CNCF Cloud Native Sustainability Month 2026 - Local Meetup Tokyo

1. 5:45 PM - 6:00 PM JST

   Opening Session（オープニング）

   in-person
   SPEAKERS
2. 6:00 PM - 6:05 PM JST

   Introduction to Cloud Native Sustainability in CNCF

   in-person

   An introduction to the Cloud Native Sustainability Working Group under the CNCF Operational Resilience TAG, covering its goals, focus areas, and key initiatives to advance sustainability across the cloud native ecosystem.

3. 6:05 PM - 6:15 PM JST

   ObservabilityにSustainabilityを - SCI for OpenTelemetry

   in-person

   Green Software Foundation (GSF) はソフトウェアの炭素強度を数値化するための ISO 標準 Software Carbon Intensity (SCI) を策定し、IT システムを環境影響の切り口から国際標準で評価できるようになりました。これにより導入検討時の製品比較から運用時のモニタリング・レポーティングまで IT システムの環境負荷を客観的な数字で表すことができるようになりました。 SCI はシステムコンポーネントのエネルギー消費量や機能単位など（トランザクション数や同時接続数など）から算出されますが、近年のモダンなシステムなら計測対象として既にモニタリングシステムに収集されているかもしれません。しかし、それら SCI の算出に使えるモニタリングシステム上のメトリクスについての明確な定義が存在していないことも確かです。 そこで GSF では OpenTelemetry Semantic Conventions SIG に提案することをゴールに、SCI for OpenTelemetry を立ち上げました。Observability 領域で広く使われている OpenTelemetry (OTel) で用いる関連メトリクスや属性などを定義することで、OTel 導入システムにおける SCI を算出しやすくします。

   本セッションでは SCI for OTel が実現する世界と現在の検討状況についてお伝えします。

   SPEAKERS
4. 6:15 PM - 6:25 PM JST

   LF Energy入門 ― LF Energy Summit Europe 2026の最新情報とともに

   in-person

   KeplerやSCI for OpenTelemetryといったクラウドネイティブのサステナビリティの取り組みは、ソフトウェアがどれだけのエネルギーを消費しているかを計測します。しかし、そのエネルギーをどのように予測し、需給を調整し、届けるかを決めているのは、メーターの向こう側にある電力系統（グリッド）のソフトウェアです。Linux Foundationのエネルギー分野向けオープンソース財団であるLF Energyは、この領域で30を超えるプロジェクトをホストしています。その対象は送電系統のモデリング、変電所の仮想化、スマートメーター連携、EV充電にまで及び、CNCFとよく似たSandboxからGraduatedまでのプロジェクトライフサイクルを採用しています。 本セッションでは、LF Energyを紹介するとともに、9月にベルリンで開催されたLF Energy Summit Europe 2026のハイライトを、CNCFコミュニティに特に関連の深いセッションを中心に共有します。あわせて、Summitで私たちが発表した、複数のLF Energyプロジェクトを一つの運用フローとしてつなぐプロトタイプを簡単に紹介し、CNCFとLF Energyのコミュニティが連携できるポイントを提案します。

   An Introduction to LF Energy, with Updates from LF Energy Summit Europe 2026

   Cloud native sustainability work, such as Kepler and SCI for OpenTelemetry, measures how much energy our software uses. But how that energy is forecast, balanced, and delivered is decided by grid software on the other side of the meter. LF Energy, the Linux Foundation's open source foundation for the energy sector, hosts more than three dozen projects on that side. They span transmission system modeling, substation virtualization, smart meter integration, and EV charging, and follow a Sandbox-to-Graduated project lifecycle much like CNCF's. In this session, I will introduce LF Energy and share highlights from LF Energy Summit EU 2026, held in Berlin in September, focusing on the sessions most relevant to the CNCF community. I will also briefly present our Summit talk, a prototype that connects several LF Energy projects into one operational chain, and suggest where the CNCF and LF Energy communities could work together.

   SPEAKERS
5. 6:25 PM - 6:35 PM JST

   Break（休憩）

   in-person
6. 6:35 PM - 6:45 PM JST

   Kepler Project Update: Technical Reset - Community Rebuild

   in-person

   Kepler (Kubernetes-based Efficient Power Level Exporter) was born to answer a fundamental cloud-native question: how much energy is your cluster actually consuming? Joining the CNCF as a Sandbox project in 2022, Kepler quickly gained momentum as a pioneer in pod-level power observability. But as cloud-native workloads evolved, we hit architectural walls around heavy root privileges, fragile eBPF dependencies, and complex power modeling. So, we fought back.

   In this fast-paced 5-minute update, I will share the story of Kepler’s ground-up v0.10+ reboot. Learn how we stripped out heavy dependencies to deliver a low-privilege, read-only core that runs safely anywhere. I'll also share our latest breakthroughs in native pod-level GPU power modeling, VM power accounting, and bare-metal Redfish BMC integration. Finally, see how our updated open governance is bringing contributors back to power sustainable cloud computing.

   SPEAKERS
7. 6:45 PM - 6:55 PM JST

   CoHDI: Optimizing GPU Utilization and Power Efficiency for AI workloads

   in-person

   Composable Disaggregated Infrastructureは、従来のようにサーバ内にリソースを固定配置するアーキテクチャではなく、GPUやメモリなどのリソースを独立して管理し、必要に応じて動的に割り当てることで、システム全体の利用効率向上と消費電力削減を実現するアーキテクチャである。

   また、オープンソースCoHDIは、AIワークロードの実行基盤として本アーキテクチャとKubernetesをシームレスに結合するCNCFのsandboxプロジェクトである。

   本セッションでは、CoHDIの最新状況およびCoHDIとAIワークロード管理向けオープンソースソフトウェアを連携させて実際に実現した、GPUリソース最適配置の事例を紹介し、GPU利用率の向上や消費電力削減などの効果を示す。

   Composable Disaggregated Infrastructure (CDI) is an architecture that improves overall system utilization and reduces power consumption by managing resources such as GPUs and memory independently and dynamically allocating them as needed, rather than statically attaching resources within individual servers.

   CoHDI is an open source CNCF Sandbox project that seamlessly integrates this architecture with Kubernetes to provide an execution platform for AI workloads.

   In this session, we will introduce the latest updates on CoHDI and present a practical example of GPU resource optimization achieved by integrating CoHDI with open source software for AI workload management. We will demonstrate its benefits, including improved GPU utilization and reduced power consumption.

   SPEAKERS
8. 6:55 PM - 7:05 PM JST

   AIBOM from Green Software Perspective

   in-person

   This talk explains the role of the AI Bill of Materials (AIBOM) as a versioned, machine-readable inventory that links AI components (models, data, infrastructure) with provenance, rights, integrity, and evaluation evidence. Key message: AIBOM does not measure energy itself but enables measured energy values to be tied to the specific model version, activity (training, fine-tuning, inference), measurement boundary, and evidence URI so the information can be passed through the supply chain for comparison and tracking. The presentation shows that energy is not a mandatory minimum SBOM-for-AI element but can be recorded using existing formats (SPDX 3.0.1’s AIPackage.energyConsumption and CycloneDX 1.7 Model Card’s energyConsumptions[]), and gives a practical example recording inference energy for an internal FAQ model. It emphasizes that operational alignment—measurement rules, common property names, and evidence review/update processes—is required for AIBOM to support meaningful cross-organization comparisons.

   SPEAKERS
9. 7:05 PM - 7:15 PM JST

   Sustainability is not all about efficiency: Research Trends in Sustainable and Ethical AI Development

   in-person

   Sustainable computing has traditionally focused on the efficient use of computing resources like CPUs, GPUs, and memory. AI has followed the same path. The speaker's own Green Software research has focused on this kind of optimization, too.

   Recently, however, AI sustainability is also becoming part of AI ethics. A ICSE (Inernational Conference on Software Engineering) 2026 study “What makes Code Generation Ethically Sourced?"(Xu et al.) identifies environmental sustainability alongside data privacy and IP rights as a core dimension of ethical AI. Efficiency alone is not enough.

   One important step toward responsible AI is therefore to make its environmental impact visible rather than leaving it as a black box. The speaker will introduce recent research on making AI's carbon footprint visible across its entire lifecycle.

   Attendees will gain a broader perspective on AI sustainability beyond models and hardware, with new insights into how the community can contribute to sustainable and ethical AI development.

   SPEAKERS
10. 7:15 PM - 7:25 PM JST

    Dead Container Walking: Post-Incident Analysis When Your Workload Is Already Gone

    virtual

    In Kubernetes, compromised containers often disappear before investigators can respond, leaving no disk, process state or live system to inspect.

    This talk explores how attackers exploit ephemeral workloads and why traditional forensic approaches fail in cloud-native environments.

    Attendees will leave with a blueprint for building forensically ready clusters where even the most ephemeral workloads leave a permanent mark.

    Join session
    SPEAKERS
11. 7:25 PM - 8:00 PM JST

    Networking Session（ネットワーキング）

    in-person

No sessions match the selected labels.