High Performance Computing Facilities
at the University of Queensland
This online documentation provides an overview of the high-performance
computing facilities available to members of the ARC Centre for Complex Systems.
Covered within is an audit of the resources available, information on how
to gain accounts on these systems, as well as information and hands-on tutorials
that describe standard practices for running large scale simulations.
This document is divided into three main sections that can be browsed using
the menu system on the left of this page:
Section 1: Systems Available and Access Procedures
This section details the various high performance computers and how to access them. In all cases, SSH (secure shell) is used as the preferred access protocol. Part of this section describes how to install and configure SSH for users unfamiliar with this method.
Section 2: Using Batch Systems
In this section, details on how to run simulations using the installed batch systems are provided. Batch systems are the preferred method of running simulations, as they are simple to use and promote fair sharing of resources between users. Batch systems are particularly useful if you are running a large number of independent simulations. This section also describes how to use the remote shell (rsh) in lieu of a batch system.
Section 3: Parallel Programming
The third section of this documentation explores the more advanced topics of socket programming and mpi (the message passing interface standard). These methods can be used to send information between machines or processors for applications that require parallel computing (such as running genetic algorithms that deal with crossing over individuals from large populations).
When to use high performance computers:
PC computers now have very fast and powerful CPUs. The High performance computing
(HPC) systems generally have multiple CPUs, large memory, fast IO(disk) to
a large storage system. Since CPU speed is generally determined by the manufacturing
technology at the time of manufacture, the HPC systems can not offer faster
single CPU execution than a recently purchased PC.
The HPC systems offer access to large memory and storage. For applications
that run well in parallel, running on 4 CPUs at speed X, may still be better
than running on a single new CPU at speed Y. However, currently, parallel
jobs need to be short (e.g., under 20 minutes on Ozone), as the batch system
(which are used for larger jobs) are generally configured only to run serial
processes.
For projects with many calculations to complete, the HPC systems provide the
opportunity to execute several at the same time, without degrading performance
(ie. each task having its own CPU). Additionally the HPC systems provide some
centrally licensed software and a platform for some packages that are not
available for PCs.