Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

QoS-Aware Load Balancing in the Computing Continuum via Multi-Player Bandits

Ivan Čilić, Ivana Podnar Žarko, Pantelis Frangoudis, Schahram Dustdar

TL;DR

The paper tackles per-client QoS guarantees in the dynamic Computing Continuum by formulating decentralized load balancing as a Multi-Player Multi-Armed Bandit problem with heterogeneous rewards. It introduces QEdgeProxy, which maintains a KDE-based QoS pool per load balancer and uses adaptive epsilon exploration with Smooth Weighted Round Robin routing to balance exploitation and exploration without central coordination. The approach yields sublinear regret in non-stationary conditions and demonstrates superior per-client QoS satisfaction (95–100%), fair load distribution, and resilience to workload and instance changes in a Kubernetes-native implementation. The results indicate that decentralized MP-MAB-based load balancing with KDE QoS estimation is effective and scalable for edge-to-cloud environments where global coordination is costly or impractical.

Abstract

As computation shifts from the cloud to the edge to reduce processing latency and network traffic, the resulting Computing Continuum (CC) creates a dynamic environment where meeting strict Quality of Service (QoS) requirements and avoiding service instance overload becomes challenging. Existing methods often prioritize global metrics and overlook per-client QoS, which is crucial for latency-sensitive and reliability-critical applications. We propose QEdgeProxy, a decentralized QoS-aware load balancer that acts as a proxy between IoT devices and service instances in the CC. We formulate the load balancing problem as a Multi-Player Multi-Armed Bandit (MP-MAB) with heterogeneous rewards: Each load balancer autonomously selects service instances to maximize the probability of meeting its clients' QoS requirements by using Kernel Density Estimation (KDE) to estimate QoS success probabilities. Our load-balancing algorithm also incorporates an adaptive exploration mechanism to recover rapidly from performance shifts and non-stationary conditions. We present a Kubernetes-native QEdgeProxy implementation and evaluate it on an emulated CC testbed deployed on a K3s cluster with realistic network conditions and a latency-sensitive edge-AI workload. Results show that QEdgeProxy significantly outperforms proximity-based and reinforcement-learning baselines in per-client QoS satisfaction, while adapting effectively to load surges and changes in instance availability.

QoS-Aware Load Balancing in the Computing Continuum via Multi-Player Bandits

TL;DR

The paper tackles per-client QoS guarantees in the dynamic Computing Continuum by formulating decentralized load balancing as a Multi-Player Multi-Armed Bandit problem with heterogeneous rewards. It introduces QEdgeProxy, which maintains a KDE-based QoS pool per load balancer and uses adaptive epsilon exploration with Smooth Weighted Round Robin routing to balance exploitation and exploration without central coordination. The approach yields sublinear regret in non-stationary conditions and demonstrates superior per-client QoS satisfaction (95–100%), fair load distribution, and resilience to workload and instance changes in a Kubernetes-native implementation. The results indicate that decentralized MP-MAB-based load balancing with KDE QoS estimation is effective and scalable for edge-to-cloud environments where global coordination is costly or impractical.

Abstract

As computation shifts from the cloud to the edge to reduce processing latency and network traffic, the resulting Computing Continuum (CC) creates a dynamic environment where meeting strict Quality of Service (QoS) requirements and avoiding service instance overload becomes challenging. Existing methods often prioritize global metrics and overlook per-client QoS, which is crucial for latency-sensitive and reliability-critical applications. We propose QEdgeProxy, a decentralized QoS-aware load balancer that acts as a proxy between IoT devices and service instances in the CC. We formulate the load balancing problem as a Multi-Player Multi-Armed Bandit (MP-MAB) with heterogeneous rewards: Each load balancer autonomously selects service instances to maximize the probability of meeting its clients' QoS requirements by using Kernel Density Estimation (KDE) to estimate QoS success probabilities. Our load-balancing algorithm also incorporates an adaptive exploration mechanism to recover rapidly from performance shifts and non-stationary conditions. We present a Kubernetes-native QEdgeProxy implementation and evaluate it on an emulated CC testbed deployed on a K3s cluster with realistic network conditions and a latency-sensitive edge-AI workload. Results show that QEdgeProxy significantly outperforms proximity-based and reinforcement-learning baselines in per-client QoS satisfaction, while adapting effectively to load surges and changes in instance availability.

Paper Structure

This paper contains 42 sections, 2 theorems, 11 equations, 11 figures, 2 tables, 4 algorithms.

Table of Contents

  1. Introduction
  2. Related work
  3. System Model
  4. MP-MAB Problem Formulation
  5. Assumptions
  6. Objective
  7. Key Properties
  8. Regret Definition
  9. Decentralized Algorithms for QoS-Aware Load Balancing in CC
  10. QoS Pool Maintenance
  11. Request Routing
  12. Handling Instance Additions
  13. Handling Instance Removals
  14. Regret Upper Bound
  15. Computational Complexity and Communication Overhead
  16. ...and 27 more sections

Key Result

Theorem 1

Given the bounded variation from Definition def:variation, the dynamic regret of load balancer $k$ satisfies matching the minimax-optimal upper bound for stochastic multi-armed bandits with non-stationary rewards established by besbes2014.

Figures (11)

  • Figure 1: Decentralized load balancing.
  • Figure 2: QEdgeProxy within a Kubernetes environment.
  • Figure 3: Client QoS satisfaction rate aggregated across five scenarios.
  • Figure 4: Load balancing fairness across five scenarios.
  • Figure 5: Distribution of client QoS success rate $\hat{\rho}$ (scenario #1).
  • ...and 6 more figures

Theorems & Definitions (5)

  • Definition 1: Variation budget per load balancer
  • Theorem 1: Regret bound per LB
  • proof
  • Corollary 1: System-wide regret
  • Remark