source: soft/giet_vm/applications/ocean/README.ocean

Last change on this file was 581, checked in by laurent, 9 years ago

Adding ocean application, some mathematics functions and distributed locks

File size: 3.8 KB
Line 
1GENERAL INFORMATION:
2
3The OCEAN program simulates large-scale ocean movements based on eddy and
4boundary currents, and is an enhanced version of the SPLASH Ocean code.
5A description of the functionality of this code can be found in the
6original SPLASH report.  The implementations contained in SPLASH-2
7differ from the original SPLASH implementation in the following ways:
8
9  (1) The SPLASH-2 implementations are written in C rather than
10      FORTRAN.
11  (2) Grids are partitioned into square-like subgrids rather than
12      groups of columns to improve the communication to computation
13      ratio.
14  (3) The SOR solver in the SPLASH Ocean code has been replaced with a
15      restricted Red-Black Gauss-Seidel Multigrid solver based on that
16      presented in:
17
18      Brandt, A. Multi-Level Adaptive Solutions to Boundary-Value Problems.
19           Mathematics of Computation, 31(138):333-390, April 1977.
20
21      The solver is restricted so that each processor has as least two
22      grid points in each dimension in each grid subpartition.
23
24Two implementations are provided in the SPLASH-2 distribution:
25
26  (1) Non-contiguous partition allocation
27
28      This implementation (contained in the non_contiguous_partitions
29      subdirectory) implements the grids to be operated on with
30      two-dimensional arrays.  This data structure prevents partitions
31      from being allocated contiguously, but leads to a conceptually
32      simple programming implementation.
33
34  (2) Contiguous partition allocation
35
36      This implementation (contained in the contiguous_partitions
37      subdirectory) implements the grids to be operated on with
38      3-dimensional arrays.  The first dimension specifies the processor
39      which owns the partition, and the second and third dimensions
40      specify the x and y offset within a partition.  This data structure
41      allows partitions to be allocated contiguously and entirely in the
42      local memory of processors that "own" them, thus enhancing data
43      locality properties.
44
45The contiguous partition allocation implementation is described in:
46
47Woo, S. C., Singh, J. P., and Hennessy, J. L.  The Performance Advantages
48     of Integrating Message Passing in Cache-Coherent Multiprocessors.
49     Technical Report CSL-TR-93-593, Stanford University, December 1993.
50
51A detailed description of both versions will appear in the SPLASH-2 report.
52The non-contiguous partition allocation implementation is conceptually
53similar, except for the use of statically allocated 2-dimensional arrays.
54
55These programs work under both the Unix FORK and SPROC models.
56
57RUNNING THE PROGRAM:
58
59To see how to run the program, please see the comment at the top of the
60file main.C, or run the application with the "-h" command line option.
61Five command line parameters can be specified, of which the ones which
62would normally be changed are the number of grid points in each dimension,
63and the number of processors.  The number of grid points must be a
64(power of 2+2) in each dimension (e.g. 130, 258, etc.).  The number of
65processors must be a power of 2.  Timing information is printed out at
66the end of the program.  The first timestep is considered part of the
67initialization phase of the program, and hence is not included in the
68"Total time without initialization."
69
70BASE PROBLEM SIZE:
71
72The base problem size for an upto-64 processor machine is a 258x258 grid.
73The default values should be used for other parameters (except the number
74of processors, which can be varied).  In addition, sample output files
75for the default parameters for each version of the code are contained in
76the file correct.out in each subdirectory.
77
78DATA DISTRIBUTION:
79
80Our "POSSIBLE ENHANCEMENT" comments in the source code tell where one
81might want to distribute data and how.  Data distribution has an impact
82on performance on the Stanford DASH multiprocessor.
83
Note: See TracBrowser for help on using the repository browser.