|
|
Agenda
Monday, 15 December 2014
|
12:00 - 13:00 |
Lunch and Registration
|
| 13:00 - 13:15 |
Greeting & Opening Remarks
Michael Resch, HLRS, University of Stuttgart
|
| 13:15 - 13:45 |
A SX-ACE-based New Computer System of Tohoku University and: Its Early Evaluation by using Real Applications
Hiroaki Kobayashi, Cyber Science Center, Tohoku University Sendai
Abstract
In the talk the new computer system SX-ACE of Tohoku University will be presented.
Its performance is discussed by using real applications developed by users of the Cyber Science Center.
In particular, new mechanisms are examined such as an enlarged ADB with MSHR (Miss Status Handling Registers)
and an Out-of-Order vector load/store mechanism to improve short-vector operations and indirect memory access operations.
What are the impacts on the sustained performances of the applications?
|
| 13:45 - 14:15 |
Code Optimization Activities toward Sustained Simulation Performance
Ryusuke Egawa, CyberScience Center, Tohoku University
Abstract
In my talk, First, I would like to introduce code optimization/user support activities in our center. Then, I am going to talking about our challenges to bring out these experiences to practical research, such as for achieving smooth code migrations, system developments, and a high sustained simulation performance, etc.
|
| 14:15 - 14:45 |
Aerodynamic simulation of flow through porous media based on Lattice Boltzmann Method
Jiaxing Qi, University of Siegen
Abstract
We have developed an efficient and parallel incompressible Navier-Stokes equation solver - MUSUBI,
to investigate the problem of sound generation through porous media.
An interpolation-based grid refinement technique for the Lattice Boltzmann Method has been proposed to handle this multi-scale aero acoustic problem.
To illustrate the efficiency of our code, we present its performance on the latest HLRS parallel system - Hornet.
We will present preliminary results of flow through a porous media.
|
| 14:45 - 15:05 |
Coffee Break
|
| 15:05 - 15:35 |
Xevolver: an extensible framework for user-defined code transformation
Hiroyuki Takizawa, Graduate School of Information Sciences, Tohoku University
Abstract
Due to the complexity of modern system architectures,
system-aware code optimizations are mandatory to exploit the potential of HPC systems.
As a result, an HPC application code is often specialized for a particular HPC system.
In general, substantial time and effort are required for migration of such an application to a new system,
because programmers have to reoptimize the code for the new system while removing the optimizations specific to the previous system.
To lower the migration cost, we are exploring an effective way to separate system-aware code optimizations from application codes,
and are developing a framework, Xevolver, for user-defined code transformations.
Instead of modifying a code for a particular system, Xevolver allows programmers to transform the code according to user-defined rules in external files.
Therefore, different rules can be used for different system configurations, resulting in high performance portability across system generations.
|
| 15:35 - 16:05 |
Semi-automatic segmentation and analysis of large vascular structures in CT-data
Nisarg Patel, HLRS, University of Stuttgart
Abstract
Segmentation and extraction of large vascular structures from CT-data is a particularly time consuming process for implementing in FE simulations.
Hence, a semi-automatic technique is developed trying to answer the vital process with the help of data processing filters along with CT-induced information.
Together with segmentation process, Fluid-Structure Interaction of the abdominal aortic aneurysm is presented to highlight the need of realistic vascular geometries.
|
| 16:05 - 16:35 |
ADVENTURE, an open-source CAE system and Lex-ADV, its underlying DDM-based solver library
Hiroshi Kawai, Tokyo University of Science-Suwa
Abstract
In this talk the ADVENTURE system and its application to structural analysis field are presented,
as well as the on-going project HDDMPPS for restructuring the system using a new iterative solver library called Lex-ADV
|
| 16:35 - 16:55 |
Coffee Break
|
| 16:55 - 17:25 |
Towards flexible supercomputing
Shinji Shimojo, Cybermedia Center, Osaka University
Abstract
Requirements for current supercomputing such as memory bandwidth, computational speed and scalability vary widely depending on the applications. It is difficult to design a supercomputer, which fits to all applications. Flexible computing is a one of challenging approach to that problem. Changing the configuration of hardware dynamically through attached hardware via network, you can have dynamic configuration of hardware of computers. We discuss about this challenge and the status.
|
| 17:25 - 17:55 |
Storage Systems for I/O-Intensive Applications in Computational Chemistry
Matthias Neuer, Stefan Wesner, Communication and Information Center Ulm
Abstract
Computations in quantum chemistry need a large amount of temporary storage so they benefit greatly from SSDs. However, SSD storage is expensive so we investigated different combinations of SSDs and HDDs to provide the best overall solution regarding performance and storage space for these type of applications.
|
| 19:00 - 22:00 |
Dinner at Restaurant "Goldener Adler" Stuttgart
https://maps.google.de/ coordinates: 48.762189,9.164196
|
Tuesday, 16 December 2014
|
| 09:00 |
Come together
|
| 9:15 - 9:45 |
Narratives of Simulation
Michael Resch, HLRS, University of Stuttgart
Abstract
Narratives are a way for communities to structure their experience and express their expectations and hopes.
Simulation technology has since the first days of computing been a source of a variety of expectations.
In this talk, we will explore the basic narratives of simulation and the change in narratives that come with the change in available technology.
|
| 9:45 - 10:15 |
Large Scale Simulation of a Helicopter Engine Jet and an Axial Fan
Onur Cetin, Alexej Pogorelov, Matthias Meinke, Wolfgang Schröder,
Institute of Aerodynamics, RWTH Aachen University, Germany
Abstract
|
| 10:15 - 10:45 |
Convection permitting channel simulation using the Weather Research and Forecasting (WRF) model
Thomas Schwitalla, Kirsten Warrach-Sagi, Volker Wulfmeyer, Institute of Physics and Meteorology, University of Hohenheim/Stuttgart
Abstract
Precipitation forecast is one of the most challenging tasks in numerical weather prediction. Most of the current limited area, high-resolution numerical weather prediction (NWP) models apply a domain which is centered over the region of interest. Due to boundary effects, errors can occur as e.g. the boundary relaxation zone can deteriorate synoptical features like low-pressure systems. The usual way to overcome these types of difficulties is to run a channel model over a certain latitude area. As the computational resources are to a greater or lesser extent limited, the model resolution is often very coarse to be able to run the simulation. As we got the opportunity to use the whole XC40 system, we set up the first convection permitting channel simulation using the WRF model on the northern hemisphere for a simulation period of 2 months. First results of this so far unique simulation will be presented.
|
| 10:45 - 11:15 |
Coffee Break
|
| 11:15 - 11:45 |
Enabling Engineering on Exascale
Sabine Roller, University of Siegen
Abstract
Engineering applications cover a wide field and often involve complex technical processes. Representing these problems properly is an increasingly demanding challenge with growing solver capabilities. Similar observations can be made for the resulting data that needs to be analyzed after the simulations. We present the APES suite, which offers an integrated parallel tool chain for complex simulations from pre- to post processing and enables such simulations in the field of engineering on large scale, highly distributed systems. It aims at facilitating all levels of parallelism found in modern machines, ranging from vectorization, over shared memory cores to distributed nodes across networks. Numerical methods are deployed that are suitable for complex simulation setups and for a scalable efficient computation on such large systems. Specifically covered is the Lattice Boltzmann method for incompressible flows and the Discontinuous Galerkin method for more general conservation laws. Both methods have properties, which make them especially attractive in this field.
|
| 11:45 - 12:15 |
Reaction and Combustion Modeling - A Key Technology for the development of industrial combustion equipment
Benedetto Risio, Recom, Stuttgart
Abstract
Combustion processes are used worldwide for electricity production,
district heating, waste incineration and process heating in various industries like the metals, minerals, power and process industry. The combustion systems in these industries are typically of a very large scale and the classical "trial and error" optimization of such systems is the field produces significant costs and slows down the development cycles. Reaction and Combustion Modelling has advanced in the past decade to a very powerful tool for the development and optimization of industrial combustion equipment, since it allows building thousands of different
design options in the virtual reality and to foresee and solve most of the problems early in advance, before the combustion equipment has even been built. The prerequisite for this development is the availability of a validated Modeling Software and Methodology that enables reliable predictions of the combustion and pollutant formation processes in industrial combustion systems. The high computational demands caused by the associated spatial resolution and physical/chemical model complexity required to generate reliable predictions for these systems makes the efficient of use of High-Performance Computing mandatory for the solution of these problems.
|
| 12:15 - 12:45 |
Large Scale Numerical Simulations of Planetary Interiors
Christian Hüttig, Ana-Catalina Plesa, Nicola Tosi and Doris Breue, DLR
Abstract
The slow creep of silicate rock inside the mantle of the Earth and other terrestrial planets is responsible for the heat transport from the deep interior and the large-scale dynamics, which in turn influence the formation of geological structures such as volcanoes, rifts and tectonic plates. In this study, we present high-resolution numerical simulations of planetary interiors using the mantle convection code Gaia.
|
| 12:45 - 14:00 |
Lunch
|
| 14:00 - 14:30 |
The Multicore Challenge: Discontinuous Galerkin Methods and High-Speed Turbulent Flows
Muhammed Atak, Claus-Dieter Munz, Institute of Aerodynamics and Gas Dynamics, University of Stuttgart
Abstract
Discontinuous Galerkin methods gain increasing importance within the CFD community as they combine arbitrary high order of accuracy in complex geometries with parallel efficiency. In this talk, we exploit the excellent HPC attributes of a discontinuous Galerkin spectral element method and present the DNS of a compressible supersonic turbulent boundary layer along a flat plate using the entire computational resource of HORNET.
|
| 14:30 - 15:00 |
Linear embedding and spectral structures of nonlinear operators
Uwe Küster, HLRS, University of Stuttgart
Abstract
Spektral analysis of linear problems is an important approach for understanding the dynamic behavior. It is not well known, that it is possible to embed any reasonable nonlinear operator into a linear setting delivering a spectral analysis for the nonlinear operator.
|
| 15:00 - 15:30 |
NEC SX-ACE Vector Supercomputer and Future Direction
Shintaro Momose, Shigeyuki Aino, Wolfgang Bez, NEC
Abstract
NEC gives an overview of the latest model vector supercomputer SX-ACE with its future direction in high-performance computing. NEC has launched SX-ACE by aiming at much higher sustained performance particularly for memory-intensive applications. It is designed based on the big core strategy targeting higher sustained performance. It provides both the world top-level single core performance of 64 GFlops and the world highest memory bandwidth per core of 64 GBytes/s. Four cores, memory controllers, and network controllers are integrated into a single CPU LSI, enabling the CPU performance of 256 GFlops and the memory bandwidth of 256 GBytes/s with a 64 GB memory capacity. In the target memory-intensive application areas, SX-ACE demonstrates both higher sustained performance and power efficiency, surpassing other supercomputers. NEC will also talk about the vision and concept of its future supercomputer products.
|
| 15:30 - 16:00 |
Coffee Break
|
| 16:00 - 16:30 |
Benchmarking the Worlds Most Powerful HPC Machines
Collin Glass, HLRS, University of Stuttgart
Abstract
The TOP 500 list is the most widely regarded ranking of modern supercomputers, based on Gflop/s measured for High Performance LINPACK (HPL). Ranking the most powerful supercomputers is important: Hardware producers hone their products towards maximum benchmark performance, while nations fund huge installations, aiming at a place on the pedestal. However, the relevance of HPL for real-world applications is declining rapidly, as the available compute cycles are heavily overrated. While relevant comparisons foster healthy competition, skewed comparisons foster developments aimed at distorted goals.
Thus, in recent years, discussions on introducing a new benchmark, better aligned with real-world applications and therefore the needs of real users, have increased, culminating in a highly regarded candidate: High Performance Conjugate Gradients (HPCG). Here, an in-depth analysis of HPCG and a model capable of predicting the performance on a given architecture are presented, the latter based solely on two inputs: the effective bandwidth between the main memory and the CPU and the highest occuring network latency between two compute units. Finally, conclusions regarding the required properties of a viable new benchmark for HPC machines are drawn.
|
| 16:30 - 17:00 |
High-productive OpenMP migration using Automatic Parallelizing Information
Kazuhiko Komatsu, Cyber Science Center, Tohoku University
Abstract
|
| 17:00 - 17:15 |
Closing Remarks
Michael Resch, HLRS, University of Stuttgart
|
| |
|
|
