All Products
Search

This Product

    Cloud Parallel File Storage:Mount CPFS AI Edition file systems on PAI

    Document Center

    Cloud Parallel File Storage:Mount CPFS AI Edition file systems on PAI

    Last Updated:Oct 09, 2026

    PAI (Platform for AI) supports mounting CPFS AI Edition in DLC, DSW, and EAS. This enables single-copy storage with multi-point access for data, code, and models, keeping paths consistent across all stages without redundant copies or code changes.

    Prerequisites

    • You have created a CPFS AI Edition file system.

    • CPFS AI Edition provides two access channels: VSC and VPC. The VSC channel uses the RDMA protocol and can fully utilize the file system's peak performance. The VPC channel uses the TCP protocol, with a default read bandwidth of 10 GB/s and write bandwidth of 5 GB/s per file system. For the complete performance specifications of both channels, see Product specifications. Complete the corresponding network configuration based on the resource type you use. For details about resource types, see PAI resource overview.

      • Lingjun AI computing resources: First determine whether the Lingjun resources and CPFS are in the same compute-storage affinity domain (see the instructions below). If they are in the same affinity domain, the VSC channel is used with no additional configuration required. If they are not in the same affinity domain, the VPC channel is used, and you need to create a VPC mount point for the CPFS file system.

      • General-purpose computing resources: The VPC channel is used. You need to create a VPC mount point for the CPFS file system, and the VPC used by the DLC/DSW/EAS task must be the same VPC as the VPC mount point, or the VPCs must be connected through Cloud Enterprise Network (CEN).

    Note

    How to determine whether Lingjun and CPFS are in the same compute-storage affinity domain

    1. Call DescribeFilesystemsAssociatedHpnZones, passing in the target CPFS file system ID, to obtain the list of Lingjun HPNZones that share the same compute-storage affinity domain with the file system.

    2. Query the HPNZone of the Lingjun resources you plan to use through the Lingjun OpenAPI or the PAI console.

    3. If the Lingjun HPNZone matches an entry in the list, the resources are in the same compute-storage affinity domain. If not, they are in different affinity domains.

    Always rely on the API response as the authoritative result. Do not judge solely based on whether the availability zone or HPNZone values are the same. Compute and storage resources with the same cluster number belong to the same compute-storage affinity domain. Resources with different cluster numbers may still belong to the same affinity domain. However, Lingjun and CPFS resources in different availability zones are never in the same compute-storage affinity domain.

    Mount CPFS in DLC

    When creating a DLC training task, you can configure CPFS storage through direct mount or dataset mount. For the complete DLC training task configuration, see Create a training job.

    Note

    • If the resource quota is general-purpose computing resources, you must configure a VPC, and the selected VPC must be the same as the VPC of the CPFS mount point. If a VPC has not been configured for the CPFS AI Edition file system, configure it throughManage VPC mount points .

    • If Lingjun resources are used and cross-affinity-domain access occurs, the system automatically switches to the VPC access channel. You also need to create a VPC mount point for the CPFS file system.

    Use Direct Mount for one-off parameter tuning. Use Dataset Mount for team collaboration or frequent task creation.

    Direct Mount

    1. Log on to the PAI console and go to the target workspace.

    2. In the left-side navigation pane, choose Distributed Training (DLC) and click Create Task.

    3. In the Environment Information section, under Storage Mount, click CPFS for AI.

    4. Configure the following mount parameters:

      Parameter

      Description

      File System

      Select the CPFS AI Edition file system instance you created.

      File System Mount Point

      (Optional) Select the VPC mount point domain name of the CPFS file system. This is required only when the computing resources and the CPFS file system are not in the same compute-storage affinity domain. If they are in the same affinity domain, access is provided through the VSC channel directly, and you do not need to fill in this field.

      For how to determine the compute-storage affinity domain, see Prerequisites.

      File System Path

      Specify the CPFS directory to mount. / mounts the entire file system. You can also specify a subdirectory.

      Mount Path

      The path within the service instance for accessing CPFS, such as /mnt/data/.

      Read-only

      Set the read/write permissions for the mounted directory.

    5. After completing other configurations, submit the task.

    Dataset Mount

    To reuse storage configurations, you can use the dataset mount method. First, Create a CPFS AI Edition dataset.

    1. Log on to the PAI console and go to the target workspace.

    2. In the left-side navigation pane, choose Distributed Training (DLC) and click Create Task.

    3. In the Dataset Mount section, click Custom Dataset.

    4. Switch to the File Storage - CPFS for AI tab, select the dataset you created, and click OK.

    5. Confirm the mount path and set Read-only as needed.

    6. After completing other configurations, submit the task.

    Mount CPFS in DSW

    When creating a DSW instance, you can configure CPFS storage through direct mount or dataset mount. For the complete DSW instance configuration, see Create a DSW instance.

    Note

    • If the resource quota is general-purpose computing resources, you must configure a VPC, and the selected VPC must be the same as the VPC of the CPFS mount point. If a VPC has not been configured for the CPFS AI Edition file system, configure it throughManage VPC mount points .

    • If Lingjun resources are used and cross-affinity-domain access occurs, the system automatically switches to the VPC access channel. You also need to create a VPC mount point for the CPFS file system.

    Use Direct Mount for one-off parameter tuning. Use Dataset Mount for team collaboration or frequent instance creation.

    Direct Mount

    1. Log on to the PAI console and go to the target workspace.

    2. In the left-side navigation pane, choose Interactive Modeling (DSW) and click Create Instance.

    3. In the Environment Information section, under Storage Mount, click CPFS for AI.

    4. Configure the mount parameters:

      Parameter

      Description

      File System

      Select the CPFS AI Edition file system instance you created.

      File System Mount Point

      (Optional) Select the VPC mount point domain name of the CPFS file system. This is required only when the computing resources and the CPFS file system are not in the same compute-storage affinity domain. If they are in the same affinity domain, access is provided through the VSC channel directly, and you do not need to fill in this field.

      For how to determine the compute-storage affinity domain, see Prerequisites.

      File System Path

      Specify the CPFS directory to mount. / mounts the entire file system. You can also specify a subdirectory.

      Mount Path

      The path within the service instance for accessing CPFS, such as /mnt/data/

      Read-only

      Set the read/write permissions for the mounted directory.

    5. After completing other configurations, create the instance.

    Dataset Mount

    Datasets save storage configurations for reuse across multiple tasks. First, Create a CPFS AI Edition dataset.

    1. Log on to the PAI console and go to the target workspace.

    2. In the left-side navigation pane, choose Interactive Modeling (DSW) and click Create Instance.

    3. In the Dataset Mount section, click Custom Dataset.

    4. Switch to the File Storage - CPFS for AI tab, select the dataset you created, and click OK.

    5. After completing other configurations, create the instance.

    Mount CPFS in EAS

    When deploying an EAS service, you can configure CPFS storage through direct mount. For the complete EAS service configuration, see Custom deployment.

    Note

    • EAS supports only direct mount and does not support dataset mount. If the resource quota is set to general-purpose computing resources, you must configure a VPC, and the selected VPC must be the same as the VPC of the CPFS mount point. If a VPC has not been configured for the CPFS AI Edition file system, first Manage VPC mount points.

    • If Lingjun resources are used and cross-affinity-domain access occurs, the system automatically switches to the VPC access channel. You also need to create a VPC mount point for the CPFS file system.

    1. Log on to the PAI console and go to the target workspace.

    2. In the left-side navigation pane, choose Elastic Algorithm Service (EAS) and select the Inference Service tab.

    3. Click Deploy Service > Custom Deployment.

    4. In the Environment Information section, under Storage Mount, click CPFS for AI.

    5. Configure the following mount parameters:

      Parameter

      Description

      Storage Permission Resource Group

      Select the storage permission resource group.

      File System

      Select the CPFS AI Edition file system instance you created.

      File System Path

      Specify the CPFS directory to mount. / mounts the entire file system. You can also specify a subdirectory.

      Mount Path

      The path within the service instance for accessing CPFS, such as /mnt/data/.

      Read-only

      Set the read/write permissions for the mounted directory.

    6. After completing other configurations, deploy the service.

    Create a CPFS AI Edition dataset

    A dataset is a template of storage configurations that records file system, directory path, mount path, and other information. After creation, it can be reused by multiple DLC/DSW tasks, eliminating the need for repetitive configuration.

    EAS does not support dataset mount.

    1. Log on to the PAI console, go to the target workspace, and choose AI Asset Management > Datasets.

    2. On the Custom Datasets tab, click Create Dataset and configure the following parameters:

      Section

      Parameter

      Description

      Dataset Configuration

      Name

      The dataset name.

      Dataset Type

      Select Basic.

      Storage Type

      Select File Storage (CPFS for AI).

      Content Type

      Select the data type: Image, Text, Audio, Video, Table, or General.

      Version Configuration

      Storage Permission Resource Group

      Select the resource group that has access permissions to the target CPFS file system.

      File System

      Select the CPFS AI Edition file system instance you created.

      File System Path

      Specify the directory in CPFS. / mounts the entire file system. You can also specify a subdirectory such as /datasets/train

      Default Mount Path

      The path within the service instance for accessing CPFS, such as /mnt/data/.

    3. Click Confirm.