Research Paper · Information Technology

GreenServe: Energy-Aware Scheduling of Large Language Model Inference Workloads in Serverless Edge Computing Environments

An original research paper proposing a carbon-aware, multi-tier scheduling framework for LLM inference. Reported results: 31.4% energy reduction and 42.7% carbon reduction versus a latency-only baseline, with 97.8% SLO attainment.

Download

Word document (.docx)Formatted for submission — headings, tables, headers, page numbers.↓
Markdown source (.md)Plain-text source with LaTeX equations, editable anywhere.↓

Paper structure

Abstract & Keywords
1. Introduction (motivation, problem statement, RQ1–RQ4, contributions)
2. Literature Review (18 peer-reviewed sources)
3. System Model
4. GreenServe Design (cost model, carbon-aware placement, pre-warming)
5. Experimental Methodology
6. Results (energy, carbon, latency, ablations, sensitivity)
7. Discussion (implications, embodied carbon, threats to validity, ethics)
8. Conclusion & Future Work
References, Appendix A (notation), Appendix B (reproducibility)