Green or Fast? Learning to Balance Cold Starts and Idle Carbon in Serverless Computing

TL;DR

This paper introduces LACE-RL, a deep reinforcement learning framework that dynamically balances cold-start latency and idle carbon emissions in serverless computing, adapting to real-time conditions.

Contribution

LACE-RL is the first to formulate serverless pod retention as a sequential decision problem using deep RL, optimizing both latency and carbon emissions jointly.

Findings

Reduces cold starts by 51.69% compared to static policies.

Cuts idle keep-alive carbon emissions by 77.08%.

Achieves better latency-carbon trade-offs than existing heuristics.

Abstract

Serverless computing simplifies cloud deployment but introduces new challenges in managing service latency and carbon emissions. Reducing cold-start latency requires retaining warm function instances, while minimizing carbon emissions favors reclaiming idle resources. This balance is further complicated by time-varying grid carbon intensity and varying workload patterns, under which static keep-alive policies are inefficient. We present LACE-RL, a latency-aware and carbon-efficient management framework that formulates serverless pod retention as a sequential decision problem. LACE-RL uses deep reinforcement learning to dynamically tune keep-alive durations, jointly modeling cold-start probability, function-specific latency costs, and real-time carbon intensity. Using the Huawei Public Cloud Trace, we show that LACE-RL reduces cold starts by 51.69% and idle keep-alive carbon emissions by 77.08% compared to Huawei's static policy, while achieving better latency-carbon trade-offs than state-of-the-art heuristic and single-objective baselines, approaching Oracle performance.