The Lyceum Linux Teaching Cluster Service

The  Linux teaching cluster, Lyceum, is a pilot service providing a powerful Linux High Performance Computing facility. It is targetted at undergraduate and MSc student projects which require substantially greater computational power or memory than is available on the Windows workstations. It is particularly designed to provide for projects where the  the number or length of runs required is impractical on individual PCs. Priority on this system will be given to teaching needs during peak periods. Research postgraduates and staff users will normally find that their research computing needs are better met on the larger Iridis research cluster, which is not available for teaching use.  

The initial pilot service has been in operation since December 2008 and we are now confident that this can be extended to cope with substantially more users. The system has been very stable so far and we are not expecting to have to make any changes that would cause significant disruption to the service. The remaining issues are mainly around fine tuning of the resource control policies which are designed to prevent domination of the system by a few users when the system is very busy. Before we move  to a full production service, we would like to investigate resource control aspects with a large active user base, and so we now encourage all interested project students to join an extended pilot service. 

Project/Course tutors who would like their students to have access to the extended pilot service on Lyceum should send a request to ServiceLine with a list of usernames. Individual students should get their tutor to submit a request to Serviceline on their behalf. 

Hardware

The Cluster is accessed via  a powerful head node with 16 processor-cores and 32 GB of RAM. This can be accessed interactively and used as a gateway to a further 16 compute nodes, each with 8 processor-cores and 16GB of memory (a cluster node is an individual computer). The individual nodes use 2.3 GHz AMD quad-core "Barcelona" processors. (The head node has four quad-core processors and the compute nodes have two.)  The peak performance of the system is ~1.3 TFlops (1.3 million million floating point operations per second!)

Operating System and Job Control

The operating system is Red Hat Enterprise Linux version 5.2. Generically Linux is an open source version of the Unix operating system. If users are not already familiar with UNIX/Linux operating systems they will need be prepared to acquire a basic knowledge. In practice this is not usually too big a problem, as with software packages such as Fluent, Ansys or Matlab, the GUI will look the same as it does under Windows.

Smaller, shorter jobs can be run on the head node in a multi-user environment. When a user needs to scale up to running larger or longer jobs, or to run several jobs at once, jobs can be submitted to run on the compute nodes via a batch queue system.  This allocates exclusive access to the node to the user for the duration of their jobs. Job scheduling policies are designed to prevent domination of the system by a small group of users. Example job templates are available for common software packages, to assist with submitting batch jobs to the compute nodes. So, often, all that is required to get started is the use of a GUI-based text editor and the use of a simple command to submit a job.