ENERGY- AND LATENCY-AWARE OPTIMIZATION OF SERVERLESS PLATFORMS USING REINFORCEMENT LEARNING

Abstract

Serverless computing has emerged as a promising cloud paradigm that enables on-demand execution of functions without requiring infrastructure management. Despite its scalability and cost-efficiency, serverless platforms often face challenges related to high latency, cold-start delays, and inefficient energy utilization, particularly under dynamic workloads. This paper proposes an Energy- and Latency-Aware Optimization Framework for serverless environments using Reinforcement Learning (RL). The proposed model formulates resource allocation and function scheduling as a sequential decision-making problem, where an intelligent RL agent dynamically adjusts resource provisioning, scaling policies, and workload placement to minimize response time and energy consumption simultaneously. A multi-objective reward function is designed to balance trade-offs between latency reduction and energy efficiency. Experimental evaluation under varying workload intensities demonstrates that the proposed approach significantly reduces average response time and power consumption compared to traditional auto-scaling mechanisms. The framework enhances operational efficiency while maintaining service-level agreements (SLAs), making it suitable for large-scale, energy-conscious cloud deployments