Szczegóły publikacji
Opis bibliograficzny
Interactive n-body simulations on the grid: HLA versus MPI / Katarzyna RYCERZ, Alfredo Tirado-Ramos, Alessia Gualandris, Simon F. Portegies Zwart, Marian BUBAK, Peter M. A. Sloot // The International Journal of High Performance Computing Applications ; ISSN 1094-3420. — 2007 — vol. 21 no. 2 spec. iss., s. 210–221. — Bibliogr. s. 220–221, Abstr. — Publikacja dostępna online od: 2007-05-01. — M. Bubak - dod. afiliacja: University of Amsterdam. — Selected papers from EuroPVM/MPI 2005 conference, Sorrento, Italy, 18–21 September 2005. — London : SAGE Publications, [2007]
Autorzy (6)
- AGHRycerz Katarzyna
- Tirado-Ramos Alfredo
- Gualandris Alessia
- Zwart Simon F. Portegies
- AGHBubak Marian
- Sloot Peter M. A.
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 37696 |
|---|---|
| Data dodania do BaDAP | 2008-02-27 |
| DOI | 10.1177/1094342007074874 |
| Rok publikacji | 2007 |
| Typ publikacji | referat w czasopiśmie |
| Otwarty dostęp |  |
| Czasopismo/seria | International Journal of High Performance Computing Applications |
Abstract
The grid paradigm for distributed computation provides an interesting framework for the simulation of dense stellar systems, which remains a great challenge for astrophysicists. In this work we apply a distributed simulation model to the astrophysical N-body problem: the High Level Architecture (HLA), using distributed federations on top of grid infrastructures for the communication between simulation and visualization components. To achieve this goal, we use a grid HLA Management system (G-HLAM) for HLA-based simulations running on the grid. We compare this setup with a direct parallelization of the N-body code using MPI on the grid. Our aim is to provide scientists with an interactive environment where, in contrast to traditional MPI-based parallel processing, seamless simulation process migration and rollbacks can be invoked on the fly to improve the overall computational performance. We found that MPI and HLA are complementary rather than competing technologies, as they can be used simultaneously to improve the performance of complex simulations on different fronts. On one hand, if the communication to computation ratio is small enough, the MPI parallelization proves to be a relatively easy and efficient method for taking advantage of grid technology. On the other hand, HLA provides advanced mechanisms to synchronize simulation and visualization components located on different grid nodes and can be used to add steering mechanisms to improve interaction. Finally, we found that HLA allows migration of running simulations to be easily managed, reducing overall communication times between simulation and the user, effectively improving the overall system performance.
