Series in Microelectronics Volume 162
Josef Nemecek
Computer-Aided Management of
Commodity Parts-Based Supercomputers
S Hartung-Gorre
Series in Microelectronics
edited by Wolfgang Fichtner, Qiuting Huang, Heinz Jäckel, Hans Melchior, George
S. Moschytz and Gerhard Tröster
Assuming the current trends in supercomputing
continue, the supercomputers of the future will be large clusters, containing
millions of off-the-shelf workstations. These «superclusters» will use
supercomputing-specific technology to make the pile of hardware and software
work together as one supercomputer, even though all products were originally
designed to be used individually and not as a part of a supercomputing system.
This dissertation presents one technology that is necessary in order to be able
to use the supercluster as one entity: Comprehensive and integrated management
software. It is the first complete research in its field.
This thesis presents a concept for
managing superclusters of various sizes, and some blueprints of management
architectures, together with a guide that allows the selection of the optimal
architecture depending on system size and user requirements. One of the
presented architectures has been implemented in the first Swiss-based
supercluster «Swiss-T1», installed at the supercomputing center CAPA of the
EPFL. This first implementation (called «COSMOS») is presented in detail,
together with the project «Swiss-Tx» that this thesis was part of. The goal of
this project was to build, develop and install a series of superclusters with 1
TFLOPS performance in Switzerland.
About the author
Josef Nemecek holds a Dipl. Ing. (M.S.) degree in computer
sciences from the Swiss Federal Institute of Technology in Zurich (ETH Zurich),
Switzerland. He joined the Electronics Laboratory of the ETH in spring 1997 as
a research and teaching assistant and has participated in several
supercomputing projects. His main work is the specification, design and
implementation of software that allows for integrated and comprehensive system
management of commodity parts-based supercomputers. In 2005, he was awarded the
Dr. sc. techn. (Ph.D.) degree by the ETH Zurich.
2006 S Hartung-Gorre Verlag • Konstanz
ISSN 0936-5362 / ISBN 3-86628-052-1 / 246 pages, EUR 64,–
Keywords: Commodity, Supercomputing, Cluster Computing,
System Management, Integration, Scalability, Availability.
Direkt bestellen bei / to order directly from: Hartung.Gorre@t-online.de