Migrate workloads from an Azure Stack Edge Pro FPGA to an Azure Stack Edge Pro GPU

This article describes how to migrate workloads and data from an Azure Stack Edge Pro FPGA device to an Azure Stack Edge Pro GPU device. The migration process begins with a comparison of the two devices, a migration plan, and a review of migration considerations. The migration procedure gives detailed steps ending with verification and device cleanup.

Important

Azure Stack Edge Pro FPGA devices will reach end-of-life in February 2024. If you are considering new deployments, we recommend that you explore Azure Stack Edge Pro GPU devices for your workloads.

About migration

Migration is the process of moving workloads and application data from one storage location to another. This entails making an exact copy of an organization’s current data from one storage device to another storage device - preferably without disrupting or disabling active applications - and then redirecting all input/output (I/O) activity to the new device.

This migration guide provides a step-by-step walkthrough of the steps required to migrate data from an Azure Stack Edge Pro FPGA device to an Azure Stack Edge Pro GPU device. This document is intended for information technology (IT) professionals and knowledge workers who are responsible for operating, deploying, and managing Azure Stack Edge devices in the datacenter.

In this article, the Azure Stack Edge Pro FPGA device is referred to as the source device and the Azure Stack Edge Pro GPU device is the target device.

Comparison summary

This section provides a comparative summary of capabilities between the Azure Stack Edge Pro GPU vs. the Azure Stack Edge Pro FPGA devices. The hardware in both the source and the target device is largely identical; only the hardware acceleration card and the storage capacity may differ.

Capability Azure Stack Edge Pro GPU (Target device) Azure Stack Edge Pro FPGA (Source device)
Hardware Hardware acceleration: 1 or 2 Nvidia T4 GPUs
Compute, memory, network interface, power supply unit, and power cord specifications are identical to the device with FPGA.
Hardware acceleration: Intel Arria 10 FPGA
Compute, memory, network interface, power supply unit, and power cord specifications are identical to the device with GPU.
Usable storage 4.19 TB
After reserving space for parity resiliency and internal use
12.5 TB
After reserving space for internal use
Security Certificates
Workloads IoT Edge workloads
VM workloads
Kubernetes workloads
IoT Edge workloads
Pricing Pricing Pricing

Migration plan

To create your migration plan, consider the following information:

  • Develop a schedule for migration.
  • When you migrate data, you may experience a downtime. We recommend that you schedule migration during a downtime maintenance window as the process is disruptive. You will set up and restore configurations in this downtime as described later in this document.
  • Understand the total length of downtime and communicate it to all the stakeholders.
  • Identify the local data that needs to be migrated from the source device. As a precaution, ensure that all the data on the existing storage has a recent backup.

Migration considerations

Before you proceed with the migration, consider the following information:

  • An Azure Stack Edge Pro GPU device can't be activated against an Azure Stack Edge Pro FPGA resource. You should create a new resource for the Azure Stack Edge Pro GPU device as described in Create an Azure Stack Edge Pro GPU order.
  • The Machine Learning models deployed on the source device that used the FPGA will need to be changed for the target device with GPU. For help with the models, you can contact Microsoft Support. The custom models deployed on the source device that did not use the FPGA (used CPU only) should work as-is on the target device (using CPU).
  • The IoT Edge modules deployed on the source device may require changes before the modules can be successfully deployed on the target device.
  • The source device supports NFS 3.0 and 4.1 protocols. The target device only supports NFS 3.0 protocol.
  • The source device support SMB and NFS protocols. The target device supports storage via the REST protocol using storage accounts in addition to the SMB and NFS protocols for shares.
  • The share access on the source device is via the IP address whereas the share access on the target device is via the device name.

Migration steps at-a-glance

This table summarizes the overall flow for migration, describing the steps required for migration and the location where to take these steps.

In this phase Do this step On this device
Prepare source device* 1. Record configuration data
2. Back up share data
3. Prepare IoT Edge workloads
Source device
Prepare target device* 1. Create a new order
2. Configure and activate
Target device
Migrate data 1. Migrate data from shares
2. Redeploy IoT Edge workloads
Target device
Verify data Verify migrated data Target device
Clean up, return Erase data and return Source device

*The source and target devices can be prepared in parallel.

Prepare source device

The preparation includes that you identify the Edge cloud shares, Edge local shares, and the IoT Edge modules deployed on the device.

1. Record configuration data

Do these steps on your source device via the local UI.

Record the configuration data on the source device. Use the Deployment checklist to help you record the device configuration. During migration, you'll use this configuration information to configure the new target device.

2. Back up share data

The device data can be of one of the following types:

  • Data in Edge cloud shares
  • Data in local shares

Data in Edge cloud shares

Edge cloud shares tier data from your device to Azure. Do these steps on your source device via the Azure portal.

  • Make a list of all the Edge cloud shares and users that you have on the source device.
  • Make a list of all the bandwidth schedules that you have. You will recreate these bandwidth schedules on your target device.
  • Depending on the network bandwidth available, configure bandwidth schedules on your device to maximize the data tiered to the cloud. That minimizes the local data on the device.
  • Ensure that the shares are fully tiered to the cloud. The tiering can be confirmed by checking the share status in the Azure portal.

Data in Edge local shares

Data in Edge local shares stays on the device. Do these steps on your source device via the Azure portal.

  • Make a list of the Edge local shares on the device.

  • Since you'll be doing a one-time migration of the data, create a copy of the Edge local share data to another on-premises server. You can use copy tools such as robocopy (SMB) or rsync (NFS) to copy the data. Optionally you may have already deployed a third-party data protection solution to back up the data in your local shares. The following third-party solutions are supported for use with Azure Stack Edge Pro FPGA devices:

    Third-party software Reference to the solution
    Cohesity https://www.cohesity.com/solution/cloud/azure/
    For details, contact Cohesity.
    Commvault https://www.commvault.com/azure
    For details, contact Commvault.
    Veritas http://veritas.com/azure
    For details, contact Veritas.
    Veeam https://www.veeam.com/kb4041
    For details, contact Veeam.

3. Prepare IoT Edge workloads

  • If you have deployed IoT Edge modules and are using FPGA acceleration, you may need to modify the modules before these will run on the GPU device. Follow the instructions in Modify IoT Edge modules.

Prepare target device

1. Create new order

You need to create a new order (and a new resource) for your target device. The target device must be activated against the GPU resource and not against the FPGA resource.

To place an order, Create a new Azure Stack Edge resource in the Azure portal.

2. Set up, activate

You need to set up and activate the target device against the new resource you created earlier.

Follow these steps to configure the target device via the Azure portal:

  1. Gather the information required in the Deployment checklist. You can use the information that you saved from the source device configuration.
  2. Unpack, rack mount and cable your device.
  3. Connect to the local UI of the device.
  4. Configure the network using a different set of IP addresses (if using static IPs) than the ones that you used for your old device. See how to configure network settings.
  5. Assign the same device name as your old device and provide a DNS domain. See how to configure device setting.
  6. Configure certificates on the new device. See how to configure certificates.
  7. Get the activation key from the Azure portal and activate the new device. See how to activate the device.

You are now ready to restore the share data and deploy the workloads that you were running on the old device.

Migrate data

You will now copy data from the source device to the Edge cloud shares and Edge local shares on your target device.

1. From Edge cloud shares

Follow these steps to sync the data on the Edge cloud shares on your target device:

  1. Add shares corresponding to the share names created on the source device. When you create the shares, make sure that Select blob container is set to Use existing, and then select the container that was used with the previous device.
  2. Add users that had access to the previous device.
  3. Refresh the share data from Azure. Refreshing the share will pull down all the cloud data from the existing container to the shares.
  4. Recreate the bandwidth schedules to be associated with your shares. See Add a bandwidth schedule for detailed steps.

2. From Edge local shares

You may have deployed a third-party backup solution to protect the local shares data for your IoT workloads. You will now need to restore that data.

After the replacement device is fully configured, enable the device for local storage.

Follow these steps to recover the data from local shares:

  1. Configure compute on the device.

  2. Add all the local shares on the target device. See the detailed steps in Add a local share.

  3. Accessing the SMB shares on the source device will use the IP addresses whereas on the target device, you'll use device name. See Connect to an SMB share on Azure Stack Edge Pro GPU. To connect to NFS shares on the target device, you'll need to use the new IP addresses associated with the device. See Connect to an NFS share on Azure Stack Edge Pro GPU.

    If you copied your share data to an intermediate server over SMB or NFS, you can copy the data from the intermediate server to shares on the target device. If both the source and the target device are online, you can also copy the data directly from the source device.

    If you have used third-party software to back up the data in the local shares, you will need to run the recovery procedure that's provided by the data protection solution of choice. See references in the following table.

    Third-party software Reference to the solution
    Cohesity https://www.cohesity.com/solution/cloud/azure/
    For details, contact Cohesity.
    Commvault https://www.commvault.com/azure
    For details, contact Commvault.
    Veritas http://veritas.com/azure
    For details, contact Veritas.
    Veeam https://www.veeam.com/kb4041
    For details, contact Veeam.

3. Redeploy IoT Edge workloads

Once the IoT Edge modules are prepared, you will need to deploy IoT Edge workloads on your target device. If you face any errors in deploying IoT Edge modules, see:

Verify data

After migration, verify that all the data has migrated and the workloads have been deployed on the target device.

Erase data, return

After the data migration is complete, erase local data and return the source device. Follow the steps in Return your Azure Stack Edge Pro device.

Next steps

Learn how to deploy IoT Edge workloads on Azure Stack Edge Pro GPU device