National Technical University of Athens (NTUA)
Orchestration of distributed applications across the computing continuum (IoT-Edge-Cloud)
Abstract
dc:descriptionThe proliferation of Internet of Things devices and the increasing prevalence of edge systems have led to the emergence of the computing continuum, where applications are executed across heterogeneous resources ranging from cloud data centers to edge servers and embedded devices. The components of complex applications can be deployed, managed, and scaled across geographically distributed resources. Efficient orchestration of these applications is essential to ensure application performance and optimize resource utilization. Therefore, we can identify specific challenges in the domain of computing continuum orchestration. In distributed settings, each service imposes a different set of requirements. Reconciling conflicting objectives to achieve multi-cluster placement remains a significant challenge. Furthermore, the automatic scaling of resources allocated to each service needs to be dynamically adjusted to respond to incoming input load fluctuations. Static solutions and optimization-based algorithms offer ease of use but are limited to coarse-grained control of resources. Lastly, integrating Internet of Things devices requires abstractions that simplify interaction with physical resources. To address these challenges, this thesis proposes a unified orchestration framework that leverages resources across the computing continuum. The framework introduces a hierarchical orchestration architecture centered around the Synergetic Meta-Orchestrator that coordinates resource management across clusters while maintaining a global view of application deployments and infrastructure state. The architecture includes cluster-level orchestrators that manage the lifecycle, monitoring, and resource allocation of applications. This hierarchy enables centralized orchestration in combination with localized control mechanisms. The proposed framework constructs deployment plans that model applications as graphs of interconnected services. This approach allows the orchestration system to account for service requirements, dependencies, and communication costs when determining deployment configurations across heterogeneous clusters. The goal of the framework is to manage the trade-off between application performance and resource utilization. In addition to placement optimization problems, the architecture supports learning-based algorithms to address the problem of dynamic resource scaling. Distributed applications frequently experience workload fluctuations that require runtime adaptation of resource allocations. To address this issue, this thesis investigates the use of a reinforcement learning agent to enable adaptive autoscaling policies. The agent continuously observes the environment's state and is trained to manage the resources of a service chain composed of serverless functions. By interacting with the environment, the agent can learn autoscaling policies that can dynamically balance performance with resource usage. Building upon the single-agent reinforcement learning setting, the thesis explores cooperative scaling strategies for applications composed of multiple services. A multi-agent reinforcement learning approach is investigated in which individual agents manage the scaling behavior of separate application components while coordinating their decisions to optimize overall system performance. Another important aspect of orchestration across the continuum is the integration of heterogeneous Internet of Things devices in distributed applications. This thesis introduces the concept of the Virtual Object, which serves as the digital counterpart and an abstraction of the physical devices. This way, Internet of Things devices provide a standardized interface that enables applications to interact with heterogeneous devices in a consistent manner. The proposed mechanisms were implemented and evaluated using a prototype orchestration platform based on container orchestration technologies and multi-cluster deployments. Experimental evaluation using distributed application workloads and serverless service chains demonstrates the feasibility of the proposed framework and highlights the potential of combining hierarchical orchestration architectures with learning-based resource management techniques. Overall, the results indicate that the proposed approach provides a flexible foundation for managing distributed applications across cloud, edge, and Internet of Things infrastructures within the computing continuum.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Filinis, Nikolaos
- Φιλίνης, Νικόλαος
Subjects
dc:subject × 20- Ενορχήστρωση
- Υπολογιστικό Συνεχές
- Διαδίκτυο των πραγμάτων
- Ενισχυτική μάθηση
- Ενισχυτική Μάθηση Πολλών Πρακτόρων
- Αυτόματη Κλιμάκωση Πόρων
- Κατανεμημένη Τοποθέτηση Εφαρμογών
- Orchestration
- Computing Continuum
- Internet of things (IoT)
- Reinforcement learning
- Multi-Agent Reinforcement Learning
- Resource Autoscaling
- Distributed Application Placement
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Επιστήμη ηλεκτρονικών υπολογιστών
- Computer and Information Sciences
- Natural Sciences
- Computer Science
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/62115