JUSTGrid: a pure Java HPCC grid architecture for multi-physics solvers using complex geometries

After the Earth Simulator, built by NEC at the Japan Marine Science and Technology Centre (JAMSTEC) on an area of 3,250 m2 (50mx65m), began it's work in March 2002 with the outstanding performance of 35,860 Gflops (40 TFIops peak) [TRIOO], numerous scientists opted in favour of such a high-performance computation and communications (HPCC) approach, suggesting to build again Cray type vector supercomputers that dominated scientific computing in the mid seventies. Today (2009) the extended Earth Simulator has a peak performance of 131 TFIops but it was outperformed by several other systems with multi-core 1 architectures. Top 1 in June 2009 is the RoadRunner build by IBM for the DOE/NNSA/LANL with a peak performance of 1456 TFIops. Multi-core processors are now build in every PC for the consumer market and not only for HPC systems. It should be remembered that the computer games industry is responsible for the revolution in high end 3D graphics cards that convert any PC into a most powerful graphics workstation. It should be obvious, despite the computational power of the Earth Simulator, that this definitely is not the road of HPCC for general scientific and engineering computation. "I hope to concentrate my attention on my research rather then how to program", says Hitoshi Sakagami, a researcher at Japan's Himeji Institute of Technology and a Gordon Bell Prize finalist for work using the Earth Simulator [TRIOO]. I fully agree with this statement, and this is one of the major reasons that I have chosen Java as high performance computing language. Programming vector computers is a difficult task, and to obtain acceptable results with regard to announced peak performance has been notoriously cumbersome. On the other hand, multi-core systems with many processors on a single chip need to be programmed in a different, namely a multi threaded way. Threads are a substantial part of the Java programming language. Java is the only general programming language that does not need external libraries for parallel programming, because everything needed is built into the language. In addition, there are major additional advantages of the Java language (object oriented, parallelization, readability, maintainability, programmer productivity, platform independence, code safety and reliability, database connectivity, internet capability, multimedia capability, GUI (graphics user interfaces), 3D graphics (Java 3D) and portability etc.) which were discussed in this thesis. The objective of this work is to build an easy to use software framework for high performance computing dealing with complex 3D geometries. The framework should also take care of all the advantages and behaviours of modern multi-core/multi-threaded hardware architectures. In view of the increasing complexity of modern hardware, working on solutions of multi-physical problems demands for software, that makes the solving process mostly independent of the available machinery.

Name dc:type.qualificationname

phd

Level dc:type.qualificationlevel

doctoral

Grantor dc:publisher.institution

University of Greenwich

Year dc:date.issued

2009

Author dc:creator

• Ludewig, Thorsten

Advisors dc:contributor.advisor

• Hauser, Jochem
• Patel, Mayur

• QA Mathematics

Language dc:language

en