Edge - Cloud Collaborative Framework for

Real-Time Industrial IoT Systems

Jayanth Siva Chidambaram , Amrish Suresh Kumar , Dhanya Sharma

Abstract— The proliferation of Industrial Internet of Things

(IIoT) devices generates vast amounts of data demanding real- execution and data processing.
time processing for critical applications. However, resource-
constrained edge devices often lack the computational power and
storage capacity for complex analytics, while relying solely on the  Efficiently managing limited resources at the edge
cloud can introduce unacceptable latency. To address these while leveraging the cloud's abundant resources.
challenges, this paper proposes an edge-cloud collaborative
framework specifically tailored for real-time IIoT systems. This  Orchestrating complex workflows involving data
framework leverages the synergistic capabilities of edge and acquisition, processing, decision-making, and
cloud computing to enable efficient data processing, intelligent
decision-making, and enhanced overall system performance. We
actuation across the edge and cloud.
outline the architecture of the proposed framework, detailing the
roles and interactions of its key components. Furthermore, we  Ensuring reliable and timely data synchronization
discuss critical aspects such as real-time task scheduling, between the edge and cloud.
resource management with task prediction, and data
synchronization across the edge and cloud. The paper aims to
provide a comprehensive foundation for developing and
deploying next-generation real-time IoT applications.

Index Terms— Edge Computing, Cloud Computing, Industrial

IoT (IIoT), Real-Time Systems, Collaborative Computing,
Resource Management

I. INTRODUCTION

The Industrial Internet of Things (IIoT) is revolutionizing

industrial processes by connecting numerous sensors,
actuators, and other devices, leading to increased efficiency,
productivity, and automation. Real-time data analytics is
paramount in IIoT applications such as predictive
maintenance, process control, and anomaly detection, where
timely insights are crucial for operational safety and
effectiveness.
Traditional cloud-centric approaches, while offering vast
computational and storage resources, often suffer from high
network latency and bandwidth limitations, hindering their Fig. 1. Edge-cloud collaborative framework for real-
applicability to real-time IIoT scenarios. Edge computing has time IoT systems based on mobility prediction and task
emerged as a promising paradigm by bringing computation
synchronization strategies
and data storage closer to the data sources, thereby reducing
latency and improving responsiveness.
The main contributions of this paper include:
Edge-cloud collaborative computing integrates the strengths
of both paradigms. Edge devices perform initial data
 A detailed architecture of an edge-cloud collaborative
processing and time-critical tasks, while the cloud handles
framework for real-time IIoT systems.
more complex analytics, long-term data storage, and global
coordination. This collaboration is particularly beneficial for
 A discussion on prediction-based resource
real-time IIoT systems, enabling low-latency responses at the
deployment at the edge to enhance real-time task
edge while leveraging the cloud's powerful capabilities.
execution.
This paper proposes a novel edge-cloud collaborative
framework designed specifically for real-time IIoT systems.
Our framework aims to address the key challenges in this  Considerations for prioritized and Pareto-efficient
domain, including: task scheduling in the collaborative environment to
meet QoS requirements.
 Meeting stringent real-time requirements for task
  Mechanisms for maintaining data consistency and
enabling seamless collaboration between the edge
and cloud layers.
C. Resource Management and Task Scheduling in Edge-
II. RELATED WORK Cloud Systems
The convergence of edge computing, cloud computing, and
Efficient resource management and task scheduling are
IoT has attracted significant research attention. Several studies
crucial in edge-cloud environments, especially for real-time
have explored the benefits and architectures of edge-cloud
applications. Su et al. proposed a prediction-based resource
collaboration for various applications.
deployment algorithm (RDAP) and a task scheduling
algorithm with Pareto improvement (TSAP) for edge-cloud
collaborative computing in IoT. Their RDAP algorithm uses
A. Edge Computing in IoT task prediction to optimize resource deployment at edge
servers, while TSAP employs Pareto improvement to balance
Edge computing addresses the limitations of cloud-only user QoS and system service effect. Ammavasai introduced a
architectures in IoT by processing data closer to the source. Dynamic Task Scheduling with Virtual Machine allocation
Premsankar et al. highlighted key IoT application scenarios (DTS-VM) strategy incorporating edge-cloud computing for
benefiting from edge computing, emphasizing the need for IoT, using RNN for task classification and SHO algorithm for
low latency and improved quality of experience. Rong et al. edge node selection. Gong et al. proposed IMAP-GCG, a
presented Sophon Edge, an industrial edge-cloud collaborative method for dynamic container resource allocation in edge
computing platform for efficient AIoT application networks using improved MAPPO and GCN-GRU for
development, focusing on reducing communication latency. workload prediction and resource scheduling.
Zhou et al. proposed energy-efficient smart routing in wireless
edge computing for IoT, highlighting the importance of D. Real-Time Considerations in IoT
efficient resource utilization at the edge.
Several works have focused on addressing the real-time
B. Cloud-Edge Collaboration requirements of IoT applications. Ghosh et al. proposed Mobi-
IoST, a mobility-aware cloud-fog-edge collaborative
The synergistic combination of edge and cloud computing framework for time-critical applications, utilizing mobility
has been explored to leverage their respective advantages. prediction to enhance QoS. Li et al. researched a cloud-edge-
Sharma and Wang proposed a framework for coordinated end collaborative computing offloading strategy in the Internet
processing between edge and cloud for live data analytics in of Vehicles based on the M-TSA algorithm, aiming to reduce
wireless IoT networks. Chen et al. presented a framework for offloading delay. Wang et al. presented ETS-TEE, an energy-
context-aware computation offloading in mobile cloud efficient task scheduling strategy in a mobile trusted
computing, which can be adapted for IIoT scenarios. Xu et al. computing environment, considering task complexity for
discussed a blockchain-enabled deduplicatable data auditing offloading decisions.
mechanism for network storage services in a cloud
environment, highlighting the importance of data integrity in
collaborative systems.

Fig. 2. Key focus areas for optimizing real-time industrial IoT

systems, highlighting edge computing, resource
management, cloud-edge collaboration, and real-time
considerations.
While these studies provide valuable insights into edge  Task Resource Deployment: Receiving and
computing, cloud collaboration, and real-time systems, our managing task-specific resources (software, models)
work aims to provide a holistic framework specifically pushed from the cloud based on predicted needs.
designed for the unique demands of real-time Industrial IoT
systems, integrating prediction-based resource management
and Pareto-efficient scheduling within a cohesive architectural
model.

III. PROPOSED EDGE-CLOUD COLLABORATIVE FRAMEWORK

The proposed edge-cloud collaborative framework for real-
time IIoT systems comprises three primary layers: the
Industrial IoT Device Layer, the Edge Computing Layer, and
the Cloud Computing Layer
A. Industrial IoT Device Layer
This layer consists of diverse IIoT devices, such as sensors,
actuators, PLCs, and other industrial equipment. These
devices generate various types of data with different sampling
rates and real-time constraints. The devices are responsible for
data acquisition and may perform minimal pre-processing
before transmitting data to the edge layer.

B. Edge Computing Layer

The edge computing layer is situated in close proximity to

the IIoT devices, typically within the industrial facility or
nearby network. It consists of edge servers with limited Fig. 3. Hierarchical architecture of Industrial IoT with
computational, storage, and network resources compared to edge and cloud computing integration for efficient data
the cloud. The key functionalities of the edge layer include: processing and analytics.

C. Cloud Computing Layer

The cloud computing layer provides virtually

unlimited computational power, storage capacity, and
advanced
 Real-time Data Processing: Performing immediate secure communication with the cloud.
analysis on incoming data to identify critical events,
anomalies, or trends that require rapid responses.

 Low-Latency Task Execution: Executing time-

sensitive tasks, such as control commands and
immediate alerts, to meet stringent real-time
deadlines.

 Resource Management: Managing local

computational and storage resources efficiently to
support the deployed applications.

 Task Scheduling: Prioritizing and scheduling tasks

based on their real-time requirements and criticality.

 Data Filtering and Aggregation: Reducing the

volume of data transmitted to the cloud by filtering
out irrelevant information and aggregating data.

 Local Data Storage: Temporarily storing processed

or raw data before forwarding it to the cloud.

 Communication Interface: Providing connectivity

and protocol conversion for diverse IIoT devices and
analytics capabilities. It performs the following key functions:

 Long-Term Data Storage: Archiving large

volumes of historical data for comprehensive
analysis and trend identification.

 Complex Data Analytics: Performing

sophisticated analysis, such as machine learning
model training and in-depth pattern recognition,
on the aggregated data from the edge.

 Global System Monitoring and

Management: Providing a centralized view
of the entire IIoT system, including device
status, edge resource utilization, and
application performance.

 Task Prediction and Resource Deployment:

Predicting future task demands at the edge based
on historical data and trends, and proactively
deploying necessary resources (e.g., software, AI
models) to the edge servers.

 Model Management and Updating: Storing,

versioning, and deploying updated machine
learning models to the edge devices for improved
inference accuracy.
 Policy Management and Orchestration: Defining
and enforcing global policies for resource
allocation,
 security, and data management across the entire
framework.

D. Data Flow and Collaboration

IIoT devices generate data that are initially processed at the edge
layer for immediate insights and actions. Processed and/or
raw data are then selectively transmitted to the cloud for long-
term storage and complex analytics. The cloud, in turn,
provides valuable insights, updated models, and resource
management directives back to the edge layer, enabling
continuous improvement and adaptation of edge applications.
This bi-directional data flow and control loop are crucial for
the effective operation of real-time IIoT systems.

Fig. 4. Edge Cloud collaborative framework for IIoT

IV. REAL-TIME TASK SCHEDULING AND
RESOURCE MANAGEMENT
To meet the stringent real-time requirements of IIoT applications, C. Collaborative Task Offloading
efficient task scheduling and resource management strategies In situations where an edge server lacks sufficient resources
are essential within the proposed edge- cloud collaborative to execute a task within its real-time constraints, task
framework. offloading to other edge servers or the cloud becomes
necessary. The decision to offload should consider factors
A. Real-Time Task Scheduling at the Edge such as network latency, available resources at potential
offloading destinations, and the criticality of the task. For less
At the edge layer, tasks arriving from IIoT devices or initiated
time-sensitive or computationally intensive tasks, offloading
locally need to be scheduled based on their real-time
to the cloud might be appropriate, leveraging its abundant
constraints (e.g., deadlines, priorities) and available resources.
resources. For more time-critical tasks that cannot be handled
We propose a priority-based scheduling mechanism where
locally, offloading to a nearby, less loaded edge server with
critical, time-sensitive tasks are assigned higher priority and
the required resources would be preferable. Mechanisms for
are executed preemptively if necessary. Furthermore,
dynamic resource monitoring and workload balancing across
considering the dual objectives of user QoS and system
the edge and cloud are crucial for effective collaborative task
efficiency, a Pareto-based approach, similar to TSAP, can be
offloading.
employed to find a balance between minimizing task
completion time and optimizing resource utilization. This
involves generating multiple scheduling schemes and selecting D. Data Synchronization and Consistency
those that offer Pareto improvements for both objectives. Maintaining data consistency between the edge and cloud is
essential for accurate global analytics and reliable system
B. Prediction-Based Resource Deployment operation. Strategies for data synchronization should consider
the trade-off between consistency and latency, especially for
Given the limited resources at the edge, proactive resource
real-time data. For critical data requiring strong consistency,
deployment based on task prediction can significantly improve
immediate synchronization mechanisms can be employed. For
the hit rate of required resources and reduce task execution
less critical data, periodic or event-triggered synchronization
delays. The cloud layer can analyze historical task patterns
might be sufficient to minimize network overhead. Techniques
and environmental conditions to predict future resource
such as data versioning and conflict resolution can be used to
demands at each edge server. Techniques such as two-
manage inconsistencies that may arise due to network delays
dimensional time series prediction, as used in RDAP, can be
or temporary disconnections.
employed to forecast the type and volume of tasks likely to be
executed at the edge. Based on these predictions, the cloud can
proactively push necessary task resources (e.g., software
components, pre-trained models) to the appropriate edge V. CONCLUSION
servers, ensuring their availability when needed. This reduces This paper presented an edge-cloud collaborative framework
the need for on-demand resource retrieval, which can designed to address the unique challenges of real-time
introduce significant delays, especially for real-time tasks. Industrial IoT systems. The proposed architecture leverages
the strengths of both edge and cloud computing to enable low-
latency processing, efficient resource utilization, and
intelligent decision-making. We discussed key aspects of the
framework, including real-time task scheduling strategies at
the edge, prediction-based resource deployment from the
cloud, collaborative task offloading mechanisms, and
considerations for data synchronization. Future research industrial IoT and production monitoring applications of industrial cyber-
directions include: physical systems. IEEE Transactions on Industrial Informatics, 18(1),
498-508.
[4] Sharma, S. K., & Wang, X. (2017). Live data analytics with collaborative
 Developing more sophisticated real-time task edge and cloud processing in wireless IoT networks. IEEE Access, 5,
scheduling algorithms specifically tailored for 4621-4635.
heterogeneous edge-cloud environments, considering [5] Ghosh, S., Mukherjee, A., Ghosh, S. K., & Buyya, R. (2019). Mobi-iost:
mobility-aware cloud-fog-edge-iot collaborative framework for time-
task dependencies and resource constraints. critical applications. IEEE Transactions on Network Science and
Engineering, 7(4), 2271-2285.
 Investigating advanced machine learning techniques for [6] Okwuibe, J., Haavisto, J., Kovacevic, I., Harjula, E., Ahmad, I., Islam, J.,
& Ylianttila, M. (2021). SDN-enabled resource orchestration for
more accurate and adaptive task prediction to optimize industrial IoT in collaborative edge-cloud networks. IEEE Access, 9,
resource deployment at the edge. 115839-115854.

 Designing efficient and reliable collaborative task

offloading strategies that minimize latency and energy
consumption while maximizing resource utilization.

 Exploring lightweight and robust data synchronization

mechanisms that can ensure data consistency between
the edge and cloud in the presence of network
variability and intermittent connectivity.

 Implementing and evaluating the proposed framework

in real-world industrial IoT testbeds to assess its
performance and feasibility.

Fig. 5. Task scheduling, resource management, and

anomaly detection approaches for optimizing real-time
industrial IoT operations

By addressing these challenges and pursuing these future

research directions, the vision of truly intelligent and
responsive real-time Industrial IoT systems enabled by edge-
cloud collaboration can be realized.

REFERENCES
[1] Rong, G., Xu, Y., Tong, X., & Fan, H. (2021). An edge-cloud
collaborative computing platform for building AIoT applications
efficiently. Journal of Cloud computing, 10(1), 36.
[2] Su, M., Wang, G., & Choo, K. K. R. (2022). Prediction‐Based Resource
Deployment and Task Scheduling in Edge‐Cloud Collaborative
Computing. Wireless Communications and Mobile Computing, 2022(1),
2568503.
[3] Hästbacka, D., Halme, J., Barna, L., Hoikka, H., Pettinen, H., Larrañaga,
M., ... & Elo, M. (2021). Dynamic edge and cloud service integration for