Background and Motivation
Edge computing has emerged as a critical paradigm for reducing latency and improving performance by bringing computation closer to data sources. This rapid adoption has created a demand for efficient orchestration frameworks capable of managing containerized applications across resource-constrained, geographically distributed nodes. At the same time, a fundamental shift is underway in how software systems are designed and operated. The rise of agentic AI — systems where autonomous agents can perceive, reason, plan, and act with minimal human intervention — is reshaping how distributed infrastructure is managed. Rather than relying on centralized, rule-based controllers, the emerging paradigm envisions networks of specialized agents that coordinate, negotiate, and make decisions collaboratively. This trend is being fueled by multi-agent systems, lightweight foundation models deployable at the edge, and new inter-agent communication protocols such as Google's Agent2Agent (A2A) protocol and Anthropic's Model Context Protocol (MCP), which aim to standardize how agents discover each other, exchange context, and delegate tasks across system boundaries. For edge orchestration, this agentic shift has significant implications. Traditional orchestration platforms like Kubernetes, while powerful, rely on centralized control planes and static scheduling strategies that often prove heavyweight for edge environments characterized by limited resources, intermittent connectivity, and heterogeneous hardware. Lightweight distributions like K3s and MicroK8s have partially addressed resource constraints, and specialized frameworks like Oakestra have introduced hierarchical orchestration tailored to edge deployments. However, none of these frameworks were designed with agentic communication in mind. The question of how autonomous agents should communicate, coordinate, and collectively manage distributed edge infrastructure remains largely unexplored. Current agent communication protocols were primarily designed for cloud-based multi-agent systems and do not account for the unique constraints of edge computing: limited bandwidth, variable connectivity, resource-scarce nodes, and the need for real-time responsiveness. Understanding how these protocols behave in constrained edge environments, what architectural bottlenecks they introduce, and how orchestration frameworks can be extended to support agentic coordination is an open and timely research challenge. This thesis aims to investigate the intersection of edge orchestration and agentic communication. The project will evaluate how existing orchestration frameworks perform in constrained edge settings, assess the suitability of emerging agent communication protocols for edge environments, and explore how agentic coordination patterns can enhance or replace traditional orchestration approaches.
Expected Outcomes
The student will work to:
- Conduct a comprehensive literature review covering edge orchestration frameworks, multi-agent systems, and agent communication protocols (A2A, MCP, ACP), identifying the gap between current protocol designs and edge computing requirements;
- Establish a controlled edge computing testbed that simulates realistic resource constraints (limited CPU, memory, network bandwidth) and deploy representative orchestration frameworks (Kubernetes, K3s, MicroK8s, Oakestra) for baseline performance evaluation;
- Analyze the architectural components of each framework (control plane overhead, container runtime, service discovery, inter-node communication) and quantify their resource footprint under varying operational stresses;
- Evaluate the feasibility of deploying agent communication protocols in edge environments, measuring their overhead in terms of latency, bandwidth consumption, and computational cost on resource-constrained nodes;
- Design and prototype an agentic coordination mechanism for one aspect of edge orchestration (e.g., distributed scheduling, adaptive scaling, fault recovery, or resource-aware agent discovery) and integrate it with an existing orchestration platform;
- Evaluate the proposed solution through experimentation, measuring improvements in responsiveness, scalability, fault tolerance, or resource efficiency compared to traditional centralized orchestration.
Requirements
- Strong understanding of container orchestration concepts;
- Proficiency in Linux system administration and containerization technologies (Docker, containerd);
- Programming skills in Python or Go for developing measurement tools and automation scripts;
- Knowledge of distributed systems concepts and network protocols;
- Familiarity with cloud-native technologies and microservices architectures;
- Interest in AI agents, multi-agent systems, or autonomous systems is a strong plus;
- Experience with infrastructure automation tools (Ansible, Terraform) for testbed setup is a plus.
Related Work
[1] Giovanni Bartolomeo, Mehdi Yosofie, Simon Bäurle, Oliver Haluszczynski, Nitinder Mohan, and Jörg Ott. 2023. Oakestra: A Lightweight Hierarchical Orchestration Framework for Edge Computing. In 2023 USENIX Annual Technical Conference (USENIX ATC 23). USENIX Association, Boston, MA, 135–148. https://www.usenix.org/conference/atc23/presentation/bartolomeo
