DocsICP FilingConsole
官方文档
首页

Create an E-HPC cluster

更新时间：
复制 MD 格式
产品详情
我的收藏

Configure hardware, software, and basic settings to create an E-HPC cluster using the console wizard.

Prerequisites

Limits and billing

Before you create an E-HPC cluster, note the following:

  • You can create up to three clusters per region. To request more, submit a ticket.

  • Cluster creation incurs E-HPC service fees and resource fees. Billing overview.

  • After creating a cluster, manage all nodes in the E-HPC console. Do not modify nodes directly in the ECS console except in special cases.

Step 1: Configure hardware

Hardware settings include the region, deployment mode, node types, networking, and storage.

Select parameters based on your requirements.

  1. Log on to the E-HPC console.

  2. In the upper-left corner of the top navigation bar, select a region.

  3. In the left-side navigation pane, click Cluster.

  4. On the Cluster page, click Create Cluster in the upper-right corner.

  5. On the Hardware Configurations page, configure the hardware settings.

    Parameter

    Description

    Zone

    Select the availability zone for the cluster.

    Note

    All nodes in a cluster must be in the same region and zone. Regions and zones.

    Pricing Model

    The pricing model for the cluster nodes. This model does not include fees for EIPs or NAS.

    • Subscription: A prepaid model where you purchase and renew resources by week, month, or year.

    • Pay-as-you-go: A postpaid model where you are charged hourly based on actual usage.

    • Preemptible Instance: Only compute nodes can be created as preemptible instances. Management and logon nodes are created as pay-as-you-go instances. This is a postpaid model.

    Billing overview.

    Deploy Mode

    The deployment mode for the cluster. Valid values:

    • Standard: The logon, management, and compute nodes are deployed on separate instances.

    • Tiny: The logon and management nodes are deployed on the same instance, while compute nodes are deployed on separate instances.

    Important

    If you want to use the Open Grid Scheduler (SGE) scheduler, you must select the Tiny deployment mode.

    Node specifications and quantity

    Configure the instance type and quantity for each type of node based on the selected deployment mode.

    • Select an instance type based on your workload. For example, select a GPU-accelerated type for molecular dynamics. Instance types and Select instance types.

      Note

      To create a cluster based on the Yitian chip platform, select a corresponding instance type, such as ecs.g8m.large. This feature is in invitational preview. To use this feature, apply on the preview page.

    • Recommended management node specifications:

      • 100 or fewer compute nodes: at least 16 vCPUs and 64 GiB of memory.

      • Up to 500 compute nodes: at least 32 vCPUs and 128 GiB of memory.

      • More than 500 compute nodes: at least 64 vCPUs and 256 GiB of memory.

    • The logon node serves as a development and debugging environment. Select an instance with the same configuration as compute nodes or with a higher memory-to-CPU ratio.

    System Disk

    The cloud disk type and capacity for all node system disks. Valid range: 40 GB to 2,000 GB.

    Note

    A capacity larger than 500 GB requires approval. You can submit a ticket to apply.

  6. Expand Advanced Configurations and configure networking and storage.

    Parameter

    Description

    Authorized Instance Configurations

    Enabled

    Binds a RAM role to the nodes, allowing them to access other Alibaba Cloud services.

    Important

    This feature is disabled by default. To use it, submit a ticket.

    After approval, configure based on your account type:

    • Alibaba Cloud account (primary account): Click Switch to RAM for authorization to authorize the system default role.

    • RAM user (sub-account): Log on to the RAM console with your Alibaba Cloud account (primary account) and grant permissions to the RAM user using one of the following methods:

      • Create the following custom policy and attach it to the RAM user. Create a custom policy and Grant permissions to a RAM user.

        {
    "Version": "1",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "ram:PassRole",
                "ram:ListRoles"
            ],
            "Resource": "*"
        },
        {
            "Effect": "Allow",
            "Action": "ecs:AttachInstanceRamRole",
            "Resource": "*"
        }
    ]
}

      • Attach the AliyunRAMFullAccess policy directly to the RAM user.

        AliyunRAMFullAccess grants full RAM management permissions, which is broader than the custom policy above. Grant permissions to a RAM user.

    Role Name

    The RAM role to bind to nodes. Recommended: AliyunECSInstanceForEHPCRole.

    Node Type

    The node types to which the RAM role will be bound. Valid values:

    • Scheduler node

    • Account node

    • Logon node

    • Compute node

    Note

    If you select compute node, E-HPC automatically binds the RAM role to any new compute nodes added during a cluster scale-out.

    Resource Group

    Resource Group

    The resource group for the cluster nodes. Use resource groups to centrally manage nodes under your account.

    Network Configuration

    EIP

    An Elastic IP Address (EIP) is a public IP address that you can purchase and hold independently for cluster access.

    • Use: Automatically creates an EIP and associates it with the logon node. You can then access the cluster over the internet.

    • Disable: You can access the cluster only through the VPC.

    Note

    EIP usage incurs charges. Billing overview.

    VPC and vSwitch

    The VPC for the cluster. VPCs are logically isolated from each other.

    The first VPC and vSwitch you created are selected by default. Ensure the vSwitch has enough IP addresses for all nodes.

    You can also click Create VPC and Create vSwitch (for subnet). Create a VPC and Create a vSwitch.

    Create Security Group

    A security group controls inbound and outbound traffic for nodes.

    • Turn on: If you turn on this switch, you must enter a name for the new security group in the Security Group Name field.

    • Turn off: If you turn off this switch, you must select an existing security group from the Select Security Group list.

    Shared Storage

    Configure by Directory

    • Turn off: Only one file system is configured for the cluster.

    • Turn on: Mounts a file system to the directories of all nodes to enhance the shared storage capacity of the cluster.

    Type

    The type of file system to mount to the cluster.

    • General-purpose NAS: Mounts a General-purpose NAS file system.

    • Extreme: Mounts an Extreme NAS file system.

    • CPFS-NFS: Mounts a CPFS file system by using the NFS protocol.

    • CPFS-POSIX: Mounts a CPFS file system by using the POSIX protocol.

    • Input: For file storage not on Alibaba Cloud, such as a self-managed NAS, you must manually enter information such as the mount protocol and mount target.

    Important

    CPFS supports two mount methods with different operations. Create an E-HPC cluster that uses a CPFS file system as shared storage.

    The ID of the file system. and Mount Point

    The first file system and mount target you created are selected by default. Ensure the file system has sufficient quota for new mount targets.

    You can also click Create a file system and Create mount point to create a new file system and mount target.

    Mount Options

    When you mount a General-purpose NAS file system, you can select the mount protocol. Both NFSv3 and NFSv4 are supported.

    Remote Directory

    The remote directory of the file system to be mounted.

Step 2: Configure software

Software settings include the image, scheduler, and domain service for the nodes.

  1. After you configure the hardware settings, click Next.

  2. Configure the software settings.

    Parameter

    Description

    ImageType and Image

    Select an image type based on your requirements. E-HPC supports the following image types:

    • Public image

    • Custom image

    • Shared image

    • Marketplace image

    • Community image

    • Compute Nest deployment

    Custom images have the following limitations:

    • You can use custom images created from official Alibaba Cloud images or imported CentOS images. When you import an image, make sure to select Check After Import. Otherwise, the image cannot be recognized in the E-HPC console.

    • Do not use a custom image created from a node of an existing E-HPC cluster. Otherwise, errors may occur when you create compute nodes for the new cluster.

    • Do not modify the yum repository configuration in the custom image. Otherwise, you may fail to create or scale out the cluster.

    • The mount path in a custom image, which is the path where the NAS file system is mounted by using the mount command, must not contain the /home or /opt directory.

    After selecting an image type, choose the specific image. The system deploys nodes based on your selection.

    Important

    The console displays only images available for the selected region, zone, account, and instance types.

    Scheduler

    The job scheduling software for the cluster.

    Available schedulers vary by image. The console shows only supported options.

    Domain Service

    The domain service for managing cluster users. Supported types: nis and ldap.

    VNC

    Enables the remote visualization service during cluster creation, providing graphical desktop access from the console.

    eRDMA

    Attaches an eRDMA network interface to compute nodes for high-performance RDMA networking. eRDMA overview.

  3. Configure queues and a post-install script.

    Parameter

    Description

    Queue Configuration

    Queue

    Group compute nodes into queues for job scheduling and filtering.

    • Default Queue: The compute nodes selected during cluster creation are automatically added to the default queue of the scheduler. The default queue is workq for PBS clusters and comp for Slurm clusters.

    • New Queue: Enter a name for the new queue in the Queue Name field. The queue is automatically created, and the selected compute nodes are added to this queue.

    Post-Install Script

    Script URL

    The URL of the custom script to run after cluster installation.

    Note

    HTTP and HTTPS are supported. Store the script in a public-read OSS bucket.

    Arguments

    The arguments for the script. Configure a post-install script.

Step 3: Configure basic information

  1. After you configure the software settings, click Next.

  2. Configure the basic information and logon settings.

    Parameter

    Description

    Name

    The cluster name displayed in the cluster list.

    Logon Password and Confirm Password

    The password for SSH access to the logon node as the root user.

    Cluster Version

    The version of the cluster. The default value is 2.0.

  3. In the Configuration List pane on the right, review the cluster configurations. Read and select Alibaba Cloud International Website Product Terms of Service, and then click OK.

Results

After the cluster is created, you can view its status on the Cluster page. The cluster is created successfully when the status of the cluster and all its nodes changes to Running.

Related topics

Previous：Create a clusterNext：Create an E-HPC cluster with CPFS storage