PreprintPDF Available

MAIA: A Microservices-based Architecture for Industrial Data Analytics

Preprints and early-stage research may not have been peer reviewed yet.


In recent decades, it has become a significant tendency for industrial manufacturers to adopt decentralization as a new manufacturing paradigm. This enables more efficient operations and facilitates the shift from mass to customized production. At the same time, advances in data analytics give more insights into the production lines, thus improving its overall productivity. The primary objective of this paper is to apply a decentralized architecture to address new challenges in industrial analytics. The main contributions of this work are therefore two-fold: (1) an assessment of the microservices' feasibility in industrial environments, and (2) a microservices-based architecture for industrial data analytics. Also, a prototype has been developed, analyzed, and evaluated, to provide further practical insights. Initial evaluation results of this prototype underpin the adoption of microservices in industrial analytics with less than 20ms end-to-end processing latency for predicting movement paths for 100 autonomous robots on a commodity hardware server. However, it also identifies several drawbacks of the approach, which is, among others, the complexity in structure, leading to higher resource consumption.
MAIA: A Microservices-based Architecture
for Industrial Data Analytics
Hai Dinh-Tuan
Service-centric Networking
Telekom Innovation Laboratories
Technische Universit¨
at Berlin
Felix Beierle
Service-centric Networking
Telekom Innovation Laboratories
Technische Universit¨
at Berlin
Sandro Rodriguez Garzon
Service-centric Networking
Telekom Innovation Laboratories
Technische Universit¨
at Berlin
Abstract—In recent decades, it has become a significant ten-
dency for industrial manufacturers to adopt decentralization
as a new manufacturing paradigm. This enables more efficient
operations and facilitates the shift from mass to customized
production. At the same time, advances in data analytics give
more insights into the production lines, thus improving its
overall productivity. The primary objective of this paper is to
apply a decentralized architecture to address new challenges
in industrial analytics. The main contributions of this work
are therefore two-fold: (1) an assessment of the microservices’
feasibility in industrial environments, and (2) a microservices-
based architecture for industrial data analytics. Also, a prototype
has been developed, analyzed, and evaluated, to provide further
practical insights. Initial evaluation results of this prototype
underpin the adoption of microservices in industrial analytics
with less than 20ms end-to-end processing latency for predicting
movement paths for 100 autonomous robots on a commodity
hardware server. However, it also identifies several drawbacks of
the approach, which is, among others, the complexity in structure,
leading to higher resource consumption.
Current manufacturing systems are still mostly based on the
5-layer architecture (ISA-95 model) [1], which was originally
developed to ease the management of interfaces between
enterprise application systems and manufacturing controllers.
However, the top-down decision making process in this model
is no longer suitable for future factories, in which the concept
of Fog Computing pushes more decisions to be made at
the lower levels of the infrastructure. At the same time,
the fierce competition among manufacturers force them to
build more agile and responsive production systems. These
factors have led to many projects worldwide to develop new
manufacturing techniques, focusing on four main strategies,
as Franciso [2] summarized: Decentralization (both physically
and logically); Vertical integration, connectivity and mobile;
Cloud Computing; and Advanced analysis.
This work contributes to such strategies by evaluating a
decentralized software architecture for industrial analytics.
Together with our industry partners, we have identified a
use case as depicted in Fig.1, illustrating a manufacturing
factory with multiple zones. In each zone, we set up a fog
computing infrastructure composed of one access point (AP)
and one server (SV) in close proximity. An autonomous robot
connects to the closest access point and, due to the limitation
Sensor Data
Zone 1 Zone 2
Sensor Data
Is robot moving from
Zone 1 to Zone 2?
Pre-transfer service from Server 1 to Server 2
Fig. 1. The use case of predictive analytics for low-latency fog computing
used to evaluate our architecture. As an autonomous robot moves from Zone 1
to Zone 2, the proposed system is expected to be able to predict this movement
and notify the fog computing infrastructure.
of computing resources and battery life, off-loads the object
recognition tasks to the server by streaming the video captured
from its camera. To avoid collisions with other robots or
objects in the shop floor, the processing latency must be always
kept as low as possible. However, as the robot moves, it might
connect to other access points along the way and the latency
increases (in this scenario, the robot is moving from Zone 1
to Zone 2, thus switching from AP1 to AP2). The low-latency
requirement is guaranteed only when the processing service
will be dynamically moved to another server that is closer to
the robot’s new location, i.e. SV2, with minimal interruption.
Therefore, we propose setting up a microservice-based sensor
data analytics component to predict robot’s movement path
and the next access point it will connect to by analyzing
location data from the robots in real-time. These predictions
are then transferred back to the fog computing infrastructure,
thus allowing parts of the service to be transferred to the new
location in advance and reducing the overall interruption.
The rest of this paper is organized as follow. In Section
II, we identify three major trends in industrial analytics and
how microservices fit into that developments. Various efforts
to apply microservices in industry in general and in manu-
facturing in particular are reviewed in Section III. Section
IV describes the design criteria, design decisions, and our
proposed architecture in detail. More about the implementation
978-1-5386-8500-6/19/$31.00 ©2019 IEEE 23
of our prototype are given in Section V. The evaluation of this
prototype and the key results are discussed in Section VI. We
conclude the findings and suggest some directions for future
research in Section VII.
Recently, advanced data analytics have been increasingly
applied across industries, covering the whole process of ex-
tracting useful knowledge, discovering patterns, and predicting
trends from the collected data [3]. Analytics, especially cloud-
based analytics will play a more significant role in the near
future, driven by three important trends.
First, alongside the increasing adoption of Industrial Inter-
net of Things (IoT), the number of sensors, and its measured
data will continue to multiply in volume, velocity, and com-
plexity [4]. On one hand, it will be a valuable source of data,
providing even more insights into underlying processes. On the
other hand, however, this huge amount of information poses
significant challenges for analytical systems, especially when
the time constraints are taken into account.
Since more data has already been made available, there is
a shift of focus from data collection toward data analytics.
Data analytics in industry can transform data into informa-
tion, knowledge, and insights, enabling multiple forms of
data-driven intelligence. However, the heterogeneous nature
of sensor data, the highly distributed data sources, and the
security requirements raise several challenging issues for the
computing platform. New researches are required to make
current infrastructures more scalable, flexible, and robust.
Third, as the data analytics techniques become more com-
plex, more computing power is required, motivating manu-
facturers to adopt cloud technologies. This technology en-
ables a number of innovative manufacturing models, such as
distributed small-scale local manufacturing, loosely coupled
manufacturing ecosystems, and agile manufacturing, paving
the way for mass customization, which is identified as the
new nature of consumer demand [5]. Again, the adoption of
cloud computing raises challenges in guaranteeing the Quality
of Service, data security and privacy, etc.
At the same time, microservices has gained attractions
from both the industry and academia due to the adoption
at multimillion users companies such as Amazon or Netflix.
According to [6], a microservices-based application is a single
application, composed of a set of services. These services are
not only small in size but also highly decoupled, indepen-
dently replaceable, upgradeable, and deployable [7], designed
following the well-known UNIX principle ”Do one thing and
do it well” [8]. Despite being highly independent, i.e., each
service has its own process and data management mechanism,
there is an extensive communication flow between them. These
interactions are not done using internal calls, but rather through
clearly defined interfaces, such as HTTP-based API. Similarly,
various types of clients can interact with a microservice-based
application using protocols such as REST.
Our hypothesis is, the distributed, and cloud-native nature
of microservices make it a potential approach for industrial
analytics. First, by decomposing an application into sev-
eral smaller services, microservices can achieve fine-grained
scalability, allowing certain services to be scaled up/down
while keeping other services intact. This means, compared to
traditional monoliths, microservices can be easily scaled in
different deployment strategies (in both cloud and edge net-
works), providing manufacturers more flexibility to optimize
the latency, energy, and resource utilization.
Also, by decomposing a software into multiple smaller
deployment units, the interdependence among microservices
is minimized. Each microservice can be deployed or updated
without redeploying the whole application. For example, new
components processing new sensor data types with different
analytics techniques can be added into an existing system with-
out any interruption. Also in case several components break,
the isolation among services can help preventing cascading
failures. Therefore, the microservices architecture improves
the availability as well as the robustness of critical systems
in manufacturing environments.
Third, the autonomy of microservices allows the most ap-
propriate technology to be applied where it can solve problems
most effectively. This capability is also referred to as technol-
ogy heterogeneity or polyglot programming and persistence,
allowing different technologies to coexist in an application.
Not only the manufacturer can maximize the efficiency of each
individual process, but also the developers can select the tool-
set that they are most familiar with to improve the quality and
the speed of service deployment.
From a business and management perspective,
microservices-based software can be delivered continuously
using agile development methodologies and cycles, allowing
manufacturers to react quickly to changing customers and
market needs. Small- and medium-sized factories also benefit
from this architecture as it enables them to leverage external
expertise and resources through outsourcing without revealing
the core business functions [9]. This feature realizes the
vision of smaller factories located closer to the customer,
thus be able to quickly deliver individualized products that
meet customers’ unique requirements.
On the other hand, we should take into consideration several
drawbacks of this concept. First of all, as a distributed model,
microservices introduce its own set of complexities, for exam-
ple, additional components to orchestrate and monitor services
are required. Developers also have to face the typical problems
of a distributed system such as data management or complex
communication patterns. Moreover, more components needs to
be implemented and deployed exposes more potential failure
points, requiring sophisticated measures to isolate the failures
and automatically restore failed components.
In addition, an application with multiple independent mi-
croservices is expected to consume more hardware resources
than a single monolith. That is because each microservice
runs in a separated process or thread, or even in a separate
virtual machine (VM) such as Java virtual machine (JVM).
Today, a common practice to fasten the deployment process
and improve the independence on underlying platform is to
adopt container or virtualization technologies. However, this
strategy requires even more hardware resources.
Ciavotta et al. [10] propose a simulation-based architec-
ture for Cyber-Physical Systems (CPSs)1at shop-floor level,
providing an environment for Digital Twins (DTs)2along the
whole plant life-cycle. The proposed platform implements a
microservice IoT-Big Data architecture supporting the publi-
cation of multidisciplinary simulation models and managing
streams of data coming from the shop-floor for real-digital
synchronization. Microservices architecture is applied in their
support infrastructure, in order to manage the DTs. In our
proposal, we extend this work by employing microservices
also for simulating physical assets to decentralize the whole
system. Rather than storing all digital copies of physical assets
in one place, building them as microservices allows more
flexibility in deployment strategies. For each service, a best
physical location to deploy could be determined based on
various criteria. This is an important foundation to enable
automated and QoS-aware deployment strategies.
Thramboulidis et al. [12] describe a 5-layer framework
based on microservices for manufacturing systems. Each phys-
ical unit of the plant is transformed to a smart entity, named a
Cyber-Physical Microservice (CPMS). The authors define two
main types of CPMS: Primitive CPMSs, which encapsulates a
physical artifact and transform it to a smart entity, and Com-
posite CPMSs, which utilizes at least one Primitive CPMS.
In this work, the authors focus on the architecture of each
individual CPMS and evaluate the overhead of microservice
orchestration. Our paper is complementary to this work, as it
proposes a complete architecture for utilizing CPMSs.
NIMBLE Collaborative Platform [13] adopts microservices
architecture to build a collaborative Industrie 4.0 platform
that enables IoT-based real-time monitoring, optimization and
negotiation in manufacturing supply chains. The microservices
in the architecture provide beyond the core business function-
alities essential supporting services such as Gateway Proxy,
Service Logging, Service Monitoring, Service Discovery, Ser-
vice Configuration, Identity Management. The implementation
uses either REST (i.e. HTTP) or messaging as the mean
of service communication. The authors describe from an
architectural viewpoint how they use microservices to build a
platform processing incoming data but they don’t address the
scalability of the platform. In our proposed architecture, we
incorporate the self-management capability for the system by
introducing a new component named Life-cycle management.
Its main responsibility is to monitor the load level of each
microservices, and scale up or down a specific microservice
1Cyber-Physical systems are systems with the seamless, real-time interac-
tion between computing elements and physical assets using intelligent data
management, analytics and computational capability [4].
2Digital Twins are representations of real-world assets (including the
assets in designing/building stage) created with the ability to collect and
synthesize data from various sources including physical data, manufacturing
data, operational data and insights from analytics software [11].
based on various criteria such as memory consumption, in-
coming message queue length, etc.
In [14], the authors propose a microservice-based reference
architecture for the Enterprise Measurement Infrastructure
(EMI) . In this architecture, there are six main components,
which are Visualization (dashboard tools to interact with
users), Calculation and Storage (provides aggregated informa-
tion for Visualization layer), Data Transport and Integration
(builds a common communication infrastructure for all ser-
vices), Data Provider (feeds data to the system), Data Adapter
(convert received data into readable formats) and Operation
(contains services that ease operating and monitoring the
application). This architecture is a basic design, as it is built to
exploit most prominent advantages of microservice: high level
of isolation between services, robust again complete system
failure, and the support for the integration of heterogeneous
systems. However, this architecture again does not address the
scalability of a microservice-based application.
The Reference Architecture Model Industrie 4.0 (RAMI
4.0) [15] is a three-dimensional model for service-oriented
architecture, combining the life-cycles of products, factories,
and machinery with the hierarchy levels of Industrie 4.0. The
authors define six layers, from top to bottom are: Business,
Functional, Information, Communication, Integration, and As-
set. The Integration layer is the connection between the digital
world (four upper layers) an the real world (the bottom layer).
As RAMI is developed as a generic framework for industrial
applications in general, a more concrete architecture designed
for decentralized analytics will provide new perspectives by
addressing specific challenges of the concept.
A. Design criteria
This work focuses on applying the decentralization
paradigm in manufacturing as an approach to be flexible
and able to react faster to market and customer demands.
Indeed, in [16], the authors have categorized four paradigms
for manufacturing systems, all of them are only possible when
the hardware and software components are reconfigurable,
which encourages the adoption of modular structures and open
architectures. Besides decentralization, we also consulted the
Industrie 4.0 design principles [17] in order to compile a list
of additional design criteria as below:
Scalability: An industrial application should be flexible
enough to scale in different dimensions: the number of nodes,
functionalities, applications, and the data volume size, among
others. Unlike other networks, a network of industrial devices
can scale up to millions of connected points. This leads
to several architectural requirements, such as selective data
transmission (ensure the data can only be sent to the necessary
parts of the system) and a well-defined set of interfaces to
support fast system integration.
Low-latency capability: Low-latency computation is an
important requirement for industrial applications in general.
Many applications are realized only with on-time decisions,
for example, autonomous robots normally require latency as
low as tens of microseconds [18]. In a distributed computing
model, this requirement is challenging to fulfill, since the time
required for processing might be affected by external factors,
such as network conditions or the communication mechanisms
between components [19].
Interoperability: One primary requirement for the design
is a high level of interoperability, which is, among others,
reflected in the use of open standard protocols or interfaces
and multi-platform technologies such as HTTP or HTML. In
addition, the application should be able to be deployed with
ease on various run environments.
Fault-tolerance: Interruptions are unacceptable for critical
mission software, therefore, the design must incorporate some
recovery mechanisms after failures.
Usability: The application must provide intuitive methods
for users to interact with. Monitoring capabilities must be
available and provide useful operation’s statistics in real-time.
B. Design decisions
Considering the previously mentioned requirements,
our proposed MAIA architecture (Microservices-based
Architecture for Industrial data Analytics) is designed based
on six key design decisions:
The decentralized nature of the design is reflected in the
adoption of microservices architectural style itself. We define
two main type of services: Functional services and Infrastruc-
ture services.Functional services are services that support spe-
cific business operations or functions, whereas Infrastructure
services support nonfunctional tasks such as authentication,
authorization, auditing, logging, and monitoring. This is an
important distinction, because Infrastructure services are not
exposed to the outside world but rather are treated as private
shared services only available internally to other services. In
contrast, Functional services provide their services externally.
We applied Domain-driven design and the Bounded context
concept to define microservices with minimal inter-services
dependency. In our design, each physical machine in the
factory has a digital replica with all the related data and
processes represented by a microservice, similar to the concept
of CPMS proposed in [12]. Although this design requires
more hardware resources comparing to have a single service
to handle multiple manufacturing units, we can tailor the
service to meet the specifications of physical entities and better
support the portability of these virtual representation. While
simple analytics and monitoring tasks can be done in each
individual digital representation of physical assets, the data
is distributed across different digital replicas. To solve more
complex problems such as optimization or predictive analysis,
aggregated data from multiple machine is required. Therefore,
we design two level of analytics, corresponding to two level
of knowledge: individual analytics and global analytics.
The main mechanism for the inter-service interaction is
asynchronous messaging to decouple services. This means ser-
vices do not communicate directly with each other, but rather
by sending messages via a message broker. One advantage of
this design is that the services don’t need to wait for other to
response to a request. Additionally, they don’t need to know
exact location of other service instances. However, this design
leads to a single point of failure, which should be compensated
by deploying the message broker in high availability clusters.
Inspired by the EMI reference architecture, we apply the
database per service pattern. Together with the asynchronous
messaging, this helps achieving a high level of autonomous
and independence between microservices. This design also
allows developers to incorporate some self-healing capabili-
ties into their system, because in the event of failure, each
microservice can independently restore its data, without af-
fecting data from other services. However, a distributed data
storage model would lead to some problems in ensuring data
consistency across services. To solve this issue, we encourage
developers to adopt the Eventual Consistency and Event-driven
model. In this concept, whenever a microservice needs to up-
date its database, it also publishes an event to an event message
queue. By subscribing to that queue, other microservices are
informed and can update their own databases accordingly.
To enable the fault-tolerance capability of the architecture,
we incorporate two mechanisms. First, a monitor service is set
up to check the health status of each deployed microservice
and restart it if error occurs. On top of that, the Circuit
Breaker pattern is applied to specify a fallback method for
microservices, thus gracefully degrading the functionalities of
the application and avoiding cascading failures.
We deploy each microservice as a containerized unit, which
can minimize the burden of incompatibility between platforms.
Using container technology also adds an additional layer of
management on top of our architecture, allowing the incor-
poration of third-party solutions for visualization, monitoring,
and management.
C. MAIA Architecture
The MAIA architecture consists of four main building
blocks: Physical Space,Digital-Physical Integration,Digital
Space, and Infrastructure Services. All components, except the
Physical Space, are composed of multiple microservices as
shown in Fig. 2:
Physical Space: The first component in this building block
is sensor-equipped machines in the shop floor. They are con-
nected with upper components via distributed gateways using
HTTP REST API or a messaging protocol. The interactions
are done bidirectionally: The machines report various types of
data and received commands, configurations adjustments, etc.
from the analytics system in return. The human operators are
also part of this MAIA component, however, they interact with
the system via a web-based interface with reports, summary,
visualized statistics in order to quickly gain insights into
processes in the shop-floor. The operators can also change
the configurations of machines directly via the same interface.
Digital-Physical Integration: This is the interface for all the
interactions between the physical and virtual world, consisting
of two bidirectional interfaces: Web-based interface for human
interaction, providing visualizations, reports, summary, etc.;
and API Gateway provides physical entities a secured channel
Physical Space
Digital-Physical Integration
Digital Space
Data Storage
Digital Models
Infrastructure Services
Robots Human
Raw Data Knowledge Base
Gateway Service
Path Prediction Service
Web Interface
DT Monitoring
DT Monitoring
HTTP MQTT messaging
Fig. 2. Our proposed Microservices-based Architecture for Industrial Data Analytics and the prototype implemented.
to exchange data with the system. With the API Gateway,
the design provides an abstraction level for the underlying
connection technologies. Since various types of connections
are installed in modern factories, from wired to wireless con-
nections, an abstraction layer helps decoupling the architecture
from physical connection technologies. Each driver for one
connection technology can be implemented as a microservice,
keeping other interactions unchanged. Other functions such
as access control, authentication, and authorization are also
implemented in this building block.
Digital Space: This component stays at the heart of the en-
tire architecture, providing representations of physical equip-
ment and related processes in virtual world. There are three
sub-components in this space, as explained below.
The Data Storage provides a distributed big data storage
infrastructure, with the ability to store a big amount of data,
with high rates of random write and access, flexible enough
to manage different data models, supporting both structured
and unstructured data. Raw data collected from the physical
systems is stored in the Raw Data and distributed to the
corresponding microservices. The knowledge extracted from
analytics processes is synthesized in the Knowledge Base,
accessible for services in the Digital-Physical Integration.
In Digital Models, the virtual representations (Digital
Twins) of the physical devices are built and modeled with all of
their dynamics. To ensure their accuracy, these representations
contain a model of the physical artifact combined with its
collected data. These digital models are structured in single
component level and composed component level, similar to
how various components compose a machine in physical
world. Individual analytics, which can be done using data from
one physical entity, are also performed here and their results
are collected at the Knowledge Base. Depending on the types
of analytics, (requirements of processing power, processing
time, and data privacy), this digital copy of assets can be
deployed either at the network edge (low latency, context
awareness), or in cloud infrastructure (slower but posses more
computational resources). Simple management functionalities
for digital models such as monitoring, inventory management,
etc. are also included here.
Advanced analytics are performed in the Aggregated An-
alytics. Such analytics require data from more than one
physical entity, for example global optimization. Although
these analytics require more time to perform, their outputs are
essential for strategic decisions. Again, similar to short-term
analytics made in each individual digital twin, the knowledge
is communicated to the Knowledge Base.
Infrastructure Services: While the three components men-
tioned above are designed to be compatible with the RAMI
4.0 [15], this component provides specific functionalities for
a microservices-based application, including but not limited
to service discovery, internal communication services, service
logging, and monitoring. In addition, we also include the Life-
cycle Management capability as a component responsible for
monitoring performance metrics such as CPU, memory load,
etc. and make decision when to scale up or scale down a
certain service by adding/removing service instances.
To evaluate the proposed architecture, we have implemented
a Java-based prototype for the low-latency fog computing
use case mentioned earlier. The containerized microservices
are categorized into Infrastructure services and Functional
services. The Functional services are implemented as below:
Gateway Service: Corresponding to the Digital-Physical
Integration in the proposed architecture, its main responsibility
is to expose a REST API for robots and access points to
update their data. Upon the receipt of an update, this service
is responsible for (1) persisting this data, and (2) creating an
event to notify other services using Eventual Consistency.
Representation Service: This service implements the Digi-
tal Models component of the MAIA architecture. Each robot is
modeled with a robot ID, current location in latitude/longitude
coordinates and associated AP. During operation, multiple
instances of this service run simultaneously, each processes the
entire data collected from an individual robot and continuously
monitors the distance from the robot to its connected AP. If this
distance exceeds a pre-defined threshold, the service assumes
this robot is leaving its current zone and send a message to
the Path Prediction Service, triggering the prediction process.
Path Prediction Service: This service implements a typical
service in the Aggregated Analytics. This service receives re-
quests from instances of Representation Service and performs
a predictive analysis to predict the movement direction of
the robot. Compared to individual instances of Representation
Service, this service has a global knowledge of APs’ locations
and coverage areas, which then used to correlate with the
robot’s predicted path to find out the next AP the robot
may connect. This service outputs recommendations, which
includes the robot ID and a list of maximum three possible
APs ranked by the recommendation’s confidence.
Knowledge Base Service: This service is responsible for
keeping track of recommendations. It maintains two connec-
tions with (1) the Web-based Interface to display recommenda-
tions received from the Path Prediction Service, and (2) the fog
infrastructures to form a feedback loop for location updates.
Web-based Interface:Provides a portal with visualized
network of microservices, together with runtime metrics such
as health status, memory, cache, system and environment
properties, etc. The recommendations are also displayed here.
The Infrastructure services consist of the following services:
Message Broker:In [12], the authors used Lightweight
M2M based on Constrained Application Protocol (CoAP) as
the main message transportation protocol. In contrast to this
work, we propose to use Message Queue Telemetry Transport
(MQTT) as the main protocol for microservices’ interactions.
Compared to CoAP, MQTT has more sophisticated reliability
and congestion control mechanisms, which becomes signifi-
cant when data is exchanged frequently [20]. In addition, when
the packet loss rate is low, MQTT performs better in terms of
latency [21]. However, it should be noted that CoAP, which is
based on UDP, is more lightweight compared to a TCP-based
protocol like MQTT. The Message Broker manages several
MQTT message queues: Representation service’s queue: Each
instance of the Representation Service has a separate queue
to receive updates from the Gateway Service;Aggregation
queue: delivers messages from the Representation Service to
the Path Prediction Service;Knowledge Based queue: delivers
recommendations from the Path Prediction Service to the
Knowledge Base Service;Data event queue: delivers data
update events created by the Gateway Service to other services,
thus allowing services to synchronized data about registered
robots and access points.
Service Registration and Discovery:This service allows
microservices to register themselves, as well as discover other
microservices dynamically during runtime. In addition, it also
keeps track of deployed microservices and provide APIs for
other management services. This service is implemented with
two smaller components: (1) an entity receives registrations
from microservices and (2) a REST client in each microservice
that registers itself with the registry. To mitigate the single-
point-of-failure risk, we deployed two instances of this regis-
trar, continuously synchronizing data with each other.
Service Monitor and Management:We create a number of
End-to-end processing time (ms)
Number of robots
Gateway Service Representation Service
Path Prediction Service Knowledge Base Service
Fig. 3. End-to-end processing time in different test cases.
HTTP endpoints for each microservices to expose operational
data such as health status or resource consumption level. A
microservice is deployed to gather all these data to monitor
the application and visualize on a interface. This service also
restarts other deployed microservices in case of failure.
Life-cycle Management:This service monitors the message
queue length and automatically scale up the subscribing mi-
croservice if the number of unprocessed messages exceed a
certain threshold. When the demand decreases, this service
also reduces the number of service instances.
Service Logging:The main responsibility of this service is
to gather all application logs from other microservices into a
single place. This service also provides useful statistics regard-
ing the request propagation between multiple microservices.
A. Results
With the evaluation, we want to investigate the performance
of microservices in terms of processing latency, as well as the
overhead of containerization. The containerized prototype is
deployed with Docker Compose in a dedicated, cloud-based
server with commodity hardware (Intel i7-6700, 64GB DDR4,
4TB SATA HDD). Robots’ location updates are sent to the
server via an Internet connection with the frequency of 1Hz.
To evaluate the latency performance of the prototype, we
perform Distributed Tracing by adding a unique end-to-end
identifier to each message, hence enable tracking the process-
ing time of each microservice. 16 test cases with the number
of robots in each case varies from 1 to 150 robots have been
conducted. In each case, the test was performed three times,
and the mean values of processing time were recorded. We
define the processing time of each microservice is the period
between the arrival of a message until that message leaves.
The end-to-end processing time for each request is the total
Size (MB)
.jar file Standard Docker image
Lightweight Docker image
Fig. 4. Overhead of containerization regarding the increased size of
deployable units.
of processing time in each microservice plus the time for
message transportation between microservices, i.e. the period
counted since the request arrival at the Gateway Service until
a recommendation is made and sent to the Web interface by
the Knowledge Base Service.
The recorded results are presented in Fig. 3 with two key
findings: (1) The end-to-end processing time increases from
10.872ms to 45.293ms; (2) The more robots are added to the
system, the more significant the message transportation (from
67.62% up to 86.97% of total processing time in the case of
1 and 150 robots, respectively).
We also evaluate the overhead of containerization microser-
vices using Docker. In this evaluation, we focus on the de-
ployment time and the size of container, as they are indicators
for the portability of microservices. We compare the size and
build time of bare java file (.jar) with standard docker base
image, and a more lightweight docker base image. The results
are shown in Fig. 4 and Fig. 5.
B. Discussion
As shown in Fig. 3, when the number of registered robot
increases up to 100, the end-to-end processing time increases
linearly, but always stays under 20ms. However, beyond this
point, the processing time increases significantly but still less
than 50ms. In the use case, where the robots move with human
walking speed (up to 7km/h), this means the recommendations
are communicated back to the fog computing infrastructure
within the robot travel distance of 4cm and 10cm, respectively.
This result confirms our concern about the latency of message-
based communication mechanism implemented in our proto-
type. When the message queues are filled up with unprocessed
messages, the end-to-end delay will increase non-linearly. In
a more distributed systems, i.e. microservices are deployed in
Build time (s)
.jar file Standard Docker image
Lightweight Docker image
Fig. 5. Overhead of containerization regarding the increased build time of
deployable units.
different physical servers, the significance of message trans-
portation will be even more noticeable. Therefore, the message
queues are important indicators for system’s performance and
it should be continuously monitored to avoid congestion.
Incorporating container technology not only simplifies the
deployment process, but also adds another layer of manage-
ment and opens up opportunities for advanced features such
as service portability. However, it also comes at a cost. As
presented in Fig. 4 and Fig. 5, containerizing services increases
both the size and the build time of deployable units, which in
turn reduces the portability of the service and increases the
deployment/redeployment time. Fortunately, to some extent,
using lightweight base images can soften the issues.
Refer back to the design criteria we specified in section IV,
one of the most noticeable changes in the use of microservices
model is that, the application is decomposed into services.
By separating between individual analytics and aggregated
analytics as well as dedicating a microservice for a physical
entity, the MAIA architecture minimizes the data dependency
among microservices by giving each service their own data,
which is only accessed by external services via API.
The self-management capability is reflected in the applica-
tion’s ability to monitor itself and adapt to load dynamically
during runtime. Although the monitor capability is imple-
mented as an integrated part of the application, it operates
independently from other components and works even when
other services fail. Running statistics are captured in real-time
and used to make scale decisions for services during operation.
Although the asynchronous messaging pattern poses chal-
lenges in guaranteeing the low-latency requirement, there are
several possible solutions. Thanks to the portability of mi-
croservices in our design, individual analytics can be deployed
close to the edge (low latency but limited computing power)
or in the cloud (more powerful hardware but also slower).
Another strategy to improve the latency is the automatically
scale up the services when more requests come, which is done
by the Life-cycle Management component in our design.
The interoperability of the prototype lies in the use of
open standards such as HTTP allowing external entities to
communicate with the application. The container technology
proves itself to be suitable for microservices, as it provides
a straightforward deployment process in different runtime
environments. At the same time, Digital-Physical Integration
is only an abstract layer without any dependency on a concrete
technology, allows additional communication technologies to
be added. Similarly, new analytics techniques can also be
incorporated as new microservices.
The monitor service is able to detect where the failures
happen and recover the corresponding microservices. In case
of stateful services, the states can be recovered from the
persisted data in host servers. In our implementation, this
feature is supported by Docker volume. With the use of Circuit
Breaker pattern, a cascading failure is also prevented.
The primary objective of this paper is to design a distributed
architecture for industrial analytics using microservices. Based
on the design, we also develop a prototype to evaluate the
feasibility of our idea in industrial contexts.
The paper identifies three important trends in industrial
analytics: the increase in volume, velocity, and complexity
of measured data; the shift of focus from data collection
toward data analytics; and the tendency of manufacturers to
adopt cloud-based services. These three trends require a new
approach for highly scalable and flexible applications. Among
others, the microservices architecture promotes the develop-
ment of applications as a set of independent, autonomous, and
self-contained units, which is in line with the trends of future
manufacturing. Therefore, we conclude that microservices is
a potential candidate for data analytics in industrial contexts.
The design and implementation of our MAIA architecture
underpins both advantages and drawbacks of the microservices
architecture. Our evaluation proves that our prototype can
achieve an end-to-end latency of less than 20ms for a scenario
with up to 100 DTs, and less than 50ms for 150 DTs. The
containerization of services does introduce overheads in terms
of building time and service size, but using a lightweight base
image can help minimizing this burden. Qualitatively saying,
the proposed architecture also meets several requirements of
decentralization, scalability, fault-tolerance.
Nevertheless, this paper raised additional questions that
need to be addressed. Among others, an effective dynamic ser-
vice allocation algorithm is required to optimize the system’s
performance in runtime by balancing between computing
power and latency. Also, new messaging techniques should
be developed to further improve the latency of microservices.
This work has received funding by the Federal Ministry
for Economic Affairs and Energy (BMWi), Germany under
grant no. 01MA17008, project Industrial Communication for
Factories (IC4F).
[1] A. ISA, “Isa-95.00. 01-2000: Enterprise-control system integration part
1: Models and terminology,Technical report, ISA, The Instrumentation,
Systems, and Automation Society, 2000.
[2] F. Almada-Lobo, “The industry 4.0 revolution and the future of manu-
facturing execution systems (mes),Journal of Innovation Management,
vol. 3, no. 4, pp. 16–21, 2016.
[3] H. Hromic, D. Le Phuoc, M. Serrano, A. Antoni´
c, I. P. ˇ
Zarko, C. Hayes,
and S. Decker, “Real time analysis of sensor data for the internet of
things by means of clustering and event processing,” in Communications
(ICC), 2015 IEEE Intnl. Conference on. IEEE, 2015, pp. 685–691.
[4] J. Lee, B. Bagheri, and H.-A. Kao, “A cyber-physical systems archi-
tecture for industry 4.0-based manufacturing systems,” Manufacturing
Letters, vol. 3, pp. 18–23, 2015.
[5] M. Vincent and M. Banovi´
c, “The future of manufacturing-precautionary
measures for industry 4.0,” Transformers Magazine, vol. 5, no. 1, pp.
70–73, 2018.
[6] J. Lewis and M. Fowler, “Microservices: a definition of this new
architectural term,” Mars, 2014.
[7] A. Krylovskiy, M. Jahn, and E. Patti, “Designing a smart city internet
of things platform with microservice architecture,” in Future Internet
of Things and Cloud (FiCloud), 2015 3rd International Conference on.
IEEE, 2015, pp. 25–30.
[8] S. Newman, Building microservices: designing fine-grained systems. ”
O’Reilly Media, Inc.”, 2015.
[9] J. Rossberg and M. Olausson, Pro application lifecycle management
with visual studio 2012. Apress, 2012.
[10] M. Ciavotta, M. Alge, S. Menato, D. Rovere, and P. Pedrazzoli,
“A microservice-based middleware for the digital factory,Procedia
Manufacturing, vol. 11, pp. 931–938, 2017.
[11] M. Grieves, “Digital twin: manufacturing excellence through virtual
factory replication,” White paper, 2014.
[12] K. Thramboulidis, D. C. Vachtsevanou, and A. Solanos, “Cyber-physical
microservices: An iot-based framework for manufacturing systems,” in
2018 IEEE Industrial Cyber-Physical Systems (ICPS). IEEE, 2018, pp.
[13] J. Innerbichler, S. Gonul, V. Damjanovic-Behrendt, B. Mandler, and
F. Strohmeier, “Nimble collaborative platform: Microservice architec-
tural approach to federated iot,” in Global Internet of Things Summit
(GIoTS), 2017. IEEE, 2017, pp. 1–6.
[14] M. Vianden, H. Lichter, and A. Steffens, “Experience on a microservice-
based reference architecture for measurement systems,” in Software
Engineering Conf. (APSEC), 2014 21st Asia-Pacific, vol. 1. IEEE,
2014, pp. 183–190.
[15] M. Hankel and B. Rexroth, “The reference architectural model industrie
4.0 (rami 4.0),” ZVEI, April, 2015.
[16] G. Zhang, R. Liu, L. Gong, and Q. Huang, “An analytical comparison on
cost and performance among dms, ams, fms and rms,” in Reconfigurable
manufacturing systems and transformable factories. Springer, 2006, pp.
[17] M. Hermann, T. Pentek, and B. Otto, “Design principles for industrie 4.0
scenarios,” in System Sciences (HICSS), 2016 49th Hawaii International
Conference on. IEEE, 2016, pp. 3928–3937.
[18] S. Schneider, “The industrial internet of things applications and taxon-
omy,Internet of Things & Data Analytics Handbook, pp. 41–81, 2017.
[19] T. Hadlich, S. H¨
ome, C. Diedrich, K. Eckert, T. Frank, A. Fay, and
B. Vogel-Heuser, “Time as non-functional requirement in distributed
control systems,” in Emerging Technologies & Factory Automation
(ETFA), 2012 IEEE 17th Conference on. IEEE, 2012, pp. 1–6.
[20] N. De Caro, W. Colitti, K. Steenhaut, G. Mangino, and G. Reali, “Com-
parison of two lightweight protocols for smartphone-based sensing,” in
Communications and Vehicular Technology in the Benelux (SCVT), 2013
IEEE 20th Symposium on. IEEE, 2013, pp. 1–6.
[21] D. Thangavel, X. Ma, A. Valera, H.-X. Tan, and C. K.-Y. Tan, “Per-
formance evaluation of mqtt and coap via a common middleware,
in Intelligent Sensors, Sensor Networks and Information Processing
(ISSNIP), 2014 IEEE Ninth Int. Conf. on. IEEE, 2014, pp. 1–6.
Powered by TCPDF (
ResearchGate has not been able to resolve any citations for this publication.
Conference Paper
Full-text available
Recent advances in ICT enable the evolution of the manufacturing industry to meet the new requirements of the society. Cyber-physical systems, Internet-of-Things (IoT), and Cloud computing, play a key role in the fourth industrial revolution known as Industry 4.0. The microservice architecture has evolved as an alternative to SOA and promises to address many of the challenges in software development. In this paper, we adopt the concept of microservice and describe a framework for manufacturing systems that has the cyber-physical microservice as the key construct. The manufacturing plant processes are defined as orchestrations of primitive cyber-physical microservices performed by the physical plant layer. Model-driven engineering is utilized to semi-automate the development process for the industrial engineer, who is not familiar with microservices; IoT technologies are used for their integration. A case study demonstrates the feasibility of the proposed approach.
Full-text available
Industry 4.0 dictates the end of traditional centralized applications for production control. Its vision of ecosystems of smart factories with intelligent and autonomous shop-floor entities is inherently decentralized. Responding to customer demands for tailored products, these plants fueled by technology enablers such as 3D printing, Internet of Things, Cloud computing, Mobile Devices and Big Data, among others create a totally new environment. The manufacturing systems of the future, including manufacturing execution systems (MES) will have to be built to support this paradigm shift.
Conference Paper
Full-text available
The Internet of Things (IoT) is being adopted in different application domains and is recognized as one of the key enablers of the Smart City vision. Despite the standardization efforts and wide adoption of Web standards and cloud computing technologies, however, building large-scale Smart City IoT platforms in practice remains challenging. The dynamically changing IoT environment requires these systems to be able to scale and evolve over time adopting new technologies and requirements. In response to the similar challenges in building large-scale distributed applications and platforms on the Web, microservice architecture style has emerged and gained a lot of popularity in the industry in recent years. In this work, we share our early experience of applying the microservice architecture style to design a Smart City IoT platform. Our experience suggests significant benefits provided by this architectural style compared to the more generic Service-Oriented Architecture (SOA) approaches, as well as highlights some of the challenges it introduces.
Full-text available
This paper introduces the concept of a " Digital Twin " as a virtual representation of what has been produced. Compare a Digital Twin to its engineering design to better understand what was produced versus what was designed, tightening the loop between design and execution.
You can have the best coders in the world working in your teams, but if your project management isn’t up to scratch, your project is almost certain to be delayed, to come in over budget, and in some cases to fail entirely. By taking precise control of your application development process, you can make changes, both large and small, throughout your project’s life cycle that will lead to better–quality finished products that are consistently delivered on time and within budget. Application lifecycle management (ALM) is an area of rapidly growing interest within the development community. Because its techniques allow you to deal with the process of developing applications across many areas of responsibility and across many different disciplines, its effects on your project can be wide ranging and pronounced. It is a project management tool that has practical implications for the whole team—from architects to designers, from developers to testers. Pro Application Lifecycle Management with Visual Studio 2012 focuses on the most powerful ALM tool available for the Microsoft .NET Framework: Visual Studio Team Foundation Server. It demonstrates the key concepts and techniques of ALM at first with a guide to the overall methodology, and then delves into architecture and testing--illustrating all of the concepts, tips and tricks using the tools TFS provides. The book serves as a complete guide to the ALM style--with no fluff and many relevant code samples and examples. After reading the book, you will understand how TFS can be used to generate continuous meaningful reporting on your project’s health for the decision makers on your team as well as for your project’s sponsors.
In recent years a considerable effort has been spent by research and industrial communities in the digitalization of production environments with the main objective of achieving a new automation paradigm, more flexible, responsive to changes, and safe. This paper presents the architecture, and discusses the benefits, of a distributed middleware prototype supporting a new generation of smart-factory-enabled applications with special attention paid to simulation tools. Devised within the scope of MAYA EU project, the proposed platform aims at being the first solution capable of empowering shop-floor Cyber-Physical-Systems (CPSs), providing an environment for their Digital Twin along the whole plant life-cycle. The platform implements a microservice IoT-Big Data architecture supporting the distributed publication of multidisciplinary simulation models, managing in an optimized way streams of data coming from the shop-floor for real-digital synchronization, ensuring security and confidentiality of sensible data.
The Industrial Internet of Things (IIoT) includes networked smart power, manufacturing, medical, and transportation. Technologically, the Consumer IoT and the Industrial IoT are more different than they are similar. This chapter presents IIoT applications that are operational Real-Time Innovations. (RTI) Connext Data Distribution Service (DDS) systems. It explains the DDS standard and highlights the breadth of the IIoT challenge. Based on extensive experience with real applications, the chapter suggests a few divisions and explores why they impact the architecture. Each of these divisions defines an important dimension of the IIoT taxonomic model. A taxonomy logically divides types of systems by their characteristics. The first problem is to choose top-level divisions. The chapter concludes that an IIoT taxonomy is not only possible but also critical to both the individual system building and the inception of a true cross-industry IIoT.
Dedicated Manufacturing Systems (DMS), Adjustable Manufacturing Systems (AMS), Flexible Manufacturing Systems (FMS) are the three basic types of manufacturing systems commonly used in manufacturing industries, while Reconfigurable Manufacturing Systems (RMS) is considered as the future of manufacturing systems. In this paper we will give an analytical and qualitative comparison among DMS, AMS, FMS and RMS from the viewpoint of cost, adaptability, complexity, production rate and ramp-up time.