CCE Ascend Mindx DL Description

Component introduction

The CCE Ascend MindX DL component is a deep learning solution built on Ascend AI processors. It offers essential features such as scheduling for Ascend AI processors and cluster performance testing, while providing foundational software support for higher-level applications like model training, deployment, and inference.

Component function

• Cluster scheduling: Improves scheduling capabilities for Ascend AI processors within Kubernetes and allows users to monitor the status of Ascend nodes and AI processors.

Application scenarios

This enables CCE clusters to seamlessly work with Ascend AI processors. Through advanced cluster scheduling tools, you can efficiently deploy and manage AI workloads on Ascend AI processors, offering containerization support for all AI tasks on these processors.

Note

Before installing the CCE Ascend MindX DL component, ensure that the conflicting component CCE AI Job Scheduler is not installed in the cluster.

Install component

1. Sign in to the Baidu AI Cloud official website and enter the management console.
2. Go to Product Services - Cloud Native - Cloud Container Engine (CCE) to access the CCE management console.
3. Click Cluster Management - Cluster List in the left navigation bar.
4. Click on the target cluster name in the Cluster List page to navigate to the cluster management page.
5. On the Cluster Management page, click O&M & Management > Component Management.
6. In the component management list, select the CCE Ascend MindX DL component and click "Install".

Component status confirmation

Use the following command to inspect Pods associated with the CCE Ascend MindX DL component in the K8S cluster. The component operates correctly only if the STATUS for the listed Pods is "Running" and the READY status is 1/1.

kubectl get pods -n volcano-system

1NAME                                                   READY   STATUS    RESTARTS        AGE
2volcano-controllers-86bxxxxxx1-xxxx1                   1/1     Running   0               038m
3volcano-controllers-86bxxxxxx2-xxxx2                   1/1     Running   0               038m
4volcano-controllers-86bxxxxxx3-xxxx3                   1/1     Running   0               138m
5volcano-node-controllers-6d8b9xxxxx-xxxxx              1/1     Running   0               09m10s
6volcano-scheduler-5bbxxxxxx1-xxxx1                     1/1     Running   0               038m
7volcano-scheduler-5bbxxxxxx2-xxxx2                     1/1     Running   0               138m
8volcano-scheduler-5bbxxxxxx3-xxxx3                     1/1     Running   0               038m

kubectl get pods -n kube-system

1NAME                                                   READY   STATUS    RESTARTS        AGE
2ascend-device-plugin-daemonset-xxxxx                   1/1     Running   0               39m
3npu-exporter-xxxxx                                     1/1     Running   0               39m

If there are no Ascend-corresponding nodes in the current cluster, the replica counts of ascend-device-plugin-daemonset and npu-exporter will be 0.

kubectl get pods -n mindx-dl

1NAME                                                   READY   STATUS    RESTARTS        AGE
2ascend-operator-manager-xxxxxxx-xxxxx                  1/1     Running   0               038m