Sitemap Home Programme
Print

Programme

 

Complete Programme

Download the final programme (as printed in the book of abstracts) as PDF:

Registered users can view the latest version of the programme (incl. last-minute changes of presentations or session chairs) in the ParCo ConfTool.


ParCo PhD Symposium

The ParCo PhD Symposium is a special forum for PhD researchers to present their work and results. Participants accepted for the PhD Symposium will receive travel support from  funds provided by industry sponsors.

The jury of the PhD Symposium congratulates Stefan Pauli, Manuel Kohler and Peter Arbenz for their paper and presentation on

"A fault tolerant implementation of Multi-Level Monte Carlo methods"

which was awarded with the ParCo2013 Best PhD Paper award. The prize includes a cutting-edge Tesla GPU sponsored by NVIDIA.

 


Mini-Symposia

Each Mini-Symposium has an organiser of its own. The organiser is responsible to decide on the application procedure, to set deadlines, and to decide whether papers presented as part of Mini-Symposia are included in the conference proceedings. Please note that abstracts and full papers have to be submitted by the regular full paper deadline (July 31) in order to be included in the book of abstracts or the proceedings, respectively. The latter applies if and only if the organiser decides that the contributions shall become part of the proceedings. To apply for a Mini-Symposium, please contact the respective organiser directly.

Below, you find a list of offered mini-symposia with a detailed description of the covered topics, organisation details, and so forth.

 


ParCo - PhD Symposium

Josef Weidendorfer (TUM)

Contact: This e-mail address is being protected from spambots. You need JavaScript enabled to view it

Submission of extended abstracts: May 17, 2013

Notification of acceptance: June 15, 2013 (postponed)

Papers published in conference proceedings: t.b.a.

As a special forum for PhD researchers to present their work and results, we will organize a PhD Symposium at ParCo 2013. Contributions are welcome on all topics relevant for ParCo: developments, applications, and future trends in parallel computing for the whole range of platforms. For a more detailed list of conference topics, see the respective ParCo website.

Participants accepted for the PhD Symposium will receive travel support (approx. 50%-90% of the conference fee) from funds provided by industry sponsors. In addition, a best paper prize is planned for the best PhD Symposium paper and presentation.

Please use the ParCo ConfTool to submit extended abstracts (2-4 pages) for the ParCo Phd Symposium. For author stylesheets and more detailed information on the conference proceedings see the respective ParCo website. Full papers for the ParCo PhD Symposium will be treated in the same way as regular papers and published in a special section of the ParCo Proceedings.

Note: for accepted papers, we will ask for a confirmation from the PhD supervisor that the PhD student is main author of the paper, and that it is based on the PhD student's research work.

 

ParCo - Symposium: Extreme Scaling on SuperMUC

Ferdinand Jamitzky (LRZ)

Contact: This e-mail address is being protected from spambots. You need JavaScript enabled to view it

Submission of extended abstracts: only accepted speakers from June workshop

Notification of acceptance: does not apply

Papers published in conference proceedings: t.b.a.

In June 2013, the Leibniz Supercomputing Centre (LRZ) organized a three day workshop on extreme scaling. The goal was the enabling of a user implemented software on SuperMUC, the 3 PetaFLOP/s machine at LRZ. Users submitted their own software as candidates for the workshop, and extreme scaling experts from IBM, Intel and LRZ were working together with the users to run the software on almost all the 147,456 cores of SuperMUC, trying to break the magic threshold of one Petaflop/s application performance. From the applicants, the most promising software packages were selected. The mini-symposium is about the lessons learned and the usage of the tools for analysis of Petascale software.

 

ParCo - Symposium: Performance Modeling and Engineering for Multi-/Many-Core Architectures

Gerhard Wellein (FAU Erlangen)

Contact: This e-mail address is being protected from spambots. You need JavaScript enabled to view it

Submission of extended abstracts: only invited speakers

Notification of acceptance: does not apply

Papers published in conference proceedings: t.b.a.

Complex, hierarchical architectures are now the standard for highly parallel computer systems. Efficient code implementation and parallelization has to consider the impact of hardware parallelism at various levels (SIMD/SIMT/SMT, socket, node, cluster) as well as multi-level memory hierarchies or heterogeneous node structures. Without a systematic understanding of the relevant performance limitations in this high-dimensional search space, any optimization attempt is just a shot in the dark.

Hence, it is vital to fully understand the performance properties and bottlenecks that come about with clustered multi-/many-core, multi-socket nodes and to develop processes or tools which allow application optimization and parallelization in a systematic way. Another aspect of modern high-end systems which attains increasing attention is the complex interplay between power constraints and the need for compute performance.

This minisymposium will discuss performance properties of modern highly parallel systems using state-of-the-art modeling and analysis techniques and real-world applications and tools.

 

ParCo - Symposium: Parallel Programming for Heterogeneous Architectures

Bettina Krammer (MoRitS, Bielefeld University & University of Applied Sciences Bielefeld), Hartmut Mix (Technische Universität Dresden), Markus Geimer (Jülich Supercomputing Centre)

Contact: bettina.krammer@uni-bielefeld.de

Submission of extended abstracts: Monday, June 24, 2013, via easychair https://www.easychair.org/conferences/?conf=parcomshetar2013); for author stylesheets and more detailed information on the conference proceedings see the respective ParCo website http://www.mac.tum.de/parco2013/infoauthors

Notification of acceptance: July 20, 2013

Papers published in conference proceedings: yes

As the demand from HPC applications in various domains requires more and more compute power, hardware accelerators such as Nvidia GPUs (Graphics Processing Units) or Intel MIC (Many Integrated Core) featuring up to thousands of processing units are attractive options for accelerating floating-point-intensive numerical algorithms. To unleash the full potential of such heterogeneous architectures, developers of HPC application codes depend on innovative and efficient programming environments. In this regard, key technologies are hybrid programming models and methods, performance measurement, prediction and correctness tools, as well as automatic code transformation to simplify the adoption of new programming techniques. The purpose of this symposium is to bring together researchers in the aforementioned topics and practitioners with diverse backgrounds developing simulation codes in order to discuss and advance the state-of-the-art in software engineering for parallel applications targeting heterogeneous HPC platforms.

   

ParCo - Symposium: Space-filling Curves in Parallel Computing

Dirk Roose (KU Leuven), Michael Bader (TUM), Tobias Weinzierl (TUM)

Contact: This e-mail address is being protected from spambots. You need JavaScript enabled to view it

Submission of extended abstracts: only invited speakers

Notification of acceptance: does not apply

Papers published in conference proceedings: t.b.a.

Space-filling curves (e.g., Hilbert or Peano curves) play an increasingly important role in parallel computing. They can be used to traverse a higher dimensional space in a locality-preserving way.  Especially in case of adaptive refined or irregular meshes,  space-filling curves produce orderings of the mesh elements with the desired characteristic that the elements that are near to each other in space are likely to be ordered near to each other as well. This ordening is beneficial for several issues in parallel computing, for example to optimize cache usage and to produce high quality mesh partitionings.

In this mini-symposium we will give an overview of some recent advances in the usage of space-filling curves in parallel computing.

     

 

ParCo - Symposium: DECI Minisymposium (PRACE - Partnership  for Advanced Computing in Europe)

Chris Johnson (The University of Edinburgh)

Contact: This e-mail address is being protected from spambots. You need JavaScript enabled to view it     

Submission of extended abstracts: only invited speakers from selected successful PRACE projects

Notification of acceptance: does not apply

Papers published in conference proceedings: one collective paper

Following on from a highly successful Mini-­‐Symposium at last year’s PARA 2012, the DECI team organises this minisymposium again under the umbrella of the ParCo’13 conference in Munich. Selected DEISA projects present their insights, results, and experiences, and a collective paper summarises these presentations later on.
      
DECI (Distributed European Computing Initiative) is a resource exchange programme which provides researchers all over Europe with access to Tier‐1-super‐computing resources. DECI first ran within the DEISA project but since mid‐2011 has operated within the PRACE environment (http://www.prace‐project.eu), so far supporting 99 projects providing approximately 300 million core hours of computing time across over 25 machines. DECI calls open at 6-­monthly intervals and successful projects are given access to resources for 1 year.

 

ParCo - Symposium: Efficient Highly Scalable Multi-level Preconditioners for Linear Systems

Matthias Bolten (U Wuppertal)

Contact: This e-mail address is being protected from spambots. You need JavaScript enabled to view it

Submission of extended abstracts: May 30, 2013 (via contact email above)

Notification of acceptance: June 10, 2013

Papers published in conference proceedings: yes

Nowadays supercomputers allow for extremely large simulations originating from applications in science and engineering. Many of these simulations are described using partial differential equations or otherwise lead to large linear systems. Most often these systems are ill-conditioned and thus iterative solvers like Krylov subspace methods require preconditioning. Efficient preconditioners often employ multi-level techniques to reduce or avoid the dependance of the performance on the system size. This includes multi-level domain decomposition and multigrid methods.

In the minisymposium the requirements of multi-level methods on large-scale parallel machines and the implications of these requirements for the mathematical foundations and the implementation of the methods will be discussed. New techniques and recent results on large-scale machines will be presented.
 

 

ParCo - Symposium: Interaction and HPC: Multi-Scale / Multi-Physics Applications

Ralf-Peter Mundani (TUM)

Contact: This e-mail address is being protected from spambots. You need JavaScript enabled to view it

Submission of extended abstracts: only invited speakers

Notification of acceptance: does not apply

Papers published in conference proceedings: t.b.a.

Due to recent advances in supercomputing, more and more application domains such as medicine or geosciences profit from high-performance computing (HPC) and come up with (new) complex and computationally intensive problems to be solved. While solving larger problems is just one objective, another one is to solve problems much faster – even in real time. The probably most prominent representative of such an approach is computational steering where a simulation back end running on a supercomputer or compute cluster is coupled to a visualisation front end for interaction. Hence, users have the possibility to manipulate certain parameters (geometry, boundary conditions etc.) in order to experience immediate feedback from a running simulation. In this mini-symposium, different aspects of interactive HPC multi-scale / multi-physics applications should be addressed, dealing with topics such as parallelisation strategies, efficient numerical algorithms, coupling interfaces, data transmission, interaction, or visualisation.

 

ParCo - Symposium: Application Autotuning for HPC (Architectures)

Siegfried Benkner (UNIVIE), Matthias Brehm, Wolfram Hesse (LRZ),
Michael Gerndt (TUM), Anna Sikora (UAB)

Contact: This e-mail address is being protected from spambots. You need JavaScript enabled to view it

Submission of extended abstracts: only invited speakers from Dagstuhl seminar

Notification of acceptance: does not apply

Papers published in conference proceedings: yes

Parallel computer systems especially for High Performance Computing (HPC) are getting increasingly complex. The growth of the supercomputer’s size and complexity increases the consumption of resources (particularly energy), leading into a more complex code optimization process. So, more challenge for code optimization is necessary for those machines. The need of optimization and tuning efforts is becoming more imperative and the tuning expertise for manual tuning of codes is scarce among developers. It can be observed, that many projects have been started recently, addressing the goal to achieve automatic tuning for better performance or to lower energy consumption of high performance computing applications, such as the FP7 Projects “European approach towards energy efficient high performance” (Mont-Blanc), “Automatic online tuning” (AutoTune) or “Dynamical Exascale Entry Platform” (Deep).


The goal of the proposed symposium is to bring together people working on performance and energy auto-tuning. The performance analysis techniques and tools based on profiling and tracing for large scale parallel systems as well as their extensions for energy measurement of applications will be covered. On the auto-tuning side, the symposium will cover self-tuning libraries, tools to automatically apply, compiler optimizations and auto-tuners for application-level parameters, as well as frameworks combining ideas of the other areas.

ParCo - Symposium: High-Dimensional Meets Parallel - Algorithms and Applications

Dirk Pflüger (U Stuttgart), Hans-Joachim Bungartz(TUM), Markus Hegland (ANU)
Universitätsstraße 38
D-70569 Stuttgart, Germany

 

Contact: This e-mail address is being protected from spambots. You need JavaScript enabled to view it

Submission of extended abstracts: May 30, 2013 (via contact email above)

Notification of acceptance: June 10, 2013

Papers published in conference proceedings: yes

 

High dimensionalities are classical problems with the need for parallelization. Due to their nature, plenty of degrees of freedom or huge amounts of samples are required for their successful treatment. Classical  approaches are of stochastic nature, tensor methods, or deal with advanced numerical discretizations such as sparse grids.

In this minisymposium we aim to discuss and present algorithms, applications and approaches to deal with the challenges arising for the parallelization of high-dimensional problems. This includes strategies for load-balancing, scalability, and resilience as well as new parallelization strategies.

 

 

ParCo - Symposium: Parallel Computing with FPGAs (ParaFPGA2013)

Erik D’Hollander, Dirk Stroobandt, Abdellah Touhafi
Ghent University
ELIS Department
St.-Pietersnieuwstraat 41
B-9000 Ghent, Belgium

Contact: This e-mail address is being protected from spambots. You need JavaScript enabled to view it

Submission of extended abstracts: June 15, 2013 (via ParaFPGA website)

Notification of acceptance: August 1, 2013

Papers published in conference proceedings: yes

Webpage: http://parafpga.elis.ugent.be

Submissions: https://www.easychair.org/account/signin.cgi?conf=parafpga2013

Field programmable gate arrays meet the demands for flexible, efficient and fast coprocessors to augment the processing power of traditional CPUs. FPGAs emerge in many areas of high performance computing, either as tailor made DSPs, embedded algorithms, systolic arrays, accelerators or application specific architectures. The massive resources of an FPGA allow to implement parallel and pipelined dataflow computations delivering many results in each clock cycle. High performance computing with FPGAs is facilitated by high-level abstractions and programming paradigms. Examples are OpenCL, smart buffers, polyhedral optimizations, domain specific IP-kernels, flexible and efficient partial reconfiguration. The ParaFPGA symposium aims to be a platform for presenting the latest developments and new directions in this field. Therefore contributions are requested in the following areas:

  • parallel computing applications with FPGAs
  • high-level synthesis languages and tools
  • adaptive and fast reconfigurable computing
  • heterogeneous FPGA-centered architectures
  • memory and streaming methodologies
  • data-reuse techniques
  • adapting multicore languages and techniques for FPGAs
  • performance evaluation
Last Updated on Sunday, 13 October 2013 15:36