Austrian Grid 2: Distributed Supercomputing in the Grid [AGRID2]

@mastersthesis{RISC3854,
author = {Amira Zaki},
title = {{Re-engineering of a Grid Aware Medical Database System Based on a Metamodel}},
language = {english},
abstract = {Over the last few decades, the amount of medical data available through the advancementsin medical and clinical researches has greatly increased. Therefore, the need for huge storagecapacities, eective data retrieval methods and high performance computations has becomesignicantly important. The aim of this work was to develop a simplied metamodel forSEE++, a software system for the computer-aided simulation of eye motility disorders andtheir treatments. The poor performance of the current over-engineered metamodel was thedriving force behind the necessary enhancements. Thus, the newly developed metamodelallows the integration into a Grid environment to achieve near real-time performance forthe numerous simulations and complex calculations performed. This was done using aninstantiation of YAMM, a tool for creating web-based database applications based on ametamodel, comprising both the data and model conjointly in a relational back-end. Ourwork involved restructuring YAMM to parse its database interactions into a generic formto query any resource. An interoperability layer was created to handle various databaseconnections. The use of a relational back-end was implemented and an interface was designedfor using any other system like a Grid-aware database. Moreover, the methodologyof instantiating YAMM to represent the SEE++ data model was investigated, resulting inthe discovery of a few problems in YAMM's expressiveness, which introduced ambiguity inthe SEE++ data model representation.},
year = {2009},
month = {July},
translation = {0},
school = { Internationaler Universitätslehrgang Informatics: Engineering & Management (ISI Hagenberg)},
length = {65}
}

@mastersthesis{RISC3458,
author = {Imre Zoltan Matko},
title = {{Grid-aware Database Support for Medical Software}},
language = {english},
abstract = {SEE-GRID is a research project based on the SEE++ software system for the biomechanical simulation of the human eye. The project aims to develop a Grid-enabled version ofSEE++ to support ophthalmologists in the treatment of common eye motility disorders.Currently there exist Grid-based implementations of computation and calculation services required by SEE++ and a prototype of a metamodel based medical database. The current database implementation has some restrictions, performance problems and its Grid integration is unfinished. The goal of this thesis is to analyze the software and find performance bottlenecks and to extend the design with a Grid interface and provide a prototype implementation.Based on the benchmarking and profiling of the SEE-GRID database components we realized that the current metamodel is too complex, causing to spend most of the runtime with database queries and data transformations. Some feasible solutions we proposed are the simplification of the metamodel or the usage of fast, XML based back-ends.We have evaluated state of the art Grid data-resource management tools and found OGSA-DAI and AMGA two promising solutions for our project. We have extended the design of the SEE-GRID database with a Globus “Web Service Resource Framework” based interface and with the services of OGSA-DAI. To prove the applicability of the design we prototyped the Grid interface and evaluated in practice OGSA-DAI with respect to the needs of SEE-GRID.iv},
year = {2008},
month = {July},
translation = {0},
school = {Internationaler Universitätslehrgang Informatics: Engineering & Management (ISI Hagenberg)},
length = {69}
}

@mastersthesis{RISC3459,
author = {Kenji Miyamoto},
title = {{Parallel Algorithms for Sparse Matrices in an Industrial Optimization Software}},
language = {english},
abstract = {Optimization problems have an important role in industry, and the finite elementmethod is a popular solution to solve optimization problems numerically. The finiteelement method rely on linear algebra, high performance linear equation solvera arethere important topic in applied science.In this thesis, we study the high performance forward/backward substitution methodby means of parallel computing.We implement various solutions and benchmark each ofthem in detail on two computers with different hardware architectures; one is a sharedmemory multicore machine, and the other one is a multicore machine with virtual sharedmemory machine. We tried two programming models, MPI and POSIX threading, andfind the difference of these programming models by run benchmarks on two computers.We achieve good speedup by integrating some solutions.},
year = {2008},
month = {July},
translation = {0},
school = {Internationaler Universitätslehrgang Informatics: Engineering & Management (ISI Hagenberg)},
length = {92}
}

@mastersthesis{RISC3444,
author = {Johannes Watzl},
title = {{Investigations on Improving the SEE-GRID Optimization Algorithm}},
language = {english},
abstract = {In this work various ways to accelerate the computation process in the SEEKID/SEE-GRID software for the biomechanical simulation of the human eyeare investigated. Both sequential and parallel strategies are discussed. Toimprove the sequential optimizer, the Broyden update method is utilized. Astrategy to interpolate function values is first found on the basis of the Delaunayalgorithm. An enhanced strategy was developed based on the specialregularity of the mesh used in the interpolation program. The strategies wereimplemented and the results of the benchmarks are presented together withthe source code.},
year = {2008},
month = {June},
translation = {0},
school = {Research Institute for Symbolic Computation (RISC), Johannes Kepler University, Linz, Austria},
length = {115}
}