|
Allan Christie
Partitioning a machine provides a mechanism to divide a single Origin 2000 into multiple distinct hosts, each with their own resources. This paper provides a brief overview of the hardware and software support for partitioning, along with an overview of valid partitioned configurations and a description of how to partition a system.
Jef Dawson
The Silicon Graphics HPC computing environment will be advancing very rapidly in the next few years. This presentation will focus on how the Silicon Graphics application environment will evolve and transition. The emphasis will be on solution areas (e.g., crash, external CFD) rather than on specific applications (e.g. MSC/NASTRAN, FLUENT).
Neal Gaarder and Monika ten Bruggencate
This paper discusses the implementation of the new OpenMP standard for Fortran shared memory parallel processing in Programming Environment Release 3.1 for Cray PVP systems. Similarities and differences between OpenMP and Autotasking compiler directives, environment variables and other control mechanisms are discussed. It becomes clear that OpenMP provides equivalents for all of the essential features of current PVP Autotasking and Microtasking. The paper then discusses where users should be careful when converting Autotasked or Microtasked PVP codes to OpenMP and provides guidelines for the conversion.
Henry Gabb, R. Phillip Bording, S.W. Bova, and Clay P. Breshears
Pthreads is a POSIX standard established to control the spawning, execution, and termination of multiple threads within a single process. Because of a much lower system overhead, use of Pthreads is an attractive approach. Under this programming paradigm on a shared-memory system, threads execute concurrently within a single address space, although multiple processors may be employed to execute the various threads. An obstacle to scientific programming with Pthreads is that no Fortran interface is defined as part of the POSIX standard. We present our current progress in defining and implementing a complete Fortran 90 interface to the Pthreads library. Also presented are many of the design decisions made and lessons learned while striving to keep the bindings as portable as possible. Initial timing results indicate that the per-processor performance may be slightly less than with compiler directives, but that the scalability is superior. In addition, combining Pthreads with MPI under Fortran is discussed. This shows promise to become a useful programming model for clusters of symmetric multiprocessors.
Miltos Grammatikakis, Mr. Helidon Dollani, and Stefan Liesche
We consider algorithms for implementing mutual exclusion on the Cray-T3E virtual shared memory using various atomic operations. Our implementations of Anderson's and MCS Lock minimize network contention and dramatically improve performance for any system with more than two processors. Improvements over the Cray shmem_lock library functions are above three orders of magnitude on a 64-processor T3E-900. Our results hold for either small or large critical sections, and make the possibility of implementing concurrent data structures on the Cray-T3E virtual shared memory a viable one.
Ramesh Menon
OpenMP is an application program interface (API) for shared-memory parallel programming. Pioneered by SGI, it is fast becoming a de facto industry standard, as evidenced by the large number of hardware and software vendors endorsing the standard. The functionality is designed to enable programmers to write coarse grain, scaleable, shared-memory parallel programs. This talk will present the why, what, and how of OpenMP.
Colin Ngam
This paper provides an overview of the design and architecture of Silicon Graphics next generation disk Logical Volume Manager(XVM). XVM is an extension of XLV, which is the current generation of SGI's disk Logical Volume Manager. This paper also addresses any differences between XVM and UNICOS/UNICOSmk disk Logical Device Drivers.
Michael Resch, Thomas Beisel, Holger Berger, Katrin Bidmon, Edgar Gabriel, Rainer Keller, and Dirk Rantzau
This paper presents an environment which enables the coupling of different supercomputers to overcome the limitations of a single computing system. This requires an extension to MPI, since MPI provides no interoperability-features. A library called PACX-MPI is presented which provides the user with a distributed MPI environment with most of the important functionality of standard MPI. First results were achieved coupling the Cray T3E at the Pittsburgh Supercomputing Center and a Cray T3E at the High Performance Computing Center in Stuttgart. More recently the coupling of two Cray T3E at the High Performance Computing Center in Juelich and one Cray T3E of Stuttgart succeded. The first application we have run on the coupled T3E's is the flow simulation package URANUS. Problems and strategies with respect to metacomputing are discussed briefly for this code. Another application using this library isa Molecular Dynamics Code called P3T-DSMC. This program was developed for general particle tracking problems. We will present in this paper results for both applications.
Barry Sharp
Many SGI/Cray sites do not have the luxury of having multiple like systems with one isolated for validating new OS releases prior to installing them as their production system. The difficulty of doing this on a single system without endangering the running production work is explored along with how Boeing has developed its processes over time to safeguard production while still being able to adequately QA new OS releases.
Table of Contents | Author Index | CUG Home Page | Home (Title Page)
|