Step 5: Install and Configure HPC Pack 2008 on a Server in the Failover Cluster

Applies To: Windows HPC Server 2008

Now that you have set up a two-server failover cluster and installed SQL Server 2005 Service Pack 2 on it, you can install HPC Pack 2008 in the failover cluster. Here is a brief summary of how this differs from installing HPC Pack 2008 when there is no failover cluster. Later sections describe these differences in more detail:

• HPC Pack 2008 Setup will detect that the server is part of a failover cluster and automatically guide you through the installation of the head node. (Other roles, such as compute node, are not supported for high availability.)

• When you are prompted by Setup for which SQL Server version to use, click Use Existing SQL instance to select your existing SQL Server installation (installed with failover clustering).

• You must run Setup on each of the two servers in the failover cluster (unlike SQL Server installation on failover clustering, which is installed just once for the cluster).

• Before HPC Pack 2008 is installed on the second server in the failover cluster, HPC Cluster Manager displays the second server as “Unknown” in the Node Management navigation pane.

• After Setup is complete on one server, the Network Configuration Wizard runs, detects that this head node is a part of a failover cluster, and automatically changes the available settings to support high availability.

• Neither server in the failover cluster can serve in an additional HPC role. That is, neither server can be a compute node or a Windows Communication Foundation (WCF) broker node. Consequently, the head node remains in the offline state under normal operating mode. It cannot be brought online.

This section describes how to install and configure HPC Pack 2008 on a server in the failover cluster:

1. Run HPC Pack 2008 Setup on a server in the failover cluster

2. Configure the head node on one of the servers

Run HPC Pack 2008 Setup on a server in the failover cluster

Before you begin Setup for HPC Pack 2008, use the Failover Cluster Management snap-in to view the clustered instance that you have configured for use by the HPC cluster. You must ensure that this clustered instance is online on the server on which you will begin installing HPC Pack 2008.

To prepare to run HPC Pack 2008 Setup on a server in the failover cluster

1. To open the failover cluster snap-in, click Start, click Administrative Tools, and then click Failover Cluster Management. (If the User Account Control dialog box appears, confirm that the action it displays is what you want, and then click Continue.)

2. In the console tree, if the failover cluster that you created is not displayed, right-click Failover Cluster Management, click Manage a Cluster, and then select the cluster on this server.

3. In the console tree, expand the failover cluster, click Services and Applications, and click the clustered instance that you configured.

4. In the center pane, view the Current Owner. If this is not the server on which you choose to install HPC Pack 2008 first, in the console tree, right-click the clustered instance and then click Move this service or application to another node. When prompted, confirm the move.

5. If the clustered instance is not already Online, in the console tree, right-click the clustered instance, and then click Bring this service or application online.

Run HPC Pack 2008 Setup on a server in the failover cluster

When you run HPC Pack 2008 Setup on a server in the failover cluster, it detects that the server is configured for failover clustering and presents options as described in the following procedure.

To run HPC Pack 2008 Setup for the first time on a server in the failover cluster

1. Run HPC Pack 2008 Setup on the server from distribution media or from a network location.

2. If you see a pop-up window with “The SQL ComputeCluster database is not hosted on this node…,” then follow the instructions in Before You Begin HPC Pack Setup to move the resources to the node that you are running the installation from.

3. If you see a pop-up window that contains “A SQL ComputeCluster database must be configured with a failover cluster…,” then Setup could not find a SQL Server instance named “ComputeCluster.” You need to rename the SQL Server instance to ComputeCluster.

4. If you see a pop-up window with “This node is not an active member of its Microsoft Windows 2008 failover cluster…,” then the wizard has detected that failover clustering has been installed, but the service is not running on this server (node). Go to the Failover Cluster Management snap-in and start the Cluster service on that node.

5. If you see a pop-up window with “The SQL ComputeCluster database is not available/online…,” then the wizard has detected that failover clustering has been installed, but the database is not running on this server. Go to the Failover Cluster Management snap-in and start the database service on that server.

6. On the Getting Started page, click Next to begin installation.

7. On the Microsoft Software License Terms page, read or print the pre-release software licensing terms. Accept or reject the terms of that agreement. If you accept the license terms, click Next.

8. If you see the High Availability Head Node Installation page (which is expected), click Next.

   If you do not see this page, then Setup has not detected that the server on which you are installing HPC Pack 2008 is a part of a failover cluster, and you must confirm that the process of creating the failover cluster succeeded.

9. On the Select Installation Location page, accept the default locations or choose other locations for the HPC Pack 2008 files. Click Next.

10. On the Use Microsoft Update to help keep your computer secure and up-to-date page, select whether to use Microsoft Update, and then click Next.

11. On the Install Required Components page, review the list of prerequisite software. Software that has been installed has a green check icon next to it. Unchecked items on this list need to be installed before installing HPC Pack 2008. Click Install.

12. When the Installation Complete wizard page appears, select Start the cluster manager console, and then click Finish. If the installation was not successful, review your configuration.

13. If the setup was successful, when you open the Failover Cluster Management snap-in, in the clustered instance under Services and Applications, new resources and file shares will be visible. The resources represent services needed by the HPC cluster, and the shares provide necessary access to the Spooler directory and for remote installations to take place.

    

Configure the head node on one of the servers

HPC Cluster Manager starts with a view of the To-do List. The Windows HPC Server 2008 Getting Started Guide (https://go.microsoft.com/fwlink/?LinkId=121228) describes how to use the To-do List to configure necessary settings on your HPC cluster. It also describes other fundamental tasks that you will likely want to perform. For Windows HPC Server 2008 running on a failover cluster, there are a few differences in the To-do List that are outlined here. For the full description, see the Getting Started Guide for Windows HPC Server 2008.

Run the HPC Pack Network Configuration Wizard

The first step in the To-do List is to run the Network Configuration Wizard, which defines the physical network topology for the cluster and configures network services such as Dynamic Host Configuration Protocol (DHCP), Routing and Remote Access service (RRAS), and the firewall. In the context of a failover cluster, the Network Configuration Wizard needs to be run only once. The following sections summarize the changes to each part of the Network Configuration Wizard when it is run in the context of a failover cluster.

Identifying physical networks

The Network Configuration Wizard prompts you to identity which networks will be the Enterprise, Application, and Private networks. You identify each network from a list of network adapters that were discovered on the head node. After the networks are identified, they do not need to be redefined for the other server in the failover cluster. In fact, if you run the Network Configuration Wizard on the other server in the failover cluster (not required), the network adapters on the server that is currently acting as the head node will still be used to identify the network.

Configuring DHCP

In the context of a failover cluster, some of the options usually offered by the Network Configuration Wizard in HPC Pack 2008 are disabled. This is because the failover cluster requires IP addresses to be defined already, before the HPC cluster is installed and configured.

If you are using the configuration recommended in this guide (which does not include creating a clustered instance of DHCP in the failover cluster), then you can enter network settings, and the wizard will configure DHCP services on both servers in the failover cluster. When the wizard configures DHCP services, it uses exclusions to allocate one half of the scope to each service.

However, if the wizard detects that the failover cluster has a clustered instance of DHCP—which is not recommended—the wizard does not allow DHCP settings to be changed within the wizard (the DHCP, Domain Name System (DNS), and Gateway settings are displayed, but cannot be changed).

• Enabling network address translation: Several of the cluster topologies isolate the compute nodes from the public network, with all traffic to and from the public network passing through the head node. If you use one of these topologies, during Setup for HPC Pack 2008, the Network Configuration Wizard enables network address translation (NAT) while configuring the head node, unless you have another server on the private or MPI networks that provide NAT or DHCP services.

  Enabling NAT on the head node enables the compute nodes to access resources on the public network. In the context of a failover cluster, NAT works correctly regardless of which server in the failover cluster is functioning as the head node at any given time. If you choose a cluster topology that connects the public network to each compute node, then NAT on the head node is not required.

• Configuring the firewall: The firewall configuration is the same in the context of a failover cluster as it is on a single server. It is summarized here for completeness. When configuring the head node, the firewall is enabled or disabled for each of the networks in the topology. When configuring the firewall, you have several options:

  • Accept the default firewall configuration for Windows HPC Server 2008: ON for the public network and OFF for the private network.

  • Leave the current firewall settings unchanged if the firewall settings are configured or controlled externally (for instance, by applied group policies).

  • Configure the firewall for each of the cluster networks to meet your security and policy requirements.

  By default, ports are open for all Windows HPC Server 2008 management services. If you have applications that require access to the cluster on certain ports, you must open those ports in the firewall.

  Enabling a firewall might adversely impact the performance of IP over InfiniBand, but it will have minimal impact on performance when Winsock Direct or NetworkDirect is enabled because enabling the firewall has no effect on remote direct memory access (RDMA) packets.

Use the following procedure to help you perform the first task in the To-do List, which is Configure your network.

To configure the cluster networking topology

1. In HPC Cluster Manager, in the To-do List, click Configure your network.

2. From the wizard navigation pane, select Network Topology Selection if it is not already selected. As described in “Network infrastructure requirements” in Overview and Requirements for Windows HPC Server 2008 in a Failover Cluster, you should have already decided on one of the four supported compute cluster topologies before setting up Windows HPC Server 2008 in the context of a failover cluster. The physical network that you used to connect the HPC cluster nodes must match the configuration you select. If it does not, a warning will be displayed, and the wizard will not continue the installation process until you choose a topology that is compatible with the physical network of your cluster. After choosing a topology, click Next.

3. On the Enterprise Network Adapter Selection page, for Network adapter, select the network adapter that you want to bind to the public network.

4. On the Private Network Adapter Selection page (if shown), for Network adapter, select the network adapter that you want to bind to the private network.

5. On the Private Network Configuration page, some settings can be changed, and some are displayed but cannot be changed:

   • The static IP address for the head node on the private network is displayed but cannot be changed.

   • If you are using the configuration that is recommended in this guide (which does not include creating a clustered instance of DHCP in the failover cluster), the wizard offers defaults for these settings and allows you to change them. The first address after the address that is used by the head node is the starting address. The wizard will create exclusions for the DHCP scope so that DHCP on each server in the failover cluster will control half of the scope.

     However, if the wizard detects that the failover cluster has a clustered instance of DHCP—which is not recommended—the wizard does not allow DHCP settings to be changed within the wizard (the DHCP, DNS, and Gateway settings are displayed, but cannot be changed).

   • The NAT configurations that are suggested by the Network Configuration Wizard can be changed or accepted.

   After configuring the settings, click Next.

6. On the MPI Network Interface Binding page (if shown), for Network adapter, select the network adapter that you want to bind to the high-speed (MPI) network.

7. On the Application Network Configuration page (if shown), either accept the DHCP and NAT configurations that are suggested by the Network Configuration Wizard, or change the settings. Then click Next.

8. On the Firewall Setup page, choose whether to automatically apply firewall settings to the enterprise network, the private network, or both. (We recommend choosing this option and applying firewall settings to at least the enterprise network.) Alternatively, select Do not manage firewall settings.

9. When the Review page appears, verify all settings.

   1. If you are satisfied with the values that you have entered, click Configure.

   2. If you want to change any setting, in the left pane, select the corresponding wizard page. After configuring settings and returning to the Review page, click Configure to apply your settings.

10. The Configuration Summary page displays a summary of the configuration values that you have entered. Click Finish to close the wizard.

Complete the remaining required tasks in the To-do List. For more information, see the Windows HPC Server 2008 Getting Started Guide (https://go.microsoft.com/fwlink/?LinkId=121228).

After all the required To-do List tasks are completed, all the navigation panes are accessible. In Node Management, you will see two head nodes. The head node you just configured will be in the “offline” state, which only indicates that no jobs can be run on it (in a failover cluster, it cannot act as a compute node, that is, it cannot run jobs). The second server will display as a head node in the Unknown state, because HPC Pack 2008 is not yet installed on it.