An Entropy-Based Load Profiling and Custom Metric Tuning Framework for Kubernetes Autoscaling
ID：62
Submission ID：385 View Protection：ATTENDEE
Updated Time：2026-07-22 16:09:33 Hits：17
Online
Start Time：2026-07-31 11:25 （Asia/Kolkata）
Duration：15min
Session：[S6] Artificial Intelligence Use Cases » [S6-4] Artificial Intelligence Use Cases
Video
No Permission
Presentation File
Tips: The file permissions under this presentation are only for participants. You have not logged in yet and cannot view it temporarily.
Abstract
Containerized microservices are increasingly unpredictable due to workload variations. Dynamic orchestration requires intelligent, high fidelity auto-scaling mechanisms to maintain performance and minimize operational costs. However, the current Kubernetes Horizontal Pod Autoscalers (HPAs) suffer from reactive lag, scaling oscillations, and imbalances in node resource utilization due to the reliance on standard resource metrics and unoptimized metric scraping intervals. This research combines maximum entropy based load profiling (for predictive capacity planning) and a custom metric tuning pipeline (to dynamically tune scaling behaviors). The proposed model reduces container overprovisioning and the cluster Total Cost of Ownership (TCO), and maintains Service-Level Objectives (SLOs) for latency. This work finally presents a strong traffic aware infrastructure model that bridges the gap between proactive traffic forecasting and reactive resource provisioning in modern edge and cloud computing environments.
Keywords
Kubernetes HPA, Maximum Entropy, Custom Metrics, Load Profiling, TCO Optimization
Speaker
Comment submit