LecturaInvestigaciónRadarMarco de inversión
Iniciar sesión / Registrarse
← Volver a la lectura
NoticiasTema · K8s动态资源分配Resumen rápido1 min de lecturaEditorial de DisconfirmAI

PyTorch: Paco Xu & Kang Zhang on Kubernetes DRA v1.36 for AI/HPC

Paco Xu and Kang Zhang present the Kubernetes Dynamic Resource Allocation (DRA) architecture in v1.36 on the PyTorch official channel, covering GPU, NIC, and NUMA topology-aware scheduling, along with an IMEX daemon case study.

2026-09-25 04:34 GMT+8

On September 24, 2026, the PyTorch official channel released a video presented by Paco Xu and Kang Zhang, detailing the architectural evolution of Kubernetes Dynamic Resource Allocation (DRA) in version 1.36. The session addresses the complex requirements of AI and High-Performance Computing (HPC) workloads for GPUs, Network Interface Cards (NICs), NUMA locality, and PCIe/fabric relationships, demonstrating how DRA unifies handling across entry points, resource modeling, scheduler allocation, binding, kubelet preparation, status, health, and access boundaries.

Key topics include topology-aware scheduling for supernode systems, where NUMA, PCIe, NVLink, or fabric placement determines performance. Additionally, the presentation features a case study on IMEX daemon/channel assignment: the design evolved from distributed status updates with conflicts to centralized updates, and finally to distributed updates on a topology-aligned state object, reducing assignment latency from minutes to seconds at thousand-device scale.

Fuentes

  • youtube.com↗
Tema

K8s动态资源分配

Kubernetes动态资源分配(DRA)在v1.36版本中正式成为通用可用(GA)功能,标志着GPU调度从传统的“整数计数”模式转向基于属性的结构化声明。这一转变解决了长期存在的AI集群资源利用率低问题,允许Pod直接请求特定型号、显存大…

Iniciar sesión / Registrarse
LecturaInvestigaciónRadarMarco de inversión

4 pasos1 piezasActualizado 2026-10-15 13:37

Cómo llegó hasta aquí

  1. 2024-04-01Kubernetes v1.30引入结构化参数(Structured Parameters),为DRA奠定基础,但此时仍主要处于Alpha/Beta阶段。v1-32.docs.kubernetes.io ↗
  2. 2025-04-01Kubernetes v1.34将DRA提升为稳定版(Stable/GA),用户无需再启用特性门控即可使用基础DRA功能。kubernetes.recipes ↗
  3. 2026-04-22Kubernetes v1.36发布,DRA全面GA,新增可分区设备(Beta)和设备绑定条件,支持动态GPU共享和更复杂的拓扑感知调度。ettayeb.fr ↗

Lo que aún está pendiente

  • 非NVIDIA厂商(如AMD、Intel)的DRA驱动何时能完全对齐v1.36的可分区设备特性?
  • 在大规模生产环境中,DRA调度器的性能开销与传统设备插件相比是否有显著差异?
  • Kubernetes v1.37中设备兼容性组(Alpha)何时会进入Beta并默认启用?
Leer el tema completo →
Continuar investigando →