Forecast-Driven Energy-Aware Orchestration in Content Delivery Networks

Abstract

Content Delivery Networks (CDNs) and edge platforms are increasingly expected to reduce energy consumption while preserving strict Quality of Service (QoS) and Service Level Agreement (SLA) targets under highly variable demand. In practice, operators often keep excess capacity online to ab-sorb sudden spikes and to hedge against warm-up delays, which leads to persistent energy waste during off-peak periods. This paper presents a forecast driven orchestration framework that couples short-horizon workload prediction with a practical server life-cycle controller managing active, warm-standby, and off pools. The controller converts uncertainty-aware forecasts into risk-calibrated capacity decisions, using hysteresis and warm-up queue dynamics to avoid oscil-lations and to prevent transient under-provisioning. We evaluate the approach in closed-loop simulation using real Points of Presence (PoP) level traces and statis-tically constrained synthetic scenarios generated with a Large Language Model (LLM) to stress-test bursty and heavy-tailed regimes beyond the observed data. Results show that probabilistic, tail-aware provisioning improves reliability under volatile demand while still enabling meaningful energy reductions compared to static provisioning.