Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Agentic Fog: A Policy-driven Framework for Distributed Intelligence in Fog Computing

Saeed Akbar, Muhammad Waqas, Rahmat Ullah

TL;DR

The paper tackles the challenge of coordinating fog/fog-edge resources under partial observability and strict latency by proposing Agentic Fog (AF), a non-LLM, policy-driven multi-agent framework. Fog Nodes operate as autonomous agents with shared memory and peer-to-peer coordination, and system-wide goals are decomposed by a Global Orchestrator into abstract sub-goals, yielding a formal exact potential game that guarantees convergence under asynchronous bounded-rational updates and resilience to node failures. The work provides a rigorous model, layered architecture, and algorithms, along with simulations showing substantial latency reductions, fast convergence, and graceful degradation under failures relative to ILP and greedy baselines. The practical impact is a scalable, analyzable, and robust framework for distributed intelligence in latency-sensitive fog and edge environments that avoids the computational and interpretability challenges of LLM-based agentic systems. The findings clarify the distinction between LLM-driven AAI and principled non-LLM agentic architectures, demonstrating that infrastructure systems can achieve emergent intelligent behavior with formal guarantees using shared memory and localized coordination.

Abstract

Fog and edge computing require adaptive control schemes that can handle partial observability, severe latency requirements, and dynamically changing workloads. Recent research on Agentic AI (AAI) increasingly integrates reasoning systems powered by Large Language Models; however, these tools are not applicable to infrastructure-level systems due to their high computational cost, stochastic nature, and poor formal analyzability. In this paper, a generic model, Agentic Fog (AF), is presented, in which fog nodes are represented as policy-driven autonomous agents that communicate via p2p interactions based on shared memory and localized coordination. The suggested architecture decomposes a system's goals into abstract policy guidance and formalizes decentralized fog coordination as an exact potential game. The framework is guaranteed to converge and remain stable under asynchronous updates, bounded-rational best-response dynamics, and node failures. Simulations demonstrate that the AF system achieves lower average latency and adapts more efficiently to varying demand than greedy heuristics and integer linear programming under dynamic conditions. The sensitivity analysis also demonstrates the capability to perform optimally under different memory and coordination conditions.

Agentic Fog: A Policy-driven Framework for Distributed Intelligence in Fog Computing

TL;DR

The paper tackles the challenge of coordinating fog/fog-edge resources under partial observability and strict latency by proposing Agentic Fog (AF), a non-LLM, policy-driven multi-agent framework. Fog Nodes operate as autonomous agents with shared memory and peer-to-peer coordination, and system-wide goals are decomposed by a Global Orchestrator into abstract sub-goals, yielding a formal exact potential game that guarantees convergence under asynchronous bounded-rational updates and resilience to node failures. The work provides a rigorous model, layered architecture, and algorithms, along with simulations showing substantial latency reductions, fast convergence, and graceful degradation under failures relative to ILP and greedy baselines. The practical impact is a scalable, analyzable, and robust framework for distributed intelligence in latency-sensitive fog and edge environments that avoids the computational and interpretability challenges of LLM-based agentic systems. The findings clarify the distinction between LLM-driven AAI and principled non-LLM agentic architectures, demonstrating that infrastructure systems can achieve emergent intelligent behavior with formal guarantees using shared memory and localized coordination.

Abstract

Fog and edge computing require adaptive control schemes that can handle partial observability, severe latency requirements, and dynamically changing workloads. Recent research on Agentic AI (AAI) increasingly integrates reasoning systems powered by Large Language Models; however, these tools are not applicable to infrastructure-level systems due to their high computational cost, stochastic nature, and poor formal analyzability. In this paper, a generic model, Agentic Fog (AF), is presented, in which fog nodes are represented as policy-driven autonomous agents that communicate via p2p interactions based on shared memory and localized coordination. The suggested architecture decomposes a system's goals into abstract policy guidance and formalizes decentralized fog coordination as an exact potential game. The framework is guaranteed to converge and remain stable under asynchronous updates, bounded-rational best-response dynamics, and node failures. Simulations demonstrate that the AF system achieves lower average latency and adapts more efficiently to varying demand than greedy heuristics and integer linear programming under dynamic conditions. The sensitivity analysis also demonstrates the capability to perform optimally under different memory and coordination conditions.
Paper Structure (19 sections, 5 theorems, 9 equations, 6 figures, 1 table, 4 algorithms)

This paper contains 19 sections, 5 theorems, 9 equations, 6 figures, 1 table, 4 algorithms.

Table of Contents

  1. Introduction
  2. Contributions and Novelty of the Proposed Work
  3. Comparison with Existing Fog Control Paradigms
  4. Agentic AI Without Large Language Models
  5. System Model
  6. Separation of System-Level Cognition and Local Autonomy
  7. Shared Memory
  8. Fog Agents
  9. Peer-to-Peer Coordination for Emergent Intelligence
  10. Decoupling Decision-Making from Execution
  11. Layered Structure for Scalability and Robustness
  12. Design Implications
  13. Simulation Results and Discussion
  14. Latency and Adaptivity
  15. Convergence Behavior
  16. ...and 4 more sections

Key Result

Proposition 1

The global objective $\mathcal{O}^{*}$ can be dynamically decomposed into sub-objectives $\{o_1,\dots,o_k\}$ such that each sub-goal corresponds to a locally creditable marginal contribution and is solvable under partial observability.

Figures (6)

  • Figure 1: Layered architecture of the proposed Agentic Fog system comprising the Intelligence Layer, the Agent Layer, and the Execution Layer. The Intelligence layer has a Global Orchestrator Agent for policy control and goal decomposition, and a Shared Memory for saving previous system states, the Agent Layer consists of distributed Fog Agents managing caching, routing, and load balancing, and finally, the Execution Layer comprises of Execution Agents responsible for localized task execution.
  • Figure 2: Comparing the latency under varying workload conditions. The proposed AF records lower average latency as compared to the baseline strategies over time.
  • Figure 3: Comparing the convergence behavior as a function of network size. The proposed AF system results in faster convergence with increasing number of fog nodes, compared with the ILP and Greedy baselines.
  • Figure 4: Performance degradation under node failures. The proposed Agentic Fog framework exhibits greater resilience to random fog node failures compared to ILP and Greedy baselines, with minimal increase in latency.
  • Figure 5: Coordination overhead relative to the number of fog nodes (FNs). The AF system incurs slightly more message exchange than the ILP and Greedy baselines.
  • ...and 1 more figures

Theorems & Definitions (12)

  • Definition 1: Agentic Fog System
  • Proposition 1: Dynamic Goal Decomposition
  • proof
  • Lemma 1: Local Rationality
  • proof
  • Definition 2: AF Game
  • Theorem 1: Existence of Potential Function
  • proof
  • Theorem 2: Convergence
  • proof
  • ...and 2 more