Introduction

HPC

High Performance Computing (HPC) is the use of computer hardware and software systems to do more computing in less real-time. HPC can be achieved through a combination of advanced processor architecture, parallel processing, and the availability of large amounts of memory and disk space.

A Sun Enterprise 10000 (also called Starfire or E10K) is the centerpiece of LIONS HPC. Configured as a single domain called Helios, the Starfire is equipped with 64 400 Mhz UltraSPARC II processors and 48 gigabytes of processor accessible memory. This architecture allows a researcher to write a program that uses all 64 processors and all available memory. Several users can simultaneously run jobs on helios, and each user can specify the number of processors required for his/her job. The load-management system will start their jobs when sufficient resources are available.

In addition to Helios (the Starfire), OCCS has a Sun Enterprise 2900 called Hydra, equipped with eight dual-core UltraSPARC IV processors and 64 gigabytes of main memory, and a Sun Enterprise 3500 called Norfolk, equipped with eight UltraSPARC II processors and two gigabytes of main memory for non-parallel applications.

Recently the LIONS HPC resources has been expanded to incorpate serveral new additions. There are now 3 Sun Grid Clusters. Each cluster contains 32 Sun Fire V20z servers with 2-processors and 4 gigabytes of processor accessible memory in each.

To take full advantage of the advanced architecture of the Starfire and SUN Clusters, your application must be developed or configured to run in parallel mode. Parallel computing is the act of dividing and distributing one large computational task into many smaller and relatively independent tasks that are run simultaneously on different processors. The processor cycles required to run your algorithm does not decrease (in fact, it will probably increase), but with the work distributed over many processors, your task is completed faster, saving you time.

Parallel computing is more accessible now than ever before. In the past, standards were scarce because parallel computing methods were specific to different proprietary hardware platforms. Today, platform-independent parallel computing software standards enhance the portability of programs among different hardware platforms. LIONS HPC supports two of these parallel software development standards: MPI (Message Passing Interface) and OpenMP. To take advantage of the parallel computing architecture, you can write programs in a high level language (Fortran, C, C++) that conform to these standards. Sun compilers for the Starfire can provide parallelization of a serial (non-parallel) program through the use of compiler options that analyze your code and create equivalent parallel code during compilation. Finally, parallel applications and libraries of parallel routines are available from Sun and other third party software developers to exploit the benefits of the parallel architecture.

OCCS provides popular applications (such as the statistics packages SAS and SPSS) and high-level programming tools from Sun. Because of LIONS' distributed file system (OpenAFS), any faculty member or student with a LIONS account can run most programs from their home directory on any LIONS computational server. For example, the departmentally-licensed software packages CFX and MATLAB installed in AFS may run on the user's local LIONS client workstation, on hydra, norfolk or on helios. LIONS High Performance Computing can grow and evolve by adding new software and servers that are accessible to LIONS users without creating additional accounts and associated overhead (e.g., passwords) for each server.

Access to the computational resources of LIONS HPC is available only in the LIONS HPC environment where Sun Grid Engine (SGE) manages the requests. We recommend that users learn more about the LIONS HPC environment and SGE. Access to this environment requires only a LIONS account. New LIONS users can "get their feet wet" with the following tutorial.

If you need more details about LIONS High Performance Computing, please check the topics listed in the HPC Topics menu or contact us.