Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Bowen Sun, Christos D. Antonopoulos, Evgenia Smirni, Bin Ren, Nikolaos Bellas, Spyros Lalis

TL;DR

LACE-RL is presented, a latency-aware and carbon-efficient management framework that formulates serverless pod retention as a sequential decision problem, achieving better latency-carbon trade-offs than state-of-the-art heuristic and single-objective baselines, approaching Oracle performance.

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.

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

TL;DR

LACE-RL is presented, a latency-aware and carbon-efficient management framework that formulates serverless pod retention as a sequential decision problem, achieving better latency-carbon trade-offs than state-of-the-art heuristic and single-objective baselines, approaching Oracle performance.

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.
Paper Structure (24 sections, 6 equations, 10 figures, 3 tables)

This paper contains 24 sections, 6 equations, 10 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Energy Modeling and Challenges
  3. Huawei Cloud Dataset
  4. Energy Model: Accounting for Serverless Carbon Footprint
  5. Challenges in Serverless Management
  6. System Design
  7. LACE-RL Overview
  8. Carbon-Latency Tradeoff via Keep-Alive Tuning
  9. Reinforcement-Learning Model Formulation
  10. Evaluation
  11. Experimental Methodology
  12. Model and Parameters Validation via Real-System Measurements
  13. Dataset and Trace Processing
  14. Energy and Carbon Modeling
  15. RL Model Configuration
  16. ...and 9 more sections

Figures (10)

  • Figure 1: Characterization of the Huawei Cloud Trace joosen2025serverless.
  • Figure 2: Impact of keep-alive timeout for two different but representative functions in the Huawei dataset. In both cases, longer timeouts reduce cold starts but increase idle carbon footprint. Depending on the function, idle carbon may even surpass (significantly) the carbon for execution (right plot).
  • Figure 3: Observed variability in carbon intensity of the electrical grid electricitymaps(a) and Huawei's function memory footprint (b).
  • Figure 4: LACE-RL overview.
  • Figure 5: General testing workload results across key metrics. The dashed line shows the optimal value.
  • ...and 5 more figures