LNAWENR
From HP-SEE Wiki
(→Activity report) |
|||
| Line 39: | Line 39: | ||
== Activity report == | == Activity report == | ||
| + | |||
| + | 1. Scientific results and impact | ||
| + | During this reporting period, we developed the pipeline in the area of mode ID and we integrated it with another stellar modelling pipeline for a complete automation of the procedure. | ||
| + | • We completed the treatment of surface effects and of the physical processes (like convection, core overshooting, diffusion of chemical elements, etc.) through successful application of the prototype pipeline to space-based data on stars observed by CoRoT and Kepler missions. | ||
| + | • We extended the fitting method to include rotation velocity as an additional parameter and validated it with data from CoRoT and Kepler missions and application of the prototype pipeline for asteroseismic data interpretation for the Kepler mission. | ||
| + | • A hybrid parallel implementation of the LNAWENR application was produced, using MPI for data-parallel distribution of the workload across various computing nodes in the NCIT-Cluster, and OpenMP for the compute intensive kernel of the application. | ||
| + | • A C version of the original Fortran code was developed and initial steps toward an OpenCL and CUDA versions of the code have been done. | ||
| + | |||
| + | 2. Scientific collaborations served | ||
| + | The data obtained from this project will be used by our team in the interpretation of data from other dedicated Space Mission like MOST, CoRoT, and Kepler. | ||
| + | |||
| + | 3. Social impact | ||
| + | Since the notoriety of the Corot and Kepler missions is so big, the social impact of the developments obtained in this project are significant. The finding of one extra solar system planet is a very important issue in the world of astronomy, and also for the public at large. | ||
| + | |||
| + | 4. Technical results | ||
| + | The current implementation of the LNAWENR application, one of the eight non-adiabatic models in use now by the international scientific community was parallelized and optimized.. The work was focused on both improving the initial, sequential version, of the implementation, and on implementing parallel versions using the MPI and OpenMP parallel programming paradigms. The results are quite promising. The MPI version scales with the number of independent components that are analyzed, while the best performance gain obtained so far for the shared-memory kernel is that the current implementation runs 5 times faster than the initial one, using OpenMP threads, on four CPU cores. | ||
| + | The work on this project is far from finished, since some further optimizations are already in the process of implementation. We will have to further rewrite the code so that some labels that are still inside the main inner loop of the program, will be moved in order to be able to separate those regions in different subroutines. By doing so, we will run in parallel some pieces of the inner loop's iterations, and then synchronize to sort and compare critical arrays and variables, and finally run the remaining work, again, in parallel. Since there are at least one hundred thousand, quite small, iterations in the inner loop, the chosen parallel environment will be OpenMP Tasks. Then, a GPU oriented parallel implementation will be attempted, either using the OpenCL or the CUDA environments. The first is more portable, while the latter should be faster. | ||
| + | |||
| + | 5. Dissemination | ||
| + | We intend to publish complete results of this research in Astronomy as well as HPC-related journals and conference proceedings, like Astrophysics and Space Science Journal, The Astrophysics Journal, the International Conference on Computational Science, etc. | ||
| + | |||
| + | 6. Running on larger systems | ||
| + | The results of this project will ensure that this application will be able to run on bigger systems because the main parallelization strategy is data-parallel and thus is suitable for large scale computing systems. Numerous instances of the MPI-versions of the LNAWENR application (parallelized with OpenMP underneath) can be run on systems of arbitrary scale. | ||
Revision as of 15:49, 13 August 2013
Contents |
General Information
- Application's name: Asteroseismic pipeline and Web based portal based on LNAWENR software code (linear, non-radial, non-adiabatic Romanian model) to be used in the KASC/KEPLER Space Mission
- Application's acronym: LNAWENR
- Scientific domain: Space Science
- Contact person: Dr. Marian Doru Suran, suran<>aira.astro.ro
- Main Developers: Astronomical Institute of the Romanian Academy, Astrophysical Department, Stellar Group, Bucharest, Romania
- Co-developers: UPB, Bucharest, Romania
- Allocation period: 01/11/2012 – 31/12/2012
- Web site: http://www.astro.ro/~suran, to be updated
Objectives of the computing project
Parallelization of the LNAWENR software code. Development of the evolutive (stellar and pulsation) grid for stars with masses [0.8Msun-6.0Msun]. Development of the metaheuristic search algorithms and the pipe-line for stellar parameters determination in the previous grid Development of the Web-based portal for the LNAWENR pipe-line, in order to be used in the Kepler Space Mission by Kepler Asteroseismic Scientific Community (KASC) Application for ~ 1000 stars observed by diffrent Space Missions in the field: MOST, CoRoT, KEPLER.
Application's description
The development of an automated pipeline in the field of asteroseismology. The pipeline will include: - the development of an extensive evolutive (stellar and pulsation) grid for stars with masses [0.8Msun- 6.0Msun] and different metallicities; - the development of an automatic routine for modes extraction and modes ID from pulsating stars light curve; - the development of a fitting method (based on the already created grid) for each star, based on metaheuristic search algorithms in order to determine the absolute elements for each star (mass, radius, luminosity, effective temperature, gravity, metallicity, age, large and small separations).
The stars (>1000) are stars observed by different Space Missions in the field of asteroseismology (MOST, CoRoT and KEPLER SPace Missions).
Results expected in the allocation period
The development of the pipeline in the area of mode ID and integrate it with stellar modeling pipeline for a complete automation of the procedure. - Complete the treatment of surface effects and of the physical processes (e.g. convection, core overshooting, diffusion of chemical elements, etc.) through successful application of the prototype pipeline to space-based data on stars observed by CoRoT (use public data from the CoRoT satellite) and Kepler missions. - Extend the fitting method to include rotation velocity as an additional parameter and validate with data from CoRoT and Kepler missions and application of the prototype pipeline for asteroseismic data interpretation for Kepler mission (launch of the PLATO mission is planned for 2017/2018).
Activity report
1. Scientific results and impact During this reporting period, we developed the pipeline in the area of mode ID and we integrated it with another stellar modelling pipeline for a complete automation of the procedure. • We completed the treatment of surface effects and of the physical processes (like convection, core overshooting, diffusion of chemical elements, etc.) through successful application of the prototype pipeline to space-based data on stars observed by CoRoT and Kepler missions. • We extended the fitting method to include rotation velocity as an additional parameter and validated it with data from CoRoT and Kepler missions and application of the prototype pipeline for asteroseismic data interpretation for the Kepler mission. • A hybrid parallel implementation of the LNAWENR application was produced, using MPI for data-parallel distribution of the workload across various computing nodes in the NCIT-Cluster, and OpenMP for the compute intensive kernel of the application. • A C version of the original Fortran code was developed and initial steps toward an OpenCL and CUDA versions of the code have been done.
2. Scientific collaborations served The data obtained from this project will be used by our team in the interpretation of data from other dedicated Space Mission like MOST, CoRoT, and Kepler.
3. Social impact Since the notoriety of the Corot and Kepler missions is so big, the social impact of the developments obtained in this project are significant. The finding of one extra solar system planet is a very important issue in the world of astronomy, and also for the public at large.
4. Technical results The current implementation of the LNAWENR application, one of the eight non-adiabatic models in use now by the international scientific community was parallelized and optimized.. The work was focused on both improving the initial, sequential version, of the implementation, and on implementing parallel versions using the MPI and OpenMP parallel programming paradigms. The results are quite promising. The MPI version scales with the number of independent components that are analyzed, while the best performance gain obtained so far for the shared-memory kernel is that the current implementation runs 5 times faster than the initial one, using OpenMP threads, on four CPU cores. The work on this project is far from finished, since some further optimizations are already in the process of implementation. We will have to further rewrite the code so that some labels that are still inside the main inner loop of the program, will be moved in order to be able to separate those regions in different subroutines. By doing so, we will run in parallel some pieces of the inner loop's iterations, and then synchronize to sort and compare critical arrays and variables, and finally run the remaining work, again, in parallel. Since there are at least one hundred thousand, quite small, iterations in the inner loop, the chosen parallel environment will be OpenMP Tasks. Then, a GPU oriented parallel implementation will be attempted, either using the OpenCL or the CUDA environments. The first is more portable, while the latter should be faster.
5. Dissemination We intend to publish complete results of this research in Astronomy as well as HPC-related journals and conference proceedings, like Astrophysics and Space Science Journal, The Astrophysics Journal, the International Conference on Computational Science, etc.
6. Running on larger systems The results of this project will ensure that this application will be able to run on bigger systems because the main parallelization strategy is data-parallel and thus is suitable for large scale computing systems. Numerous instances of the MPI-versions of the LNAWENR application (parallelized with OpenMP underneath) can be run on systems of arbitrary scale.
