Next: About this document ...
Up: Parallel Implementation of the
Previous: Summary and Conclusions
- Allard, F., & Hauschildt, P. H. 1995, ApJ, 445, 433
- Allard, F., Hauschildt, P., Alexander, D. R., & Starrfield, S. 1997, ARAA, 36, in press
- [Baron et al.(1995)Baron, Hauschildt, Branch, Austin, Garnavich, Ann, Wagner, Filippenko, Matheson, and Liebert]b93j3Baron, E., Hauschildt, P. H., Branch, D., Austin, S., Garnavich, P., Ann, H. B., Wagner, R. M., Filippenko, A. V., Matheson, T., & Liebert, J. 1995, ApJ, 441, 170
- Baron, E., Hauschildt, P. H., & Young, T. R. 1995, Physics Reports, 256, 23
- [Baron et al.(1996a)Baron, Hauschildt, Branch, Kirshner, and Filippenko]b94i1Baron, E., Hauschildt, P. H., Branch, D., Kirshner, R. P., & Filippenko, A. V. 1996a, MNRAS, 279, 779
- Baron, E., Hauschildt, P. H., Nugent, P., & Branch, D. 1996b, MNRAS, 283, 297
- [Cowan et al.(1997)Cowan, McWilliam, Sneden, and Burris]cowanetal97Cowan, J. J., McWilliam, A., Sneden, C., & Burris, D. L. 1997, , 480, 246
- Gropp, W., Lusk, E., Doss, N., & Skjellum, A. 1996.
-
- MPICH Model MPI Implementation Reference Manual.
Technical report, Argonne National Laboratory, Argonne
- Hauschildt, P. H. 1992, JQSRT, 47, 433
- Hauschildt, P. H. 1993, JQSRT, 50, 301
- [Hauschildt & Baron(1995)Hauschildt and Baron]hbfe295Hauschildt, P. H., & Baron, E. 1995, JQSRT, 54, 987
- Hauschildt, P. H., Starrfield, S., Shore, S. N., Allard, F., & Baron, E. 1995, ApJ, 447, 829
- [Hauschildt et al.(1996)Hauschildt, Baron, Starrfield, and Allard]phhnovfe296Hauschildt, P. H., Baron, E., Starrfield, S., & Allard, F. 1996, ApJ, 462, 386
- Hauschildt, P. H., Baron, E., & Allard, F. 1997, ApJ, 483, 390
- [Message Passing Interface Forum(1995)Message Passing Interface Forum]mpistdMessage Passing Interface Forum 1995.
- MPI: A Message-Passing Interface Standard, Version 1.1, Knoxville, TN: Univ. of Tennessee
- Mihalas, D., & Mihalas, B. W. 1984.
- Foundations of Radiation Hydrodynamics, Oxford: Oxford University
- Nugent, P., Baron, E., Hauschildt, P., & Branch, D. 1995a, ApJ, 441, L33
- [Nugent et al.(1995b)Nugent, Phillips, Baron, Branch, and Hauschildt]nugseq95Nugent, P., Phillips, M., Baron, E., Branch, D., & Hauschildt, P. 1995b, ApJ, 455, L147
- Nugent, P., Baron, E., Branch, D., Fisher, A., & Hauschildt, P. 1997, ApJ, 485, in press
- [Prochaska & Wolfe(1997)Prochaska and Wolfe]pwlya97Prochaska, J. X., & Wolfe, A. M. 1997, ApJ, 474, 140
- Schwarz, G., Hauschildt, P. H., S. Starrfield, E. B., Allard, F., Shore, S., & Sonneborn, G. 1997, MNRAS, 284, 669
-
Figure 1:
The basic design of our parallelization
method, groups of processors are divided up into wavelength clusters
which will work on individual wavelength points, the wavelength
clusters are further divided into worker nodes, where each worker node
is assign a set of specific (e.g., spatially distributed) tasks. Our design
requires that each worker node on all wavelength clusters work on
exactly the same set of tasks, although additional inherently serial
operations can be assigned to one particular master worker, or master
wavelength cluster. This method reduces communication between clusters
to its absolute minimum and allows the maximum speedup.
|
Figure:
Pseudo-code for the global layout of
PHOENIX. The processing that is required before and after the
radiative transfer is indicated. Both pre- and post-processing phases
can be executed in parallel and independently for all clusters.
|
Figure 3:
Scalability of the static Vega model atmosphere
test run as function of the number of nodes (processing elements or
nodes) used. The y-axis gives the speedup obtained relative to the serial
run. The different symbols show the results for different numbers of
worker tasks for each wavelength cluster.
|
Figure 4:
Scalability of the Nova model atmosphere test run
as function of the number of nodes (processing elements or nodes) used.
The y-axis gives the speedup obtained relative to the serial run. The
different symbols show the results for different numbers of worker tasks
for each wavelength cluster.
|
Figure 5:
Scalability of the Supernova model atmosphere test run
as function of the number of nodes (processing elements or nodes) used.
The y-axis gives the speedup obtained relative to the serial run. The
different symbols show the results for different numbers of worker tasks
for each wavelength cluster.
|
Next: About this document ...
Up: Parallel Implementation of the
Previous: Summary and Conclusions
Peter H. Hauschildt
4/27/1999