1 |
$Header: /u/gcmpack/models/MITgcmUV/doc/README,v 1.11 1998/07/10 18:48:48 cnh Exp $ |
2 |
|
3 |
|
4 |
|
5 |
MITgcmUV Getting Started |
6 |
======================== |
7 |
|
8 |
o Introduction |
9 |
|
10 |
This note is a guide to using the MIT General Circulation Model Ultra Verstaile |
11 |
implementation, MITgmcUV. MITgcmUV is a Fortran code that implements the |
12 |
algorithm described in Marshall et. al. 1997, Hill, Adcroft, ... |
13 |
The MITgcmUV implementation is designed to work efficiently on all classes |
14 |
of computer platforms. It can be used in both a single processor mode |
15 |
and a parallel processor mode. Parallel processing can be either multi-threaded |
16 |
shared memory such as that found on CRAY T90 machines or it can be multi-process |
17 |
distributed memory. A set of "execution enviroment" support routines are |
18 |
used to allow the same numerical code to run on top of a single-process, multi-threaded |
19 |
or distributed multi-process configuration. |
20 |
|
21 |
o Installing |
22 |
To setup the model on a particular computer the code tree must be created |
23 |
and appropriate compile and run scripts set up. For some platforms |
24 |
the necessary scripts are included in the release - in this case follow |
25 |
the steps below: |
26 |
|
27 |
1. Extract MITgcmUV from the downloadable archive |
28 |
tar -xvf checkpoint12.tar |
29 |
|
30 |
2. Create platform specific make file |
31 |
For example on a Digital UNIX machine the script "genmake.dec" can |
32 |
be used as shown below |
33 |
|
34 |
cd bin |
35 |
../tools/genmake |
36 |
cp Makefile.alpha Makefile ( On Alpha machine) |
37 |
|
38 |
3. Now create header file dependency entries |
39 |
make depend |
40 |
|
41 |
4. Compile code |
42 |
make |
43 |
|
44 |
5. Copy input files |
45 |
cp ../verification/exp2/[a-z]* ../verification/exp2/*bin . |
46 |
|
47 |
6. Run baseline test case |
48 |
setenv PARALLEL 1 |
49 |
dmpirun -np 2 ../exe/mitgcmuv ( Under Digital UNIX ) |
50 |
mpirun.p4shmem ../exe/mitgcmuv -np 2 ( Under Solaris + mpich) |
51 |
|
52 |
|
53 |
This runs a 4 degree global ocean climatological simulation. |
54 |
By default this code is set to use two porcessors splitting |
55 |
the model domain along the equator. Textual output is written |
56 |
to files STDOUT.* and STDERR.* with one file for each process. |
57 |
Model fileds are written to files suffixed .data and .meta |
58 |
These files are written on a per process basis. The .meta |
59 |
file indicates the location and shape of the subdomain in |
60 |
each .data file. |
61 |
|
62 |
o Running |
63 |
|
64 |
- Input and output files |
65 |
|
66 |
Required files |
67 |
============== |
68 |
The model is configured to look for two files with fixed names. |
69 |
These files are called |
70 |
"eedata" and "data". |
71 |
The file eedata contains "execution environment" data. At present |
72 |
this consists of a specification of the number of threads to |
73 |
use in X and Y under multithreaded execution. |
74 |
|
75 |
- Serial execution |
76 |
|
77 |
- Parallel execution. Threads |
78 |
nSx, nSy |
79 |
setenv PARALLEL n |
80 |
nTx=2, nTy=2 |
81 |
|
82 |
- Parallel execution. MPI |
83 |
mPx, nPy |
84 |
dmpirun |
85 |
|
86 |
- Parallel execution. Hybrid |
87 |
|
88 |
o Cutomising the code |
89 |
|
90 |
Model structure |
91 |
=============== |
92 |
The "numerical" model is contained within a execution |
93 |
environment support wrapper. This wrapper is designed |
94 |
to provide a general framework for grid-point models. |
95 |
MITgcmUV is a specific numerical model that uses the |
96 |
framework. |
97 |
Under this structure the model is split into |
98 |
execution environment support code and conventional |
99 |
numerical model code. The execution environment |
100 |
support code is held under the "eesupp" directory. |
101 |
The grid point model code is held under the |
102 |
"model" directory. |
103 |
Code execution actually starts in the eesupp |
104 |
routines and not in the model routines. For this |
105 |
reason the top level main.F is in the eesupp/src |
106 |
directory. End-users should not need to worry about |
107 |
this level. The top-level routine for the numerical |
108 |
part of the code is in model/src/the_model_main.F. |
109 |
|
110 |
|
111 |
o References |
112 |
Web sites - HP |
113 |
for doc Digital |
114 |
SGI |
115 |
Sun |
116 |
Linux threads |
117 |
CRAY multitasking |
118 |
PPT notes |