All Products
Search

This Product

    Platform For AI:Create and manage slots

    Document Center

    Platform For AI:Create and manage slots

    Last Updated:Oct 31, 2024

    Before you can use Dataset Accelerator of Platform for AI (PAI) to accelerate training data, you need to create a slot to associate the storage address of the data source. After you create slots, the system preprocesses the associated datasets based on data type, data size, framework, and model used for the training to accelerate dataset access. This topic describes how to create and manage slots.

    Prerequisites

    An accelerator is created. For more information, see Create and manage accelerators.

    Create a slot

    1. Log on to the PAI console.

    2. Go to the Create Slot panel.

      You can go to the Create Slot panel by using one of the following methods.

      • Method 1: Go to the Slots tab.

        1. In the left-side navigation pane, choose AI Acceleration > Dataset Accelerator.

        2. On the Slots tab, click Create Slot.

      • Method 2: Go to the details page of the accelerator.

        Note

        If you want to create a slot on the details page of the accelerator, set the value of the Associated Instance parameter to the selected accelerator.

        1. In the left-side navigation pane, choose AI Acceleration > Dataset Accelerator.

        2. On the Accelerators tab, click the name of the accelerator that you want to view.

        3. On the Slots tab, click Create Slot.

    3. In the Create Slot panel, configure the parameters. The following table describes the parameters.

      Parameter

      Description

      Associated Instance

      Select an existing accelerator.

      Associated Cloud Services

      You can associate slots with the cloud service that you specified for the Storage Type parameter of the accelerator. Valid values:

      Alibaba Cloud OSS

      • Read/Write Attributes: the read and write permissions on Object Storage Service (OSS). Valid values: Read-only and Read/Write.

      • OSS Path: Select the OSS directory in which the dataset is stored.

      General-purpose NAS and Extreme NAS

      • Select File System: the ID of an existing File Storage NAS (NAS) file system.

      • Mount Target: the mount target that you want to use to access the file system.

      • File System Path: an existing storage path in the NAS file system. Default value: /.

      Capacity

      The capacity of the slot. You can configure this parameter based on the capacity of the dataset that you want to accelerate.

      Important

      The slot capacity must be greater than or equal to the dataset capacity.

      Mount Target

      This parameter is available only if you set the Associate Cloud Services parameter to Alibaba Cloud OSS.

      You can use the mount target to mount the accelerator to the training cluster of the training job. When you train a model in a training cluster, the training job can access the slot by using the mount target and read the dataset associated with the slot. This improves the efficiency of dataset access.

      You can configure the mount target by using one of the following methods:

      • Select Mount Target: select an existing mount target.

      • Create Mount Target: Configure the following parameters to create a mount target.

        • Type: Only VPC is supported.

        • VPC: Select an existing virtual private cloud (VPC). If no VPC is available, you can click Create VPC to create a VPC.

        • vSwitch: Select an existing vSwitch. If no vSwitch is available, you can click Create vSwitch to create a vSwitch.

      Note

      • Dataset Accelerator supports only mount targets of the VPC type.

      • A mount target of a VPC can be used by training clusters that use different vSwitches in the same VPC.

      • When you create a mount target for a file system, the file system is assigned an IP address. We recommend that you select a vSwitch that has a large number of available internal IP addresses.

      • When you use the mount target, you can mount only training clusters that belong to the same VPC as the mount target.

      Enable Automatic Recycling

      If you enable automatic recycling, you can configure the running duration or stop time of the slot. The system automatically stops the slot at the specified time.

    4. Click Submit.

      The system initializes the slot. When the Status changes to Running, the slot is ready for use.

    Manage a slot

    You can manage a slot on the Slots page or on the details page of an accelerator.

    • Click the name of the slot. On the slot details page, view the basic information about the slot.

    • Move the pointer over ② to view the storage type of the dataset in the slot.

    • You can stop or delete a slot that you no longer use or clone a slot. You can also click Automatic Recycling to configure the running duration or stop time of the slot.

    Manage a mount target

    Click the name of a slot to go to the slot details page. On this page, you can manage mount targets. image.png

    • Move the pointer over the name of a mount target to view the configurations of the mount target.

    • Click the name of the mount target. The Deployment Configurations page appears. In the Deployment Configurations panel, click Download to download the YAML file of the Dataset Accelerator client. You can use the YAML file when you configure the accelerator in the training cluster. image

    What to do next

    You can enable the acceleration feature of a dataset when you create a Data Science Workshop (DSW) instance or submit a Deep Learning Containers (DLC) job to accelerate access to the dataset. For more information, see Use Dataset Accelerator.