Introduction
MindCluster Ascend Deployer provides automated downloading and one-click installation of OS dependencies and Docker, and supports online and offline downloading, installation, and upgrade of software packages such as drivers, firmware, CANN, MindCluster components (performance testing, fault diagnosis, and cluster scheduling), AI frameworks (TensorFlow, MindSpore, and PyTorch), and MindIE images.
Intended Audience
This document is primarily intended for general developers.
If Huawei technical support engineers need to install Ascend software, please select the appropriate Cluster Delivery Guide based on the actual hardware scenario.
Key Features
Downloading Ascend Software
MindCluster Ascend Deployer can download software on Linux and Windows systems for Ascend hardware in inference and training scenarios. It supports downloading Ascend software, Docker, and system dependencies that are compatible with the hardware and OS versions.
Installing Ascend Software
MindCluster Ascend Deployer can install Ascend software on a single node or in batches for Ascend hardware in inference and training scenarios. Running MindCluster Ascend Deployer enables automated deployment on multiple devices simultaneously, saving installation and deployment time.
Upgrading Ascend Software
MindCluster Ascend Deployer provides single-node and batch upgrade capabilities for Ascend hardware in inference and training scenarios. It currently supports upgrading the following software: NPU driver and firmware, MCU firmware, CANN, and MindCluster components (performance testing, fault diagnosis, and cluster scheduling).
Constraints and Precautions
- The OS dependencies pre-installed by MindCluster Ascend Deployer are only applicable to initial server deployment scenarios (i.e., the server has only completed OS installation without deploying other service components).
- When using MindCluster Ascend Deployer to perform installation or upgrade operations, core services such as Containerd, Docker, and Kubelet may be restarted, causing temporary interruption or abnormal status of related container services in the environment. Please conduct a risk assessment in advance.
- MindCluster Ascend Deployer does not support online capacity expansion or online upgrade operations. Before performing installation or upgrade operations, you must stop all related services first, and resume service running after the operations are completed.
Workflow
The detailed workflow for using MindCluster Ascend Deployer is shown in the following figure.
