This is an archived page of the 2000 conference


Linux Clusters: the HPC Revolution

Title IA32 Architecture and IA64 Architecture
Presenter Sunil Saxena
Presenter Inst Intel
Time 8:30am - 12:00pm
Description This tutorial will provide an overview of the IA-32 / IA-64 architecture and performance. We will discuss the new processor features of IA-32 architecture. We will go into detailed architecture and micro-architecture overview for IA-64. We will review examples of how one can write good code and how to tune code for best performance.
Title The Linux Kernel and Internals for Scientific Cluster Users
Presenter Pete Beckman and Trammi Hudson
Presenter Inst TurboLinux, Inc.
Time 1:30am - 5:00pm
Description Clusters and Linux are hot news. It's similar to every other Unix, although migrating from other UN*X platforms to Linux does require some adjustment of ideology, philosophy and file locations. This tutorial will give an overview of the Linux kernel and Linux cluster computing issues relevant to scientific users. We will describe the kernel internals, the boot process and network setup for common distributions, high-speed networking, and cluster programming issues. A discussion of present and future filesystems in the Linux kernel will be followed by pontificating and prognostication on the Open Source community and future directions of Linux clusters, the kernel, and community.
Title Effective Compiler Utilization
Presenter Eric Stoltz
Presenter Inst The Portland Group
Time 8:30pm - 9:45pm
Description This tutorial will provide methods for tuning scientific codes utilizing parallelization techniques with the PGI compiler. These techniques include OpenMP, automatic parallelization, MPI, and vector registers on current superscalar architectures. Other general methods for increasing code performance will be covered as time permits.
Title DEEP Tutorial: Parallel Performance Profiling at the Source Code Level
Presenter David McNamara
Presenter Inst Pacific-Sierra Research
Time 10:00am - 12:00pm
Description EEP/MPI is a development tool for analyzing MPI programs. DEEP/MPI consists of profilers which instrument C and Fortran programs at the source level and a sophisticated GUI which displays useful information about the MPI program. The goal is to help programmers quickly identify performance bottlenecks in their parallel source code. Beyond MPI information, DEEP also provides performance counter information such as cache misses at the loop level by making use of the PerfAPI interface. In this tutorial, effective ways of using DEEP/MPI to improve application performance will be discussed. A step-by-step approach to improving an MPI program will be demonstrated.
Title TotalView Debugger
Presenter Luiz DeRose
Presenter Inst IBM, Inc
Time 1:30pm - 3:15pm
Description The TotalView Multiprocess Debugger provides debugging and analysis capabilities for Linux, AIX, and other Unix platforms. This tutorial will show you how to
  • get started using TotalView,
  • access language specific features in F90 and perhaps C++
  • use the visualizer to graphically view data in your program
  • debug parallel programs
The tutorial will focus on helping you find the most important features within TotalView, such that you can further explore new and additional features on your own.
Title Vampir MPI Performance Profiling
Presenter Werner Krotz-Vogel
Presenter Inst Pallas
Time 3:45am - 5:30pm
Description Vampir, the leading MPI performance analysis tools, is now available in a new and improved version. Vampir features a streamlined user-interface, additional displays and a source-code display, while keeping all the unique features of previous Vampir releases.

This presentation will cover
  • brief introduction to Pallas, a leading european vendor of software tools for parallel computing.
  • Vampir 2.5, visualization and analysis of MPI programs
  • Vampirtrace 2.0, low overhead MPI profiling library
  • an on-line demonstration showing how to start using Vampir and how to use it efficiently in the loop of parallel software development and performance tuning.