Latency Hiding and Adaptability for Parallel Iterative Algorithms
Switching from the MPI process model to a threaded model in the parallel environment via user level threads takes advantage of the existing concurrence in applications. Major aims for this work are to face the computational demands of scientific applications by exploiting concurrency, minimizing latency due to message passing, and to present and evaluate strategies to implement adaptability when imbalance occurs.
KeywordsLoad Balance Load Balance Strategy Single Program Multiple Data Jacobi Algorithm Physical Processor
Work partially supported through grants MEC-TIN 2005-00447 and JA-P06-TIC-01426.
- 1.Pancake CM (1993) Multithreaded languages for scientific and technical computing. In: Proceedings of the IEEE 81(2)Google Scholar
- 3.Kak AC, Slaney M (2001) Principles of computerized tomographic imaging. SIAM Society for Industrial and Applied MathematicsGoogle Scholar
- 4.Fernández J, Lawrence AF, Roca J, García I, Ellisman MH, Carazo JM (2002) High performance electron tomography of complex biological specimens. J Struct Biol 138:6–20Google Scholar
- 5.Huang Ch, Zheng G, Kumar S, Kale LV (2006) Performance evaluation of Adaptive MPI. In: Proceedings of ACM SIGPLAN Symposium on Principles and Practice of Parallel ProgrammingGoogle Scholar