CSE09
From Autocompwiki
←Older revision | Newer revision→
Contents |
Computational Science and Engineering Workshop (CSE 2009)
8. and 9. January 2009, Seehotel Rust
Introduction
This is the first workshop where scientist from the TU Vienna, BOKU, and University of Vienna get together to discuss novel challenges in the area of Computational Science Engineering.
Organization Team
Contact Schahram Dustdar for further information.
Motivation
Various technologies have been developed for research on computational science and engineering; however, still most computational applications are written using OpenMP, MPI, Fortran, etc., and are most likely to be executed on high performance parallel machines (see e.g., report from www.prace-project.eu). From the point of view of parallel programming languages/models and parallel machines used, in our view, there has not been a lot of changes in the way that computational scientists develop their applications. It does not mean that there is no advantage in the parallelization and optimization of these applications (in fact, performance of scientific applications has been improved substantially, due to various factors).
The goal of this first Workshop on Computational Science and Engineering at the TU Wien aims at focusing more on sharing application capabilities and data since these possibilities are still not well-researched. Recent Grid, service-oriented-, and cloud computing technologies have fostered the data and computation sharing among different organizations and research disciplines. They open the door for multi-disciplinary research; it is, however, still the case that until now those scientists mostly harness computational resources for executing their applications and fewer efforts are undertaken on sharing data and application capabilities with others in Grid and cloud environments.
We would like to jointly discuss solutions that on the one hand will help computational scientists to continue optimizing their codes with their best knowledge but on the other hand also allowing them to share their computational capabilities and data with other scientists in a seamless, transparent manner enabling multiple-disciplinary research.
Background
Various computing paradigms like Cluster, Grid and more recently Cloud computing are revolutionizing scientific computing. They are not only speeding up the application execution but also facilitating the access to and management of scientific applications simplifying the work of scientists. However, the potential of different computing paradigms is usually underestimated by computational scientists. Consequently, the focus of this workshop is to:
- To classify and identify application characteristics used on our new cluster infrastructure
- To identify and discuss research issues of computational scientists and computer science engineers and identify overlaps and synergies between them
- To foster different paradigms and decide how specific applications can be improved based on a particular paradigms
- To identify and kick-off collaborations within groups where potential synergy effects are present.
Some Research Questions
- How can we speed up the application execution except by buying faster clusters?
- How can we simplify access to HPC applications (beyond SSH ) without changing the “native code”?
- How can we identify the granularity of HPC applications and fragment application to be executed on multiple machines/clusters?
- How can we improve the organization of the application in time (data flow, control flow, etc.) and in space (components, interfaces, etc.)?
- Can we reuse data products for/of a specific application as input for another run or as input for another scientific applications? How could an infrastructure for the data reuse look like?
- Where and how "reducing language dependencies" can be made for scientific applications?
- How do we abstract code of scientific applications in a way that their lifecycle (requirement analysis, code development, maintenance) in both internal view (sequential, multi-threaded, MPI, mixed OpenMP+MPI code) and external view (component view, software-as-a-service) can evolve in an optimal way?
- Can we abstract existing computational applications into "ecosystem services" (e.g., fundamental infrastructure) for computational science and engineering?
- How do we share application capabilities and data between different research disciplines?
- How do data-as-a-service and software-as-a-service concepts impact modern computational science and engineering?
- How to simplify the modeling and integration of CSE code?
- How to develop and exploit domain-specific modeling and engineering techniques for CSE?
- How to support the end-user (scientists) with simple, domain-specific programming environments?
Guidelines for presenters
Submit your presentation in PDF focusing on 3 to 5 slides as follows:
- Purpose of the research (e.g., modeling condensed matter, Quantum mechanics, device simulation,..)
- Classification of your research: (1 slide)
*Dense Linear Algebra / Sparse Linear Algebra / Spectral methods / particle methods / Structured Grid / Unstructured Grid /Map Reduce Algorithms
- Classification on Data Products: (1 slide)
*Intermediary Data / Data Size / Data management
- Research issues with the focus on sharing and integration (1 slide)
Position papers
- The “external view” and “internal view” of real scientific applications
- software-as-a-service and data-as-a-service for scientific applications
- Sharing data among different disciplinary
- Collaborative working environments for computational scientists
What we do not intend to discuss in this workshop is how to obtain maximum performance of the new cluster, since this can be discussed in other forums.
Submission
Submit your slides and optionally any supporting material in PDF to Schahram Dustdar via email to: dustdar@infosys.tuwien.ac.at until 15 December 2008 in an IEEE style with a maximum of 8 pages. The workshop papers will be published as a Technical Report.
Participants
Program
Report
To appear
