Features

  • A set of tools to ease the different tasks of setting up and maintaining a high performance cluster
  • A variety of Linux distributions and system configurations are supported
  • Master node set up to provide all information and services needed to run a cluster
  • Golden client to provide basic computational infrastructure
  • Install and update many compute nodes with pre-defined setup
  • Flexible node type concept

The ParaStation ClusterTools as part of the ParaStationV5 cluster suite offer a set of tools to ease the different tasks of setting up and maintaining a high performance compute cluster. It supports a variety of Linux distributions and system configurations.

Beside a minimal setup for the compute nodes and a complete setup for the master node, a couple of packages are installed by default, transforming the bunch of servers into a high performance compute cluster:

  • ParaStation MPI as a robust environment to run parallel and serial jobs within the cluster.

  • Torque/Maui for batch processing of jobs within the cluster.

  • ParaStation GridMonitor to monitor all parts and activities of the system.

  • NFS to provide system-wide file systems for user and scratch data.

  • SystemImager as a basis for system installation and updates.

Basic installation and configuration

Setting up a HPC cluster from scratch requires careful preparation - lots of packages have to be installed and configurations must be adjusted to form a homogeneous cluster. The ParaStation ClusterTools simplify these tasks to do essentially three different things:

  • Set up a master node to provide all the information and services necessary to run a cluster.

  • Set up one compute node ("golden client") to provide the basic computational infrastructure.

  • Install and update many compute nodes with a previously defined setup.

Furthermore, the ParaStation ClusterTools may be used to update compute nodes to new or modified configurations.

The entire cluster is managed from a so called master node. This node runs all the necessary services to organize all nodes into a homogeneous high performance compute cluster. The master node and the compute nodes are interconnected by an administration network, typically Gigabit Ethernet using one internal NIC. Providing NAT, the master node also acts as a gateway for the compute nodes. Usually only the master node is connected to a higher-ranking network.

On a typical HPC cluster an additional data network (MPI network) running application data will be configured. The MPI network may be based on Gigabit Ethernet, or on more advanced technologies like Infiniband.

The system installation and configuration of the compute nodes is based on images, pre-defined or set up on one or more golden clients. The images are maintained and distributed using the SystemImager suite. On top of this, the ParaStation ClusterTools defines an additional layer describing the entire cluster.