Cluster Scheduling Scenario

Scenario

In this scenario, you already have a Kubernetes cluster and need to manage new NPU servers. In addition, features such as NPU Device Management, NPU Scheduling Optimization, Resumable Training, and Rescheduling Upon Inference Card Faults can be used. You need to deploy the NPU management component on the master node of the existing Kubernetes cluster and deploy the NPU management component on the worker node on the newly managed NPU node.

List of Components to Be Installed

Table 1 Components to be installed in the cluster scheduling scenario

Component

Description

Ascend Docker Runtime

Allows containers to use Ascend NPUs.

Ascend Device Plugin

Supports NPU device management.

Volcano

Supports NPU scheduling optimization, resumable training, and rescheduling upon inference card faults.

(Optional) HCCL-Controller

Generates the ranktable file (also called the hccl.json file) for NPU training jobs. Install it when you need to use the function.

(Optional) NodeD

Supports resumable training upon node faults. Install it when you need to use the function.

(Optional) NPU-Exporter

Supports monitoring of NPU device management status. Install it when you need to use the function.