Tutorials
2001 TutorialsLinux Clusters: the HPC Revolution
Last updated: 20 April 2001
| Title | Cluster-In-A-Box | |
Presenter |
Jeremy Enos | |
| Presenter Inst | NCSA | |
| Time | 8:00am - 12:00pm | |
| Description | OSCAR (Open Source Cluster Application Resource) is a cluster-in-a-box package developed to lower the entry barriers into the cluster computing world and to eliminate the need for a cluster expert. OSCAR provides a single CD with a wizard-driven interface to building a cluster. It is designed to provide the most popular and universal cluster software available in open source. At the end of the wizard process, the cluster will be ready to run MPI or PVM jobs using the Portable Batch System (PBS). C3 tools are included for management and ssh for secure transactions. The build process and future plans for the cluster-in-a-box tecnologies will be covered in the tutorial. | |
| Title | Using the TotalView Debugger | |
| Presenter | Roy Heimbach | |
| Presenter Inst | NCSA | |
| Time | 8:00am - 12:00pm | |
| Description | This half-day tutorial will demonstrate the interactive features of the Totalview debugger from Etnus. Common MPI programming errors will be reviewed. A series of short example programs will be used to demonstrate diagnosis of MPI programming errors with the help of Totalview. | |
| Title | Performance Tuning Techniques for IA32- and IA64-Based Clusters | |
| Presenter | Troy Baer | |
| Presenter Inst | Ohio Supercomputing Center | |
| Time | 1:00pm - 5:00pm | |
| Description | This half-day tutorial will demonstrate
several techniques for improving the single-processor
performance of applications on clusters of IA32-
and IA64-based systems, such as those in service
at OSC, NCSA, and AHPCC. These techniques involve
taking advantage of features common to most modern
microprocessors, including multi-level caches
and multiple pipelined functional units. The
tutorial will concentrate primarily on techniques
for IA32 based systems using the GNU and Portland
Group compilers, but will also discuss some techniques
applicable to IA64 based systems using the Intel
and SGI compilers. Topics covered in this tutorial include basics of microprocessor architectures, loop unrolling, hierarchical memory systems, avoiding cache thrashing, prefetching, inlining, high-performance numerical libraries, and performance analysis tools. |
|
| Title | Low-Cost Linux Clusters for Biomolecular Simulations Using NAMD | |
| Presenter | Jim Phillips | |
| Presenter Inst | University of Illinois at Urbana-Champaign | |
| Time | 1:00pm - 5:00pm | |
| Description | This tutorial will familiarize
the participant with the design, assembly, and
use of a small Linux cluster for a specific biomolecular
application. We will begin with a discussion
of currently available technology, the design
of a cluster for a small research group, and
cost-benefit tradeoffs. The group will be broken
into groups of mixed experience, each of which
will assemble and install a small cluster using
the Scyld Beowulf cluster operating system (http://www.scyld.com/). Our motivating application will be molecular dynamics simulations of large biomolecular systems using the program NAMD, distributed free of charge with source code (http://www.ks.uiuc.edu/Research/namd/). After an introduction to molecular dynamics and NAMD, we will experiment with NAMD performance on the just-built small clusters, a 32-processor cluster, and the large clusters at NCSA. We will conclude with a demonstration of interactive molecular dynamics, connecting the visualization program VMD, a haptic interface device, and a NAMD simulation running on a Linux cluster. We will encourage open discussion of any questions on cluster design and use during the tutorial. |

