Concepts and Terminology

From UniCluster-UC5

This section describes concepts and terms that appear throughout the UniCluster documentation. We suggest you have a clear understanding of these items before continuing.

  • CFM
CFM stands for Configuration File Management and is used to distribute configuration changes to the various node groups. Generally, this is not something that users or administrators need to interact with directly, although in some cases interaction is necessary.
  • Cluster
A cluster is a collection of nodes, or computers.
  • Component
A component is a chunk of software that runs on one machine. If a software package has a master and a compute piece, then there will be a master component and a compute component. Components must be selected individually for installation on specific node groups. More information can be found under the profile editing article.
  • Compute Network
To manage the cluster, deploy software, and so forth, the installer node must be able to reach each compute node. The compute nodes may also need to communicate with each other. In the simplest configuration, all nodes in a cluster (the installer node and the compute nodes) connect to a single network: the compute network. More complicated configurations might have a separate compute network for each node group. We strongly suggest that the compute network be a dedicated network segment, separate from any user or desktop traffic.
  • Hardware Profile
A hardware profile is a non-overlapping, logical group of nodes that are provisioned with the identical set of kernel, network configuration, disk partitioning, and device modules. A hardware profile configures all the components of a node that cannot be changed without rebooting the hardware or adding/removing devices from the hardware.
  • IPS Repository
IPS repositories manage OpenSolaris packages. OpenSolaris UniCluster manages IPS repositories for distributing software to cluster nodes.
  • Kit
A kit, in UniCluster terms, is a complete software package. If more than one chunk of software makes up a particular software package (such as SGE, which has a master piece and a compute piece), then they are all included with the kit. Software is brought down to the installer node in the form of a kit. A kit contains components (see Component), and the components are what get deployed to nodes.
  • Network Boot
See PXE Boot.
  • Node
A node is a single computer within a cluster. There are three different types of nodes:
  • Primary Installer Node
There is one primary installer node in a cluster. It holds the provisioning and configuration data for all nodes in the cluster. Its purpose is to provision and deploy software to the other nodes (compute nodes or subordinate installers). You perform all provisioning and configuration on the primary installer node, which then "pushes" these changes out to the other nodes. System administrators and users typically have direct network access to the primary installer node, which acts as a gateway between the users and the compute nodes. The primary installer node may also be referred to as the "master" or "main" node.
  • Subordinate Installer Node
There can be many subordinate installer nodes in a cluster. The subordinate installer nodes are provisioned from the primary installer node. The subordinate installer nodes communicate directly with the primary installer and allow UniCluster to scale beyond 1,000 nodes in a single cluster. The subordinate installer nodes have three main functions: provide scalability beyond 1,000 nodes, handle complex network configurations, and provision different operating systems in the same cluster. The primary installer node maintains a single central database and synchronizes the database with all subordinate installer nodes. The subordinate installer nodes maintain their own local database configuration for nodes that they provision.
  • Compute Nodes
Compute nodes are the workers in the cluster; they run the deployed applications. They are typically connected to a single, common network. Note: System administrators and users typically do not have direct access to this network. User access to a compute node is usually accomplished through the gateway provided by the installer node.
  • Public Network
System administrators and users access the cluster through the gateway provided by the primary installer node. Typically, the primary installer node has two network interfaces: one that connects to the compute network and one that connects to the public network for admin/user/desktop access.
  • PXE Boot
PXE is a standard of booting up a computer whereby the computer will use a form of DHCP (Dynamic Host Configuration Protocol) to request an initial kernel to boot from. The initial kernel is used to provision nodes in UniCluster. Note: It is sometimes necessary to change the boot order in the BIOS so that "Network Boot" or "PXE Boot" is the first item.
  • Repository
See Yum Repository and IPS Repository.
  • Resource Adapter
Resource adapters provide the interface between UniCluster and a provisioning technology. The default resource adapter in UniCluster provisions physical machines attached to an installer using PXE-boot. Other resource adapters are available for deploying VMs (virtual machines) on hypervisors or machines in a Cloud.
  • Software Profile
A Software Profile is a non-overlapping, logical group of nodes that are provisioned with identical sets of software.
  • YUM Repository
YUM stands for "Yellow dog Updater, Modified" and is the way a number of Linux distributions (including RedHat) allow users to install or update software. Univa UD hosts YUM repositories so that Linux UniCluster can connect to them and download new or updated software. Once software has been downloaded to the installer node, UniCluster 5 hosts YUM repositories internally on the installer node for the compute nodes to update against. See the http://en.wikipedia.org/wiki/Yellow_dog_Updater%2C_Modified link for more information on how YUM repositories work.

[ UniCluster Documentation ]