Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

QR-MO: Q-Routing for Multi-Objective Shortest-Path Computation in 5G-MEC Systems

Annisa Sarah, Rosario G. Garroppo, Gianfranco Nencioni

TL;DR

This work tackles multi-objective routing for MOSP in 5G-MEC by extending Q-routing to QR-MO, a reinforcement-learning approach that maintains multi-attribute Q-values and uses a dominance-based heuristic to produce a Pareto-approximate set of paths. It compares QR-MO against the exact Multi-objective Dijkstra Algorithm (MDA) across four network topologies, showing near-optimal Distance to Pareto Set (DPS) and high correctness after 100 episodes, particularly in sparse graphs with low to moderate degrees. QR-MO generates exactly $K=J$ solutions, one per objective, enabling real-time, context-aware routing in dynamic 5G-MEC environments, albeit with higher per-episode computation than MDA in static conditions. The results suggest QR-MO’s potential for near-real-time Pareto-front approximation and adaptive routing, with future work focusing on dynamic networks and faster policy updates (e.g., ~5 ms per solution after convergence).

Abstract

Multi-access edge computing (MEC) is a promising technology that provides low-latency processing capabilities. To optimize the network performance in a MEC system, an efficient routing path between a user and a MEC host is essential. The network performance is characterized by multiple attributes, including packet-loss probability, latency, and jitter. A user service may require a particular combination of such attributes, complicating the shortest-path computation. This paper introduces Q-Routing for Multi-Objective shortest-path computation (QR-MO), which simultaneously optimizes multiple attributes. We compare the QR-MO's solutions with the optimal solutions provided by the Multi-objective Dijkstra Algorithm (MDA). The result shows the favorable potential of QR-MO. After 100 episodes, QR-MO achieves 100% accuracy in networks with low to moderate average node degrees, regardless of the size, and over 85% accuracy in networks with high average node degrees.

QR-MO: Q-Routing for Multi-Objective Shortest-Path Computation in 5G-MEC Systems

TL;DR

This work tackles multi-objective routing for MOSP in 5G-MEC by extending Q-routing to QR-MO, a reinforcement-learning approach that maintains multi-attribute Q-values and uses a dominance-based heuristic to produce a Pareto-approximate set of paths. It compares QR-MO against the exact Multi-objective Dijkstra Algorithm (MDA) across four network topologies, showing near-optimal Distance to Pareto Set (DPS) and high correctness after 100 episodes, particularly in sparse graphs with low to moderate degrees. QR-MO generates exactly solutions, one per objective, enabling real-time, context-aware routing in dynamic 5G-MEC environments, albeit with higher per-episode computation than MDA in static conditions. The results suggest QR-MO’s potential for near-real-time Pareto-front approximation and adaptive routing, with future work focusing on dynamic networks and faster policy updates (e.g., ~5 ms per solution after convergence).

Abstract

Multi-access edge computing (MEC) is a promising technology that provides low-latency processing capabilities. To optimize the network performance in a MEC system, an efficient routing path between a user and a MEC host is essential. The network performance is characterized by multiple attributes, including packet-loss probability, latency, and jitter. A user service may require a particular combination of such attributes, complicating the shortest-path computation. This paper introduces Q-Routing for Multi-Objective shortest-path computation (QR-MO), which simultaneously optimizes multiple attributes. We compare the QR-MO's solutions with the optimal solutions provided by the Multi-objective Dijkstra Algorithm (MDA). The result shows the favorable potential of QR-MO. After 100 episodes, QR-MO achieves 100% accuracy in networks with low to moderate average node degrees, regardless of the size, and over 85% accuracy in networks with high average node degrees.

Paper Structure

This paper contains 11 sections, 7 equations, 2 figures, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Works
  3. Problem Definition and MDP Representation
  4. Proposed Solution
  5. Q-learning
  6. QR-MO
  7. Complexity Analysis
  8. Experimental Results
  9. Performance Metrics
  10. Simulation Setting and Result Discussion
  11. Conclusion and Future Works

Figures (2)

  • Figure 1: Average DPS of QR-MO of network (a) 25N50E, (b) 100N150E, (c) MCC (30N35E), and (d) 50N50E
  • Figure 2: Comparison of (a) average correctness of QR-MO and (b) the average number of correct solutions of QR-MO for different topologies