1 |
% $Header: /u/gcmpack/manual/part6/mnc.tex,v 1.13 2004/10/12 21:44:59 edhill Exp $ |
2 |
% $Name: $ |
3 |
|
4 |
\section{NetCDF I/O Integration: MNC} |
5 |
\label{sec:pkg:mnc} |
6 |
\begin{rawhtml} |
7 |
<!-- CMIREDIR:package_mnc: --> |
8 |
\end{rawhtml} |
9 |
|
10 |
The \texttt{mnc} package is a set of convenience routines written to |
11 |
expedite the process of creating, appending, and reading NetCDF files. |
12 |
NetCDF is an increasingly popular self-describing file format |
13 |
\cite{rew:97} intended primarily for scientific data sets. An |
14 |
extensive collection of NetCDF reference papers, user guides, |
15 |
software, FAQs, and other information can be obtained from UCAR's web |
16 |
site at: |
17 |
\begin{rawhtml} <A href="http://www.unidata.ucar.edu/packages/netcdf/"> \end{rawhtml} |
18 |
\begin{verbatim} |
19 |
http://www.unidata.ucar.edu/packages/netcdf/ |
20 |
\end{verbatim} |
21 |
\begin{rawhtml} </A> \end{rawhtml} |
22 |
|
23 |
|
24 |
\subsection{Using MNC} |
25 |
|
26 |
\subsubsection{MNC Configuration} |
27 |
|
28 |
As with all MITgcm packages, MNC can be turned on or off at compile time |
29 |
using the \texttt{packages.conf} file or the \texttt{genmake2} |
30 |
\texttt{-enable=mnc} or \texttt{-disable=mnc} switches. |
31 |
|
32 |
While MNC is likely to work ``as is'', there are a few compile--time |
33 |
constants that may need to be increased for simulations that employ |
34 |
large numbers of tiles within each process. Note that the important |
35 |
quantity is the maximum number of tiles \textbf{per process}. Since |
36 |
MPI configurations tend to distribute large numbers of tiles over |
37 |
relatively large numbers of MPI processes, these constants will rarely |
38 |
need to be increased. |
39 |
|
40 |
If MNC runs out of space within its ``lookup'' tables during a |
41 |
simulation, then it will provide an error message along with a |
42 |
recommendation of which parameter to increase. The parameters are all |
43 |
located within \filelink{pkg/mnc/mnc\_common.h}{pkg-mnc-mnc_common.h} |
44 |
and the ones that may need to be increased are: |
45 |
|
46 |
\begin{center} |
47 |
{\footnotesize |
48 |
\begin{tabular}[htb]{|l|r|l|}\hline |
49 |
\textbf{Name} & |
50 |
\textbf{Default} & \textbf{Description} \\\hline |
51 |
& & \\ |
52 |
\texttt{MNC\_MAX\_ID} & 1000 & |
53 |
\textbf{IDs for various low-level entities} \\ |
54 |
\texttt{MNC\_MAX\_INFO} & 400 & |
55 |
\textbf{IDs (mostly for object sizes)} \\ |
56 |
\texttt{MNC\_CW\_MAX\_I} & 150 & |
57 |
\textbf{IDs for the ``wrapper'' layer} \\\hline |
58 |
\end{tabular} |
59 |
} |
60 |
\end{center} |
61 |
|
62 |
In those rare cases where MNC ``out-of-memory'' error messages are |
63 |
encountered, it is a good idea to increase the too-small parameter by |
64 |
a factor of \textbf{2--10} in order to avoid wasting time on an |
65 |
iterative compile--test sequence. |
66 |
|
67 |
|
68 |
\subsubsection{MNC Inputs} |
69 |
|
70 |
For run-time configuration, most of the MNC--related model parameters |
71 |
are contained within a Fortran namelist file called \texttt{data.mnc}. |
72 |
If this file does not exist, then the MNC package will interpret that |
73 |
as an indication that it is not to be used. If the \texttt{data.mnc} |
74 |
file does exist, then it may contain the following parameters: |
75 |
|
76 |
\begin{center} |
77 |
{\footnotesize |
78 |
\begin{tabular}[htb]{|l|c|l|l|}\hline |
79 |
\textbf{Name} & \textbf{T} & |
80 |
\textbf{Default} & \textbf{Description} \\\hline |
81 |
& & & \\ |
82 |
\texttt{useMNC} & L & \texttt{.FALSE.} & |
83 |
\textbf{overall MNC ON/OFF switch} \\ |
84 |
\texttt{mnc\_echo\_gvtypes} & L & \texttt{.FALSE.} & |
85 |
echo pre-defined ``types'' (debugging) \\ |
86 |
\texttt{mnc\_use\_outdir} & L & \texttt{.FALSE.} & |
87 |
create a directory for output \\ |
88 |
\texttt{mnc\_outdir\_str} & S & \texttt{'mnc\_'} & |
89 |
output directory name \\ |
90 |
\texttt{mnc\_outdir\_date} & L & \texttt{.FALSE.} & |
91 |
embed date in the output dir name \\ |
92 |
\texttt{pickup\_write\_mnc} & L & \texttt{.FALSE.} & |
93 |
use MNC to write (create) pickup files \\ |
94 |
\texttt{pickup\_read\_mnc} & L & \texttt{.FALSE.} & |
95 |
use MNC to read pickup files \\ |
96 |
\texttt{mnc\_use\_indir} & L & \texttt{.FALSE.} & |
97 |
use a directory (path) for input \\ |
98 |
\texttt{mnc\_indir\_str} & S & \texttt{''} & |
99 |
input directory (or path) name \\ |
100 |
\texttt{snapshot\_mnc} & L & \texttt{.FALSE.} & |
101 |
write \texttt{snapshot} (instantaneous) w/MNC \\ |
102 |
\texttt{monitor\_mnc} & L & \texttt{.FALSE.} & |
103 |
write \texttt{monitor} w/MNC \\ |
104 |
\texttt{timeave\_mnc} & L & \texttt{.FALSE.} & |
105 |
write \texttt{timeave} w/MNC \\ |
106 |
\texttt{autodiff\_mnc} & L & \texttt{.FALSE.} & |
107 |
write \texttt{autodiff} w/MNC \\\hline |
108 |
\end{tabular} |
109 |
} |
110 |
\end{center} |
111 |
|
112 |
Additional MNC--related parameters are contained within the main |
113 |
\texttt{data} namelist file and in some of the namelist files for |
114 |
individual packages. These options are: |
115 |
\begin{center} |
116 |
{\footnotesize |
117 |
\begin{tabular}[htb]{|l|c|l|l|}\hline |
118 |
\textbf{Name} & \textbf{T} & |
119 |
\textbf{Default} & \textbf{Description} \\\hline |
120 |
\multicolumn{4}{|c|}{\ } \\ |
121 |
\multicolumn{4}{|c|}{Main namelist file: |
122 |
``\textbf{data}''} \\\hline |
123 |
\texttt{snapshot\_ioinc} & L & \texttt{.FALSE.} & |
124 |
write \texttt{snapshot} ``inclusively'' \\ |
125 |
\texttt{timeave\_ioinc} & L & \texttt{.FALSE.} & |
126 |
write \texttt{timeave} ``inclusively'' \\ |
127 |
\texttt{monitor\_ioinc} & L & \texttt{.FALSE.} & |
128 |
write \texttt{monitor} ``inclusively'' \\ |
129 |
\texttt{the\_run\_name} & C & ``name...'' & |
130 |
name is included in all MNC output \\\hline |
131 |
\multicolumn{4}{|c|}{\ } \\ |
132 |
\multicolumn{4}{|c|}{Diagnostics namelist file: |
133 |
``\textbf{data.diagnostics}''} \\\hline |
134 |
\texttt{diag\_mnc} & L & \texttt{.FALSE.} & |
135 |
write \texttt{diagnostics} w/MNC \\ |
136 |
\texttt{diag\_ioinc} & L & \texttt{.FALSE.} & |
137 |
write \texttt{diagnostics} ``inclusively'' \\\hline |
138 |
\end{tabular} |
139 |
} |
140 |
\end{center} |
141 |
|
142 |
By default, turning on MNC for a particular output type will result in |
143 |
turning off all the corresponding (usually, default) MDSIO or STDOUT |
144 |
output mechanisms. In other words, output defaults to being an |
145 |
exclusive selection. To enable multiple kinds of simultaneous output, |
146 |
flags of the form \texttt{NAME\_ioinc} have been created where |
147 |
\texttt{NAME} corresponds to the various MNC output flags. When a |
148 |
\texttt{NAME\_ioinc} flag is set to \texttt{.TRUE.}, then multiple |
149 |
simultaneous forms of output are allowed for the \texttt{NAME} output |
150 |
mechanism. The intent of this design is that typical users will only |
151 |
want one kind of output while people debugging the code (particularly |
152 |
the I/O routines) may want simultaneous types of output. |
153 |
|
154 |
This ``inclusive'' versus ``exclusive'' design is easily applied in |
155 |
cases where three or more kinds of output may be generated. Thus, it |
156 |
can be readily extended to additional new output types (eg. HDF5). |
157 |
|
158 |
Input types are always exclusive. |
159 |
|
160 |
\subsubsection{MNC Output} |
161 |
|
162 |
While NetCDF files are supposed to be ``self-describing'', it is |
163 |
helpful to note the following: |
164 |
|
165 |
\begin{itemize} |
166 |
\item The constraints placed upon the ``unlimited'' (or ``record'') |
167 |
dimension inherent with NetCDF v3.x make it very inefficient to put |
168 |
variables written at potentially different intervals within the same |
169 |
file. For this reason, MNC output is split into a few file ``base |
170 |
names'' which try to reflect the nature of their content. |
171 |
|
172 |
\item All MNC output is currently done in a ``tile-per-file'' fashion |
173 |
since most NetCDF v3.x implementions cannot write safely within MPI |
174 |
or multi-threaded environments. This tiling is done in a global |
175 |
fashion and the tile numbers are appended to the base names |
176 |
described above. Some scripts to ``assemble'' output are available |
177 |
(\texttt{MITgcm/utils/matlab}). More general manipulations can be |
178 |
accomplished with the |
179 |
\begin{rawhtml} |
180 |
<A href="http://nco.sourceforge.net"> |
181 |
\end{rawhtml} |
182 |
\begin{verbatim} |
183 |
NetCDF Operators (or ``NCO'') at http://nco.sourceforge.net |
184 |
\end{verbatim} |
185 |
\begin{rawhtml} </A> \end{rawhtml} |
186 |
which is a very powerful and convenient set of tools for working |
187 |
with all NetCDF files. |
188 |
|
189 |
\item On many systems, NetCDF has practical file size limits on the |
190 |
order of 2--4GB (the maximium memory addressable with 32bit |
191 |
pointers) due to a lack of operating system, compiler, and/or |
192 |
library support. In cases where this limit is reached, it is |
193 |
generally a good idea to reduce write frequencies or restart from |
194 |
pickups. |
195 |
|
196 |
\item MNC does not (yet) provide a mechanism for reading information |
197 |
from a single ``global'' file as can be done with the MDSIO |
198 |
package. This is in progress. |
199 |
|
200 |
\end{itemize} |
201 |
|
202 |
|
203 |
\subsection{MNC Internals} |
204 |
|
205 |
The \texttt{mnc} package is a two-level convenience library (or |
206 |
``wrapper'') for most of the NetCDF Fortran API. Its purpose is to |
207 |
streamline the user interface to NetCDF by maintaining internal |
208 |
relations (look-up tables) keyed with strings (or names) and entities |
209 |
such as NetCDF files, variables, and attributes. |
210 |
|
211 |
The two levels of the \texttt{mnc} package are: |
212 |
\begin{description} |
213 |
|
214 |
\item[Upper level] \ |
215 |
|
216 |
The upper level contains information about two kinds of |
217 |
associations: |
218 |
\begin{description} |
219 |
\item[grid type] is lookup table indexed with a grid type name. |
220 |
Each grid type name is associated with a number of dimensions, the |
221 |
dimension sizes (one of which may be unlimited), and starting and |
222 |
ending index arrays. The intent is to store all the necessary |
223 |
size and shape information for the Fortran arrays containing |
224 |
MITgcm--style ``tile'' variables (that is, a central region |
225 |
surrounded by a variably-sized ``halo'' or exchange region as |
226 |
shown in Figures \ref{fig:communication_primitives} and |
227 |
\ref{fig:tiling-strategy}). |
228 |
|
229 |
\item[variable type] is a lookup table indexed by a variable type |
230 |
name. For each name, the table contains a reference to a grid |
231 |
type for the variable and the names and values of various |
232 |
attributes. |
233 |
\end{description} |
234 |
|
235 |
Within the upper level, these associations are not permanently tied |
236 |
to any particular NetCDF file. This allows the information to be |
237 |
re-used over multiple file reads and writes. |
238 |
|
239 |
\item[Lower level] \ |
240 |
|
241 |
In the lower (or internal) level, associations are stored for NetCDF |
242 |
files and many of the entities that they contain including |
243 |
dimensions, variables, and global attributes. All associations are |
244 |
on a per-file basis. Thus, each entity is tied to a unique NetCDF |
245 |
file and will be created or destroyed when files are, respectively, |
246 |
opened or closed. |
247 |
|
248 |
\end{description} |
249 |
|
250 |
|
251 |
\subsubsection{MNC Grid--Types and Variable--Types} |
252 |
|
253 |
As a convenience for users, the MNC package includes numerous routines |
254 |
to aid in the writing of data to NetCDF format. Probably the biggest |
255 |
convenience is the use of pre-defined ``grid types'' and ``variable |
256 |
types''. These ``types'' are simply look-up tables that store |
257 |
dimensions, indicies, attributes, and other information that can all |
258 |
be retrieved using a single character string. |
259 |
|
260 |
The ``grid types'' are a way of mapping variables within MITgcm to |
261 |
NetCDF arrays. Within MITgcm, most spatial variables are defined |
262 |
using two-- or three--dimensional arrays with ``overlap'' regions (see |
263 |
Figures \ref{fig:communication_primitives}, a possible vertical index, |
264 |
and \ref{fig:tiling-strategy}) and tile indicies such as the following |
265 |
``U'' velocity: |
266 |
\begin{verbatim} |
267 |
_RL uVel (1-OLx:sNx+OLx,1-OLy:sNy+OLy,Nr,nSx,nSy) |
268 |
\end{verbatim} |
269 |
as defined in \filelink{model/inc/DYNVARS.h}{model-inc-DYNVARS.h} |
270 |
|
271 |
The grid type is a character string that encodes the presence and |
272 |
types associated with the four possible dimensions. The character |
273 |
string follows the format |
274 |
\begin{center} |
275 |
\texttt{H0\_H1\_H2\_\_V\_\_T} |
276 |
\end{center} |
277 |
where the terms \textit{H0}, \textit{H1}, \textit{H2}, \textit{V}, |
278 |
\textit{T} can be almost any combination of the following: |
279 |
\begin{center} |
280 |
\begin{tabular}[h]{|ccc|c|c|}\hline |
281 |
\multicolumn{3}{|c|}{Horizontal} & Vertical & Time \\ |
282 |
\textbf{H0}: location & \textbf{H1}: dimensions & \textbf{H2}: halo |
283 |
& \textbf{V}: location & \textbf{T}: level \\\hline |
284 |
\texttt{-} & xy & Hn & \texttt{-} & \texttt{-} \\ |
285 |
U & x & Hy & i & t \\ |
286 |
V & y & & c & \\ |
287 |
Cen & & & & \\ |
288 |
Cor & & & & \\\hline |
289 |
\end{tabular} |
290 |
\end{center} |
291 |
A example list of all pre-defined combinations is contained in the |
292 |
file |
293 |
\begin{center} |
294 |
\texttt{pkg/mnc/pre-defined\_grids.txt}. |
295 |
\end{center} |
296 |
|
297 |
The variable type is an association between a variable type name and the |
298 |
following items: |
299 |
\begin{center} |
300 |
\begin{tabular}[h]{|l|l|}\hline |
301 |
\textbf{Item} & \textbf{Purpose} \\\hline |
302 |
grid type & defines the in-memory arrangement \\ |
303 |
\texttt{bi,bj} dimensions & tiling indices, if present \\\hline |
304 |
\end{tabular} |
305 |
\end{center} |
306 |
and is used by the \texttt{mnc\_cw\_*\_[R|W]} subroutines for reading |
307 |
and writing variables. |
308 |
|
309 |
|
310 |
\subsubsection{Using MNC: Examples} |
311 |
|
312 |
Writing variables to NetCDF files can be accomplished in as few as two |
313 |
function calls. The first function call defines a variable type, |
314 |
associates it with a name (character string), and provides additional |
315 |
information about the indicies for the tile (\texttt{bi},\texttt{bj}) |
316 |
dimensions. The second function call will write the data at, if |
317 |
necessary, the current time level within the model. |
318 |
|
319 |
Examples of the initialization calls can be found in the file |
320 |
\filelink{model/src/ini\_mnc\_io.F}{model-src-ini_mnc_io.F} |
321 |
where these function calls: |
322 |
{\footnotesize |
323 |
\begin{verbatim} |
324 |
C Create MNC definitions for DYNVARS.h variables |
325 |
CALL MNC_CW_ADD_VNAME('iter', '-_-_--__-__t', 0,0, myThid) |
326 |
CALL MNC_CW_ADD_VATTR_TEXT('iter',1, |
327 |
& 'long_name','iteration_count', myThid) |
328 |
|
329 |
CALL MNC_CW_ADD_VNAME('model_time', '-_-_--__-__t', 0,0, myThid) |
330 |
CALL MNC_CW_ADD_VATTR_TEXT('model_time',1, |
331 |
& 'long_name','Model Time', myThid) |
332 |
CALL MNC_CW_ADD_VATTR_TEXT('model_time',1,'units','s', myThid) |
333 |
|
334 |
CALL MNC_CW_ADD_VNAME('U', 'U_xy_Hn__C__t', 4,5, myThid) |
335 |
CALL MNC_CW_ADD_VATTR_TEXT('U',1,'units','m/s', myThid) |
336 |
CALL MNC_CW_ADD_VATTR_TEXT('U',1, |
337 |
& 'coordinates','XU YU RC iter', myThid) |
338 |
|
339 |
CALL MNC_CW_ADD_VNAME('T', 'Cen_xy_Hn__C__t', 4,5, myThid) |
340 |
CALL MNC_CW_ADD_VATTR_TEXT('T',1,'units','degC', myThid) |
341 |
CALL MNC_CW_ADD_VATTR_TEXT('T',1,'long_name', |
342 |
& 'potential_temperature', myThid) |
343 |
CALL MNC_CW_ADD_VATTR_TEXT('T',1, |
344 |
& 'coordinates','XC YC RC iter', myThid) |
345 |
\end{verbatim} |
346 |
} |
347 |
{\noindent initialize four \texttt{VNAME}s and add one or more NetCDF |
348 |
attributes to each.} |
349 |
|
350 |
The four variables defined above are subsequently written at specific |
351 |
time steps within |
352 |
\filelink{model/src/write\_state.F}{model-src-write_state.F} |
353 |
using the function calls: |
354 |
{\footnotesize |
355 |
\begin{verbatim} |
356 |
C Write dynvars using the MNC package |
357 |
CALL MNC_CW_SET_UDIM('state', -1, myThid) |
358 |
CALL MNC_CW_I_W('I','state',0,0,'iter', myIter, myThid) |
359 |
CALL MNC_CW_SET_UDIM('state', 0, myThid) |
360 |
CALL MNC_CW_RL_W('D','state',0,0,'model_time',myTime, myThid) |
361 |
CALL MNC_CW_RL_W('D','state',0,0,'U', uVel, myThid) |
362 |
CALL MNC_CW_RL_W('D','state',0,0,'T', theta, myThid) |
363 |
\end{verbatim} |
364 |
} |
365 |
|
366 |
While it is easiest to write variables within typical 2D and 3D fields |
367 |
where all data is known at a given time, it is also possible to write |
368 |
fields where only a portion (\textit{eg.} a ``slab'' or ``slice'') is |
369 |
known at a given instant. An example is provided within |
370 |
\filelink{pkg/mom\_vecinv/mom\_vecinv.F}{pkg-mom_vecinv-mom_vecinv.F} |
371 |
where an offset vector is used: {\footnotesize |
372 |
\begin{verbatim} |
373 |
IF (useMNC .AND. snapshot_mnc) THEN |
374 |
CALL MNC_CW_RL_W_OFFSET('D','mom_vi',bi,bj, 'fV', uCf, |
375 |
& offsets, myThid) |
376 |
CALL MNC_CW_RL_W_OFFSET('D','mom_vi',bi,bj, 'fU', vCf, |
377 |
& offsets, myThid) |
378 |
ENDIF |
379 |
\end{verbatim} |
380 |
} |
381 |
to write a 3D field one depth slice at a time. |
382 |
|
383 |
Each element in the offset vector corresponds (in order) to the |
384 |
dimensions of the ``full'' (or virtual) array and specifies which are |
385 |
known at the time of the call. A zero within the offset array means |
386 |
that all values along that dimension are available while a positive |
387 |
integer means that only values along that index of the dimension are |
388 |
available. In all cases, the matrix passed is assumed to start (that |
389 |
is, have an in-memory structure) coinciding with the start of the |
390 |
specified slice. Thus, using this offset array mechanism, a slice |
391 |
can be written along any single dimension or combinations of |
392 |
dimensions. |
393 |
|