Jump to content United States-English
HP.com Home Products and Services Support and Drivers Solutions How to Buy
» Contact HP
More options
HP.com home
HP XC System Software: XC Installation Guide > Appendix F Description of Node Roles, Services, and the Default Configuration

Special Considerations for Modifying Node Role Assignments

» 

Technical documentation

Complete book in PDF
» Feedback
Content starts here

 » Table of Contents

 » Glossary

 » Index

This section presents information you need to consider before accepting the default role assignments made by the cluster_config utility. After reading this information, you might decide to modify the default assignments.

Appendix G describes how to modify and verify node role assignments.

General Considerations

Before deciding whether you want to accept the default node role assignments, consider the following:

  • By default, only one node is configured with the resource_management role. The resource_management role consists of one or more SLURM controller daemons and the LSF execution host. If only one node has the resource_management role, both SLURM and LSF controller daemons run on that node, and no fail over of these components is possible.

    HP recommends that you configure at least two nodes with the resource_management role to distribute the work of these components and provide a failover configuration.

  • If you expect LSF to be accessible from outside the HP XC system, all nodes with the resource_management role must also be configured with the external role and have the appropriate hardware and wiring to directly access the external network.

  • You must assign the disk_io role to any node that is exporting SAN storage.

  • By default, no nodes are configured with the login role. Assigning a login role to a node enables the LVS director service. You must assign a login role to each node on which you expect users to be able to log in and use the system.

  • By default, no nodes are configured with the nis_server role. Assigning a nis_server role to a node establishes the node as a NIS slave server. Any node assigned with the nis_server role must also have an external network connection defined.

  • You must assign an external role to any node that has an external network connection configured.

  • By default, no nodes are assigned with the cisco_hsm or voltaire_hsm roles, which are required only if your InfiniBand hardware configuration does not contain a functioning subnet manager.

    You can use a host based subnet manager from Cisco or Voltaire in certain InfiniBand switch configurations where there is no InfiniBand managed switch running a subnet manager. A functioning subnet manager is required for a functioning InfiniBand fabric.

    A typical use for a host based subnet manager is for small server blade configurations that only contain blade InfiniBand switches, which are unmanaged. In this situation, using a host based subnet manager can eliminate the cost of an extra external switch, however, additional license costs for the host based subnet managers apply reducing the cost advantage.

    Another instance where a host based subnet manager is required is in a hardware configuration with only large (> 24 port) Cisco InfiniBand switches that do not run subnet managers.

    See Section  and Section  for more information about the cisco_hsm and voltaire_hsm roles.

Special Considerations for Hardware Configurations with 63 or Fewer Nodes

Before deciding whether you want to accept the default configuration for hardware configurations with 63 or fewer nodes, consider that a compute role is assigned to the head node by default. Therefore, when users submit jobs, it is possible that the jobs run on the head node. In that situation, less than optimal performance is obtained if interactive users are also on the head node.

Consider removing the compute role from the head node to prevent it from being configured as a compute node.

Special Considerations for Hardware Configurations with 64 or More Nodes

Before deciding whether you want to accept the default configuration provided for hardware configurations with 64 or more nodes, consider that the cluster is optimized for computation by default; that is, compute nodes have no additional services on them. Therefore, consider whether you want more compute nodes overall at the expense of impact to other services on those nodes.

Special Considerations for Improved Availability

Special considerations for assigning roles for improved availability of services are documented in Table 1-2.

Printable version
Privacy statement Using this site means you accept its terms Feedback to webmaster
© 2003–2007 Hewlett-Packard Development Company, L.P.