecture (scales
down to the simplest, lowest cost, lowest power
consumption devices). Mbed supports only devices based
on ARM Cortex-M microcontroller. Key principles are
security, connectivity and manageability (uses OMA
Lightweight M2M, a popular protocol for monitoring and
managing embedded devices). The architecture consists of
mbed OS, Device connector (works with REST APIs),
TLS (includes cryptographic and SSL/TLS capabilities in
embedded products), client (library that connects devices
to device connector service and mbed server – free, high-
level C++ API) and server (essentially, a middleware, also
hosted as a cloud service, connects IoT devices to web
applications).
Nimbits is data logging service and rule engine
platform for M2M connectivity. It provides nimbits.io
open source Java library for developing Java, web and
Android solutions that use Nimbits Server as a backend
platform. Backend platform collects geo and time-stamped
data and executes rules on this data, such as calculations,
email alerts, xmpp messages, push notifications and
others. Free and Enterprise editions of the server are
available.
Particle.io (former Spark.io) offers hardware
development kits for building the firmware for the
devices, by using web-based IDE and deploying this
firmware over the air. Then, ParticleJS and Mobile SDK
libraries can be used to build web and mobile apps, based
on the collected data.
Autodesk SeeControl is IoT cloud service which
virtualizes machines, links them with reporting devices
and use analytics to unlock their data. No-coding, drag-
and-drop approach is implemented. Platform is focused to
the needs of the manufacturing industries, in specific to
generating product performance data, predicting a product
failure, performing maintenance and optimizing supply
chain and material replenishment costs. It provides a large
library of existing protocol/vendor device adapters.
Service also includes light ERP modules and business
management tools.
SensorCloud is a cloud IoT platform for acquisition,
visualization and analysis of data. The platform natively
supports connectivity with LORD MicroStrain’s wireless
and wired sensors. Visualization tools are available. It is
possible to setup simple alerts, triggered by the data
threshold values. MathEngine analysis tools are provided,
with a simple interface which facilitates common
operations such as FFTs, smoothing, filtering and
interpolation.
PTC ThingWorx. In the platform, each device is
represented by so-called Thing Template. Template
defines properties (for example, temperature), services
(for example, posting to Facebook) and events (for
example, malfunction). Devices use agents to connect to
IoT platform; different agents are used for the different
types of devices. Composer application is used to model
the things, business logic, visualization, data storage,
collaboration and security required for IoT application.
Mashup can be assembled by using different thing
templates, namely, UI widgets which are pre-wired to the
thing templates. The mashups are then used as interactive
IoT applications, real-time dashboards, collaborative
workspaces and mobile interfaces. BPM component is
included to enable definition and execution of the
processes, starting with an alert or event from a remote
connected device. Device asset management tool is also
included to facilitate remote diagnostics, control and
scheduled software update of things. Free use is possible,
with limited functionality.
ThingSpeak is another IoT platform, with features very
similar to SensorCloud. It features open channels with
available data from different devices, published by the
users. Platform enables actuation, namely talking back to
the device, which is done over HTTP.

IV. DISCUSSION
Today’s IoT platforms are typically cloud based,
delivering PaaS (Platform-as-a-Service). Communication
with devices is made by installing or developing light
clients, which only purpose is to facilitate connectivity to
the central platform. Pervasive (over any communication
channel, including cellular, WiFi and satellite) and reliable
(where connection quality is considered as a primary
criteria) connectivity is a key feature.
Typical features of IoT platforms are: connectivity as a
service, monitoring and maintenance of devices (including
firmware updates), data visualization, data analytics, basic
application logics through alerts and triggers.
Connectivity as a service is achieved by enabling the
unconditional (with installed client) access to devices, no
matter if they are located behind the firewall, a NAT or
mobile network router. The service should work with any
device that provides a TCP socket.
The following categories of existing IoT platforms were
identified:
 Domain-specific platforms are the IoT platforms
which facilitate specific domain scenarios. Often,
these platforms are built on the top of the generic
M2M connectivity providers. Examples of such
platforms are rachio for smart irrigation, nest for
home automation, getcleverpet, fishbed and others.
 Technology-specific platforms are the platforms
which take into account only specific set of
devices. These platforms are sometimes closed, in
the sense that they are based on the devices with
proprietary technology. Examples of such
platforms are Mbed, which supports only devices
based on ARM Cortex-M microcontroller, Zatar,
Nest and others.
 M2M connectivity providers offer connectivity as a
service as a core service, with only a few other
features, mostly related to data analytics. Their
primary objective is data acquisition and analysis.
 Full scale generic IoT middlewares (for example,
ThingWorx) provide full range of connectivity
services, but they also facilitate the application
development, based on data collected by the
devices and transformed by analytical tools. Such
development is possible by using integrated
development environments (IDE), API’s or even
language interpreters.
 Some platforms offering supporting services were
considered as important. They do not offer M2M
connectivity services, so they are not IoT platforms.
However, they offer functionality which can be
useful for IoT scenarios. Examples of such
platforms are GrooveStreams, a data analytics
cloud platform, and ifttt, interoperability-as-a-
service platform.
With the rise of IoT platforms, cross-platform
interoperability and reuse is starting to emerge. There are
cases where the domain-specific IoT platforms are made
by using M2M connectivity providers. For example,
getcleverpet, groovelabs.io and fishbit are implemented by
using particle.io. Similarly, first cases of collaboration
between platforms appear, with interoperability solutions.
For example, ThingSpeak platform is connected to
realtime.io. Further collaboration may be facilitated by
increasing number of stable open-source solutions. In fact,
they already exist on market (Kaaproject, LinkSmart)
offering significant opportunities for development of
complex systems over existing core communication
management and communication platform. Finally, IoT
ecosystem will certainly benefit from the further
development of supporting services (interoperability-as-a-
service, storage-as-a-service, data analytics and
visualization, etc.).
HTTP and REST will probably be the way to
communicate with devices, as well as between platforms.
Guinard et al [4] found that RESTful Web Services are
more suitable for programming access to IoT devices, than
WS-* service architecture. However, it is highlighted that
the latter are better choice when complex requirements
related to advanced security and QoS are considered.
IoT at scale is a crucial technical requirement for
further development, scalable performance (concurrency),
storage and connectivity being the most important topics.
Scaling at a client side is considered with least priority;
devices are still the most sensitive and weakest
components of IoT architecture in terms of reliability
(power consumption).
A. Issues and challenges of current cloud IoT platforms
Centralized approach to managing IoT ecosystem may
pose challenges to the devices’ reusability in multiple
contexts/applications, due to potential conflicts between
clients. It may also affect the future of IoT ecosystem
architecture and pose an approach characterized by
exclusive ownership over a device, where services and
data of this device would be offered through a central
platform that controls that device, forming a network of
networks. This may lead to application silos, with
potential risk of interoperability issues. Is P2P
communication between devices worth considering as an
alternative, even through their digital identities stored in a
cloud (like in Arrayent platform)? Obviously, current
devices’ energy consumption issue makes the case for
centralized approach, but for how long?
Though convenient, turnkey solutions, such as the
group of full scale generic IoT platforms may affect the
development of IoT market. Namely, when considering
current cloud computing applications, the only feature that
distinguishes IoT platforms from the other cloud services
platforms is M2M connectivity as a service. All the other
services, such as analytics, visualization and application
development are value-added, non-IoT core services and
they may be provided by the third-parties. However, the
use of non-core services by the M2M connectivity
providers is not at the significant rate.
Finally, the third major issue is lack of support to
complex data structures and business logics (beyond the
level of simple triggers or rules), to be used for the
development of applications, based on collected and/or
transformed data.
V. CONCLUSIONS
Although Gartner’s analysis of the emerging
technologies positions IoT platforms at the very early
phase of development, experiences from this survey show
that cloud-based M2M connectivity services offer is well
established. Some characteristic market niches are already
recognizable, namely, M2M connectivity, data storage and

analysis, data visualization, Interoperability-as-a-Service
and others. What is clearly missing at this point is IoT
ecosystem application building environment.
While the objective of this paper was to identify the gap
in the current state of art of IoT platforms, comparing to
the theoretical foundations and vision of IoT, its
motivation was to setup the novel design of IoT platform
which core feature will be exactly application
development.
Based on the survey, following main principles for the
development of formal model-driven IoT software
execution platform (InoTEP) are defined:
- InoTEP is web application for devices in IoT
which enables composition and realization of IoT
scenarios, by using peer-to-peer approach (multiple
InoTEP instances installed on multiple devices,
communicating over REST).
- InoTEP provides Application-as-a-Service service
which will interpret any formal model
(RDF/RDFS/OWL ontology) in a runtime and
deliver CRUD (create/read/update/delete data)
application.
- InoTEP enables formal definition of the device’s
capability to sense and/or actuate, by using
Capabilities ontology.
- InoTEP uses RDF as a transport protocol for
communication between devices (over REST).
- InoTEP tries to match any data received through its
own REST interface (external data), with domain
and capabilities ontologies.
The above listed principles are further used in selection
of the enablers of the key components of InoTEP.
Application-as-a-Service component will be implemented
by using OntoApp system [6]; W3C Sensor ontology [7]
is being extended to develop a Capabilities Ontology;
Active Semantic Model [8] approach will be used for a
matching engine.
ACKNOWLEDGMENT
This work presented in this paper has been co-funded
by the projects VIHOS (III41017, supported by the
Government of Republic of Serbia), H2020-644715
“AquaSmart”, supported by the European Commission
and “Towards Context-aware Smart Cyber-Physical
Ecosystems” (451-03-01765/2014-09/23, supported by
Bilateral scientific co-operation programme - Serbia and
Portugal).
REFERENCES
[1] Stankovic, J.A. (2014) Research Directions for the Internet of
Things. Internet of Things Journal, 1(1)3-9
[2] Machina Research press release, Machine-to-Machine connections
to hit 12 billion in 2020, generating EUR714 billion revenue.
https://machinaresearch.com/static/media/uploads/machina_resear
ch_press_release_m2m_global_forecast_analysis_2010_20.pdf
[3] Nair, S., Why is “Internet of Everything” a future growth driver
for Cisco? http://finance.yahoo.com/news/why-internet-
everything-future-growth-130020843.html
[4] Gluhak, A., Krco, S., Nati, M., Pfisterer, D., Mitton, N. et al. A
Survey on Facilities for Experimental Internet of Things Research.
IEEE Communications Magazine, Institute of Electrical and
Electronics Engineers, 2011, 49 (11), pp.58-67
[5] Stankovic, J.A. (2014) Research Directions for the Internet of
Things. Internet of Things Journal, 1(1)3-9
[6] Zdravković, M., Trajanović, M. (2015) On the Runtime Models
for Complex, Distributed and Aware Systems. 5th International
Conference on Information Society and Technology (ICIST 2015),
March, 8-11 2015, Kopaonik, Serbia. In: Zdravkovic, M.,
Trajanovic, M., Konjovic, Z. (Eds.): ICIST 2015 Proceedings,
pp.235-240, 2015
[7] Compton, M., Barnaghi, P., Bermudez, L., Garcia-Castro, R.,
Corcho, O., Cox, S., Graybeal, J., Hauswirth, M., Henson, C.,
Herzog, A., Huang, V., Janowitz, K., Kelsey, D., Le Phuoc, D.,
Lefort, L., Leggieri, M., Neuhaus, H., Nikolov, A., Page, K.,
Passant, A., Sheth, A., Taylor, K.: The SSN Ontology of the W3C
Semantic Sensor Network Incubator Group, Web Semantics:
Science, Services and Agents on the World Wide Web, Vol. 17,
25-32. (2012)
[8] Stojkovic, M., Trifunovic, M., Misic, D., Manic, M. (2015)
Towards Analogy-Based Reasoning in Semantic Network.
Computer Science and Information Systems, 12(3) 979–1008