Department of Informatics Technische Universität München Informatik X: Rechnertechnik und Rechnerorganisation / Parallelrechnerarchitektur Prof. Dr. Arndt Bode , Prof. Dr. Hans Michael Gerndt

ESPRIT Project 29907: NEPHEW

Goal and Motivation of this project

The NEPHEW projects intends to promote cluster computing as a stable platform for parallel computing which allows applications to be executed with both high performance and cost effictiveness. This will be shown by implementing the following three computationally intensive applications:

• Restoration of historic film material (Joanneum Research / Austria)
• Low-cost flight simulation for pilot training (ELCO Sistemas / Spain)
• PET image reconstruction (Technische Universität München / Germany)

However, not only hardware cost affect the implementation and deployment of parallel software on clusters. In addition, also the effort of implementing and maintaining the parallel application imposes significant costs. In order to also attack this problem, the Nephew projects utilizes PeakWare from the Realtime Division of Matra Systems & Information. This tool, which will be ported within Nephew to Windows NT and enhanced for the use with clusters, allows to graphically model the communication within an application and then maps the specified dependencies onto the concrete hardware platform.

In this project the Technische Universität München provides a computationally intensive application from the area of medical imaging, the reconstruction of PET images for clinical usage. This project is based on a cooperation between the Lehrstuhl für Rechnertechnik und Rechnerorganisation (LRR-TUM / the Chair for Computing Systems: Technology, Organization and Design, Parallel Computers) from the Department of Computer Science and the Nuklearmedizinische Klinik und Polyklinik (MRI-NM / the Institute for Nuclear Medicine) from the Departmant of Medicine and its university hospital Klinikum Rechts der Isar.

Tomographic imaging has become a key technology for modern medical diagnosis. Recently, technologies like positron-emission-tomography (PET) have been developed, which enable the direct measurement of function whereas traditional technologies such as computer tomography (CT) are only capable of imaging the morphological structure.

PET images are acquired by measuring the decay of radio-isotopes attached to molecules with known physiological properties such as water or sugar. Reconstructing the measured projection data is a complex task - it is mentioned in the Blue Book edition of 1996 as a 'Grand Challenge' for high performance computing.

Within the Nephew project we will implement a parallel version of the image reconstruction application adapted to the Nephew hardware and tool environment. The utilization of extremely cost effective hardware in combination with a high-performance interconnection fabric, namely SCI, will allow the usage of a new and better class of PET reconstruction algorithms in clinical routine without buying expensive massively parallel processors. The core of these reconstruction algorithms that will be used within this project have been implemented at the University of Michigan by Prof. J. Fessler and are provided through the ASPIRE library.

Example images

Clustering using SCI

One of the key components of this project and for applications in the cluster environment in general is the deployment of a high performance interconnection fabric with high bandwidth and low latency characteristics. In Nephew one of the leading technologies in this sector, the Scalable Coherent Interface (SCI), is deployed using the leading-edge products of Dolphin Interconnection Solution, also a partner within the Nephew consortium.

More background information about SCI cluster computing can be obtained from the homepage of the SMiLE project (Shared Memory in a Lan-like environment). This group, which also participates in the Nephew project, has a long tradition in cluster computing research using SCI.

Contact persons / Staff at TUM

• Dr. Wolfgang Karl (LRR-TUM), email: karlw@in.tum.de
• Frank Munz (MRI-NM), email: munz@in.tum.de
• Martin Schulz (LRR-TUM), email: schulzm@in.tum.de
• Sybille Ziegler (MRI-NM), email: tbb03ab@sunmail.lrz-muenchen.de

NEPHEW Partners

	Dolphin ICS (Norway)
	ELCO Sistemas (Spain)
	Matra Systems & Information (France)
	Joanneum Research (IIS) (Austria)
	Technische Universität München (Germany)

TUM home page	Computer Science	Medicine	LRR-TUM	Nuclear Medicine	SMiLE Project