Technische Universität München, Fakultät für Informatik
Lehreinheit für Rechnertechnik und Rechnerorganisation / Parallelrechnerarchitektur

Martin Schulz: Publications

Evaluating a Novel Hybrid--DSM Framework in Real--World Scenarios Using Medical Imaging Applications

Martin Schulz
Parallel-Algorithmen, -Rechnerstrukturen und -Systemsoftware (PARS), München, Germany (ISSN 0177-0454) , October 2001

Abstract:

Cluster architectures with NUMA interconnection fabrics provide a good tradeoff between scalability and hardware support for shared memory programming, while maintaining the cost benefits of commodity clusters. However, a novel type of Hybrid-DSM system is required to directly provide the hardware support to user applications. Such a DSM system, embedded into a general, multi-programming model framework called HAMSTER, is presented and evaluated in this work. In contrast to earlier work in this area, which focused on simulation and/or small benchmark kernels, this evaluation is done using real-world applications from the area of medical imaging working on large data sets. This provides a realistic evaluation environment allowing a comprehensive assessment of the chosen approach.

Download:

• Compressed Postscript

Back to my Publications/Talks/Conferences.

Back to my homepage.