All of the IBM supplied compilers produce unoptimized code by default. Specific optimization command line options must be supplied to the compilers in order for them to produce optimized code. In this talk, several of the more useful optimization options for the IBM Fortran, C, and C++ compilers are described and recommendations will be given on which of them are most useful.
It is clear that we gain a great savings by analyzing and tuning our codes. To be effective in our efforts to improve performance, we need a versatile, robust, and portable set of performance tools. The NERSC/LBNL-led Performance Evaluation Research Center (PERC) SciDAC project is developing an interoperable suite of measurement, analysis, and tuning tools to respond to this need. The PERC tool suite features an infrastructure of low-level APIs, and a set of end-user tools built on top of those APIs. In this talk we discuss the PERC tool infrastructure and take an in-depth look at the SvPablo tool for performance instrumentation and browsing. We will also look at the future roadmap for the PERC tools and at how the PERC tool suite will affect the the future of performance analysis and tuning.
Totalview is an essential tool for debugging codes. This talk will focus on the usage of totalview with FORTRAN and C/C++. Many examples will be provided to show how to use this tool effectively. Examples using MPI, OpneMp will be provided.
Computer security is a vast topic, however this talk will give an overview of the inherent threat level currently on the Internet, how computer security is done at NERSC and techniques you can employ to ensure your computing experience at NERSC is a safe and secure one.
After the decommissioning of the Crays, what will become of the all the data stored in Cray unformatted files? In this talk we will review some of the options available in Cray's Flexible File I/O library that enable Cray files and data to be converted to formats usable on the IBM SP. Also, there will be an overview of the NCARU library, available on the SP, which has similar capabilities.
Problems unique to large concurrency MPI applications are presented along with means for addressing the problems. An overall approach to dealing with scaling issues is presented along with a set of frequently encountered problems.
The ACTS (Advanced Computational Testing and Simulation) Toolkit is a set of computational tools developed primarily at DOE laboratories and is aimed at simplifying the solution of common and important computational problems on parallel machines. The tools fall into four broad categories: numerical, and support for code development, execution and library development. In the talk we will describe the tools currently installed on NERSC platforms and the levels of support provided to users.
As a follow-up for the report on the ACTS Toolkit, in this talk we will show examples of how ACTS tools have been used on NERSC platforms and discuss lessons that we have learned.
NERSC “Visualization Greenbook”: Future Visualization Needs of the DOE Computational Science Community Hosted at NERSC
This report presents the findings and recommendations that emerged from a one-day workshop held at Lawrence Berkeley National Laboratory (LBNL) on June 5, 2002, in conjunction with the NERSC User Group (NUG) Meeting.