Experience with automatic, dynamic load balancing and adaptive finite element computation
- Sandia National Labs., Albuquerque, NM (United States)
- Rensselaer Polytechnic Inst., Troy, NY (United States). Dept. of Computer Science
- New Mexico Univ., Albuquerque, NM (United States). Dept. of Computer Science
Distributed memory, Massively Parallel (MP), MIMD technology has enabled the development of applications requiring computational resources previously unobtainable. Structural mechanics and fluid dynamics applications, for example, are often solved by finite element methods (FEMs) requiring, millions of degrees of freedom to accurately simulate physical phenomenon. Adaptive methods, which automatically refine or coarsen meshes and vary the order of accuracy of the numerical solution, offer greater robustness and computational efficiency than traditional FEMs by reducing the amount of computation required away from physical structures such as shock waves and boundary layers. On MP computers, FEMs frequently result in distributed processor load imbalances. To overcome load imbalance, many MP FEMs use static load balancing as a preprocessor to the finite element calculation. Adaptive methods complicate the load imbalance problem since the work per element is not uniform across the solution domain and changes as the computation proceeds. Therefore, dynamic load balancing is required to maintain global load balance. We describe a dynamic, fine-grained, element-based data migration system that maintains global load balance and is effective in the presence of changing work loads. Global load balance is achieved by overlapping neighborhoods of processors, where each neighborhood performs local load balancing. The method utilizes an automatic element management system library to which a programmer integrates the application`s computational description. The library`s flexibility supports a large class of finite element and finite difference based applications.
- Research Organization:
- Sandia National Labs., Albuquerque, NM (United States)
- Sponsoring Organization:
- USDOE, Washington, DC (United States)
- DOE Contract Number:
- AC04-94AL85000
- OSTI ID:
- 10102903
- Report Number(s):
- SAND-93-2172C; CONF-940136-2; ON: DE94001422; BR: GB0103012
- Resource Relation:
- Conference: 27. Hawaii international conference on system sciences,Maui, HI (United States),4-7 Jan 1994; Other Information: PBD: [1993]
- Country of Publication:
- United States
- Language:
- English
Similar Records
Dynamic load balancing of applications
A massively parallel adaptive finite element method with dynamic load balancing