Research Statement · Energy-Aware Cloud Computing

Convincing Kubernetes to Care About Energy

Menra Wedwang Romial
IMT ATLANTIQUE · INRIA · STACK TEAM · NANTES, FR
Abstract

We investigate why running a container has to cost so much electricity — and demonstrate empirically that it doesn’t have to. This thesis develops energy‑aware scheduling, power capping, and SLA management for cloud clusters, treating energy as a first‑class constraint rather than an afterthought. Experiments on Grid’5000 with Intel RAPL show that clusters can do more with less — less heat, less waste, less electricity bill — without sacrificing performance. Somewhere between a PhD thesis and a power meter is where the author resides.

Keywords — Kubernetes · Energy Proportionality · RAPL · Grid’5000 · SLA · Green Computing
PORTRAIT · GREEN-IT.WEBP
Fig. 1. The author, mid‑argument with a scheduler. Contains ≈ 60 % coffee by volume.

On the ethics of idle wattage

My job is to convince Kubernetes to care about energy. It doesn’t listen. I insist. We compromise. Eventually, something moves.

The question I wake up with every morning: why does running a container have to cost so much electricity? Spoiler: it doesn’t have to. That is the whole point of my PhD at Inria, in the STACK team.

I work on energy‑aware scheduling, power capping, and SLA management in cloud clusters. Concretely: I poke servers with Intel RAPL, run large‑scale experiments on Grid’5000, and write schedulers that treat energy as a first‑class constraint, not an afterthought.

The result: clusters that do more with less — less heat, less waste, less electricity bill. Not by sacrificing performance. By being smarter about how resources are used.

If you want to talk cloud, green computing, or Kubernetes internals, I’m always up for it.


2 · News & Recent Activity
Apr 16, MMXXVI Welcoming Clement Obama as a new M2 research intern (April–September 2026) working on Reactive control of energy constraints in Kubernetes clusters: a multi‑lever re-adaptation policy with SLA management.
Dec 04, MMXXV Volunteering at UCC & BDCAT 2025 — My Experience. Notes on organising a conference when you’re technically still the intern.
Oct 03, MMXXV Invited talk at the Nantes Green Infrastructure Seminar“Power capping without tears: why your SRE team will thank you.”
Jul 22, MMXXV Paper accepted at UCC ’25: Energy-Proportional Scheduling via Adaptive RAPL Budgets. Camera-ready submitted.

From the blog

Oct 23, MMXXVA comprehensive guide to powercap-utils on Linux
Sep 22, MMXXVBuilding a Kubernetes controller with Kubebuilder from scratch
Aug 24, MMXXVGuide to building a custom Kubernetes scheduler
Jun 11, MMXXVNotes on measuring joules with honesty

Résumé of work

  1. M. W. Romial, J. Pastor, A. Lebre. Energy‑proportional scheduling via adaptive RAPL budgets. UCC ’25, Nantes, 2025. [pdf · bib]
  2. M. W. Romial et al. Multi-lever re-adaptation of SLA under a power cap. BDCAT ’25, 2025. [pdf · bib]
  3. M. W. Romial, A. Lebre. Position paper: Kubernetes needs an energy descriptor. HotCarbon, 2024. [pdf · bib]

see full list


5 · Correspondence
Mail
[email protected]

Expect a reply within ≤ 3 working days, shorter if you mention scheduling.

Address
IMT Atlantique
4 Rue Alfred Kastler
44300 Nantes, France
Elsewhere
GitHub
Google Scholar
ORCID 0000-0002-XXXX-XXXX