Enforcing SLA Compliance in Edge-Cloud Offloading: Replacing Simple Thresholds with an Adaptive PI Controller

Abstract

Manually configured rules for offloading computational tasks from edge devices—such as smartphones or IoT gateways—to the cloud are often rigid and ineffective under real-world dynamics. Static policies—e.g., “offload if the local queue exceeds 10 tasks”—frequently degrade performance when cloud connectivity is slow, congested, or unpredictable, leading to repeated violations of Service Level Agreements (SLAs). To address this, we propose an SLA-aware adaptive controller that dynamically adjusts offloading decisions based on real-time performance metrics, including latency, cost, and resource utilization. Our system employs an optimized PI (Proportional-Integral) control strategy, enhanced with intelligent fault tolerance, conservative safety bounds, and a brief learning phase to ensure stability without oscillation. Using a high-fidelity simulator that models real-world uncertainties—such as network jitter, traffic bursts, and transient failures—we evaluated three approaches: a static rule, a basic adaptive algorithm, and our SLA-aware controller. Results show that the proposed system achieves 100% SLA compliance, reduces latency variance by over 60%, and improves cost-performance efficiency—all while maintaining full transparency, configurability, and production-grade stability. Crucially, it eliminates erratic or overreactive decision-making by converging to a stable policy after initial adaptation, making it suitable for mission-critical edge-cloud deployments.