Use Microsoft HPC Pack 2012 R2 and Azure to set up a small, hybrid high performance computing (HPC) cluster. The cluster will consist of an on-premises head node (a computer running the Windows Server operating system and HPC Pack) and some compute nodes you deploy on-demand as worker role instances in an Azure cloud service. You can then run compute jobs on the hybrid cluster.
This tutorial shows one approach, sometimes called cluster "burst to the cloud," to use scalable, on-demand compute resources in Azure to run compute-intensive applications.
This tutorial assumes no prior experience with compute clusters or HPC Pack 2012 R2. It is intended only to help you deploy a hybrid compute cluster quickly for demonstration purposes. For considerations and steps to deploy a hybrid HPC Pack cluster at greater scale in a production environment, or to use HPC Pack 2016, see the detailed guidance. For other scenarios with HPC Pack, including automated cluster deployment in Azure virtual machines, see HPC cluster options with Microsoft HPC Pack in Azure.
- Azure subscription - If you don't have an Azure subscription, you can create a free account in just a couple of minutes.
An on-premises computer running Windows Server 2012 R2 or Windows Server 2012 - This computer will be the head node of the HPC cluster. If you aren't already running Windows Server, you can download and install an evaluation version.
- The computer must be joined to an Active Directory domain. For a test scenario with a fresh installation of Windows Server, you can add the Active Directory Domain Services server role and promote the head node computer as a domain controller in a new domain forest (see the documentation for Windows Server).
- To support HPC Pack, the operating system must be installed in one of these languages: English, Japanese, or Chinese (Simplified).
- Verify that important and critical updates are installed.
HPC Pack 2012 R2 - Download the installation package for the latest version free of charge and copy the files to the head node computer or to a network location. Choose installation files in the same language as your installation of Windows Server.
If you want to use HPC Pack 2016 instead of HPC Pack 2012 R2, additional configuration is needed. See the detailed guidance.
- Domain account - This account must be configured with local Administrator permissions on the head node to install HPC Pack.
- TCP connectivity on port 443 from the head node to Azure.
Install HPC Pack on the head node
You first install Microsoft HPC Pack on your on-premises computer running Windows Server that will be the head node of the cluster.
- Log on to the head node by using a domain account that has local Administrator permissions.
- Start the HPC Pack Installation Wizard by running Setup.exe from the HPC Pack installation files.
On the HPC Pack 2012 R2 Setup screen, click New installation or add new features to an existing installation.
- On the Microsoft Software User Agreement page, click Next.
On the Select Installation Type page, click Create a new HPC cluster by creating a head node, and then click Next.
The wizard runs several pre-installation tests. Click Next on the Installation Rules page if all tests pass. Otherwise, review the information provided and make any necessary changes in your environment. Then run the tests again or if necessary start the Installation Wizard again.
On the HPC DB Configuration page, make sure Head Node is selected for all HPC databases, and then click Next.
Accept default selections on the remaining pages of the wizard. On the Install Required Components page, click Install.
After the installation completes, uncheck Start HPC Cluster Manager and then click Finish. (You will start HPC Cluster Manager in a later step.)
Prepare the Azure subscription
Use the Azure classic portal to perform the following steps with your Azure subscription. These are needed so you can later deploy Azure nodes from the on-premises head node. Detailed procedures are in the next sections.
- Upload a management certificate (needed for secure connections between the head node and the Azure services)
- Create an Azure cloud service in which Azure nodes (worker role instances) will run
Create an Azure storage account
Also make a note of your Azure subscription ID, which you will need later. Find this in the classic portal by clicking Settings > Subscriptions.
Upload the default management certificate
HPC Pack installs a self-signed certificate on the head node, called the Default Microsoft HPC Azure Management certificate, that you can upload as an Azure management certificate. This certificate is provided for testing purposes and proof-of-concept deployments.
- From the head node computer, sign in to the Azure classic portal.
- Click Settings > Management Certificates.
On the command bar, click Upload.
Browse on the head node for the file C:\Program Files\Microsoft HPC Pack 2012\Bin\hpccert.cer. Then, click the Check button.
You will see Default HPC Azure Management in the list of management certificates.
Create an Azure cloud service
For best performance, create the cloud service and the storage account (in a later step) in the same geographic region.
- In the classic portal, on the command bar, click New.
- Click Compute > Cloud Service > Quick Create.
Type a URL for the cloud service, and then click Create Cloud Service.
Create an Azure storage account
- In the classic portal, on the command bar, click New.
- Click Data Services > Storage > Quick Create.
Type a URL for the account, and then click Create Storage Account.
Configure the head node
To use HPC Cluster Manager to deploy Azure nodes and to submit jobs, first perform some required cluster configuration steps.
- On the head node, start HPC Cluster Manager. If the Select Head Node dialog box appears, click Local Computer. The Deployment To-do List appears.
Under Required deployment tasks, click Configure your network.
In the Network Configuration Wizard, select All nodes only on an enterprise network (Topology 5).
This is the simplest configuration for demonstration purposes, because the head node only needs a single network adapter to connect to Active Directory and the Internet. This tutorial does not cover cluster scenarios that require additional networks.
- Click Next to accept default values on the remaining pages of the wizard. Then, on the Review tab, click Configure to complete the network configuration.
- In the Deployment To-do List, click Provide installation credentials.
In the Installation Credentials dialog box, type the credentials of the domain account that you used to install HPC Pack. Then click OK.
HPC Pack services only use installation credentials to deploy domain-joined compute nodes. The Azure nodes you add in this tutorial are not domain-joined.
- In the Deployment To-do List, click Configure the naming of new nodes.
In the Specify Node Naming Series dialog box, accept the default naming series and click OK.
The naming series only generates names for domain-joined compute nodes. Azure worker nodes are named automatically.
- In the Deployment To-do List, click Create a node template. You will use the node template to add Azure nodes to the cluster.
In the Create Node Template Wizard, do the following:
a. On the Choose Node Template Type page, click Azure node template, and then click Next.
b. Click Next to accept the default template name.
c. On the Provide Subscription Information page, enter your Azure subscription ID (available in your Azure account information). Then, in Management certificate, click Browse and select Default HPC Azure Management. Then click Next.
d. On the Provide Service Information page, select the cloud service and the storage account that you created in a previous step. Then click Next.
e. Click Next to accept default values on the remaining pages of the wizard. Then, on the Review tab, click Create to create the node template.
By default, the Azure node template includes settings for you to start (provision) and stop the nodes manually, using HPC Cluster Manager. You can optionally configure a schedule to start and stop the Azure nodes automatically.
Add Azure nodes to the cluster
You now use the node template to add Azure nodes to the cluster. Adding the nodes to the cluster stores their configuration information so that you can start (provision) them at any time as role instances in the cloud service. Your subscription only gets charged for Azure nodes after the role instances are running in the cloud service.
For this tutorial you will add two Small nodes.
In HPC Cluster Manager, in Node Management (called Resource Management in recent versions of HPC Pack), in the Actions pane, click Add Node.
In the Add Node Wizard, on the Select Deployment Method page, click Add Windows Azure nodes, and then click Next.
On the Specify New Nodes page, select the Azure node template you created previously (called by default Default AzureNode Template). Then specify 2 nodes of size Small, and then click Next.
For details about the available sizes, see Sizes for Cloud Services.
On the Completing the Add Node Wizard page, click Finish.
Two Azure nodes, named AzureCN-0001 and AzureCN-0002, now appear in HPC Cluster Manager. Both are in the Not-Deployed state.
Start the Azure nodes
When you want to use the cluster resources in Azure, use HPC Cluster Manager to start (provision) the Azure nodes and bring them online.
In HPC Cluster Manager, in Node Management (called Resource Management in recent versions of HPC Pack), click one or both nodes and then, in the Actions pane, click Start.
In the Stop Windows Azure nodes dialog box, click Start.
The nodes transition to the Provisioning state. View the provisioning log to track the provisioning progress.
- After a few minutes, the Azure nodes finish provisioning and are in the Offline state. In this state the role instances are running but will not yet accept cluster jobs.
To confirm that the role instances are running, in the classic portal, click Cloud Services > your_cloud_service_name > Instances.
You will see two worker role instances are running in the service. HPC Pack also automatically deploys two HpcProxy instances (size Medium) to handle communication between the head node and Azure.
To bring the Azure nodes online to run cluster jobs, select the nodes, right-click, and then click Bring Online.
HPC Cluster Manager indicates that the nodes are in the Online state.
Run a command across the cluster
To check the installation, use the HPC Pack clusrun command to run a command or application on one or more cluster nodes. As a simple example, use clusrun to get the IP configuration of the Azure nodes.
- On the head node, open a command prompt.
Type the following command:
clusrun /nodes:azurecn* ipconfig
You will see output similar to the following.
Run a test job
Now submit a test job that runs on the hybrid cluster. This example is a very simple parametric sweep job (a type of intrinsically parallel computation). This example runs subtasks that add an integer to itself by using the set /a command. All the nodes in the cluster contribute to finishing the subtasks for integers from 1 to 100.
In HPC Cluster Manager, in Job Management, in the Actions pane, click New Parametric Sweep Job.
In the New Parametric Sweep Job dialog box, in Command line, type
set /a *+*(overwriting the default command line that appears). Leave default values for the remaining settings, and then click Submit to submit the job.
- When the job is finished, double-click the My Sweep Task job.
Click View Tasks, and then click a subtask to view the calculated output of that subtask.
To see which node performed the calculation for that subtask, click Allocated Nodes. (Your cluster might show a different node name.)
Stop the Azure nodes
After you try out the cluster, stop the Azure nodes to avoid unnecessary charges to your account. This stops the cloud service and removes the Azure role instances.
In HPC Cluster Manager, in Node Management (called Resource Management in recent versions of HPC Pack), select both Azure nodes. Then, in the Actions pane, click Stop.
In the Stop Windows Azure nodes dialog box, click Stop.
The nodes transition to the Stopping state. After a few minutes, HPC Cluster Manager shows that the nodes are Not-Deployed.
To confirm that the role instances are no longer running in Azure, in the classic portal, click Cloud Services > your_cloud_service_name > Instances. No instances will be deployed in the production environment.
This completes the tutorial.