Debian Codegraph @

packages/seacas/applications/nem_slice/nem_slice.1

raw
1 C Copyright(C) 1999-2020 National Technology & Engineering Solutions
2 C of Sandia, LLC (NTESS). Under the terms of Contract DE-NA0003525 with
3 C NTESS, the U.S. Government retains certain rights in this software.
4 C
5 C See packages/seacas/LICENSE for details
6
7 .TH NEM_SLICE 1 LOCAL
8 .SH NAME
9 nem_slice - generate a load-balance file from an ExodusII geometry file
10 .SH SYNOPSIS
11 .B nem_slice
12 [
13 .B -h
14 ] [
15 .B -V
16 ] [
17 .B -a
18 .I inpfile
19 ] [ <
20 .B -n
21 |
22 .B -e
23 >
24 .B -v
25 .B -m
26 .I machine_desc
27 .B -l
28 .I loadbal_desc
29 .B -s
30 .I eigensolv_desc
31 ] [
32 .B -w
33 .I weighting_desc
34 ] [
35 .B -g
36 .I group_list
37 ] [
38 .B -f
39 ] [
40 .B -c
41 ] [
42 .B -o
43 .I outfile
44 ] [
45 .I exoIIfile
46 ]
47 .SH DESCRIPTION
48 .I nem_slice
49 reads in a FEM description of the geometry of a problem from an
50 .I ExodusII
51 file,
52 .I exoIIfile
53 , generates either a nodal or elemental graph of the problem,
54 calls
55 .I Chaco
56 to load balance the graph, and outputs a
57 .I NemesisI
58 load-balance file.
59 .SH OPTIONS
60 The
61 .B -h
62 option causes
63 .I nem_slice
64 to print out usage information. If the
65 .B -h
66 option is specified all other options are ignored.
67 .PP
68 The
69 .B -V
70 option causes
71 .I nem_slice
72 to print out the version number. If the
73 .B -V
74 option is specified all other options are ignored.
75 .PP
76 If the
77 .B -a
78 option is specified all parameters will be set from an ASCII input
79 file. Other command line options may be used with
80 .B -a
81 and such options will override anything respecified in the ASCII input
82 file.
83 .PP
84 The
85 .B -n
86 option specifies that a nodal decomposition should be performed on
87 the input FEM mesh, while the
88 .B -e
89 option specifies that an elemental decomposition should be
90 performed. One of either
91 .B -e
92 or
93 .B -n
94 is required.
95 .PP
96 The
97 .B -v
98 option is used to indicate that you wish to output an ExodusII file
99 which will allow you to visualize the load balance results. If the
100 mesh consists of one type of element then the visualization is done by
101 element blocks. If the mesh consists of mixed element types then the
102 visualization is done by assigning nodal results variables to each
103 FEM node in the mesh, the value of which corresponds to the processor
104 ID that node is assigned to.
105 .PP
106 The
107 .B -m
108 option is used to describe the parallel architecture for which the
109 load balance will be performed. There are currently three choices for
110 this option,
111 .I mesh
112 for an n-dimensional mesh architecture,
113 .I hcube
114 for a hypercube architecture, or
115 .I cluster
116 for a cluster architecture. All machine type specifications require
117 a sub-option giving the dimension(s) of the machine, and thus the
118 number of processors, for which the load balance is to be performed.
119 .PP
120 The
121 .B -l
122 option is used to specify which method to utilize in generating a load
123 balance for the given problem. In addition, a sub-option of
124 .B kl
125 can be specified if Kernighan-Lin refinement is desired and the main
126 method does not use KL by default. Another sub-option is the number of
127 sections to use (see the Chaco User's Guide for further
128 information). This can be specified with the sub-option,
129 .B num_sects=value.
130 An additional sub-option requires that the load balanced domains
131 are connected. This can be specified with the sub-option,
132 .B cnctd_dom.
133 Supported methods for load balancing are:
134 .PP
135 .B multikl
136 .br
137 .B spectral
138 .br
139 .B inertial
140 .br
141 .B linear
142 .br
143 .B random
144 .br
145 .B scattered
146 .br
147 .B zpinch
148 .br
149 .B infile=filename
150 .br
151 The last option,
152 .B infile,
153 reads the assignment vector from and input file. When using this option,
154 no other load balancing is performed. In addition, the is a sub-option,
155 .B outfile=filename,
156 that will write out the assignment vector after the load balancing is
157 complete.
158 .PP
159 The
160 .B -s
161 option is used to specify parameters for the eigensolver, if the
162 method utilizes an eigensolver. Parameters which can be specified are:
163 .PP
164 .B tolerance=value
165 .br
166 .B use_rqi
167 .br
168 .B vmax=value
169 .PP
170 The
171 .B -o
172 option gives the name of the
173 .I NemesisI
174 output file. By default the output file name is generated from the
175 input file name with different additions depending on the method being
176 used for load balancing and the number of processors for which the
177 load balance was generated.
178 .PP
179 The
180 .B -w
181 option is used to specify how to weight the graph of the problem once
182 it is generated. Currently
183 .I read
184 ,
185 .I eb
186 , and
187 .I edges
188 sub-options are supported. The
189 .I read
190 sub-option allows the reading in of weights from an ExodusII file.
191 Nodal values are read for nodal decompositions and elemental values
192 for elemental decompositions. Along with the
193 .I read
194 sub-option other sub-options which can be specified are the name of the
195 variable, either nodal or elemental depending on the method requested,
196 as it exists in the ExodusII input file. This variable can either be
197 specified via it's name with the
198 .I var_name
199 sub-option. Or it can be specified with it's index via the
200 .I var_index
201 sub-option.
202 In addition the time index from which to read the given variable can
203 also be specified with the
204 .I time_index
205 sub-option. The
206 .I eb
207 sub-option is used to give a specific weight to a given element block.
208 The format to give block
209 .I n
210 a weight of
211 .I m
212 is
213 .br
214 eb=n:m
215 .br
216 If the sub-option
217 .I eb
218 comes after the
219 .I read
220 sub-option in the weighting specifications, then the value for the
221 element block overwrites the values read in from the ExodusII file.
222 If the
223 .I eb
224 sub-option comes before the
225 .I read
226 sub-option, then it is ignored. The
227 .I edges
228 sub-option is used to turn on edge weights for an elemental decomposition.
229 The default is that edge weights are not calculated. This sub-option is
230 ignored for a nodal decomposition.
231 .PP
232 The
233 .B -g
234 option allows the user to break up the element blocks into groups
235 that will be passed separately to Chaco. If the group descriptor
236 has white spaces in it, then it must be enclosed in quotes (").
237 The following rules apply to the descriptor:
238 .br
239 - Blocks are grouped using the slash "/" character.
240 .br
241 - Ids are separated with white space, comma, or by the hyphen "-"
242 character. The hyphen, "-", is used to include all element blocks between
243 the two given ID's.
244 .br
245 - Any blocks not included in the list, are added to a separate group.
246 .br
247 - Duplicates in the list are permitted, but the last group to which a
248 block is placed is where the block will go.
249 .br
250 - Block IDs not in the exodus file are quietly ignored.
251 .PP
252 The
253 .B -c
254 option turns off some of the error checking that nem_slice does while
255 finding elemental communication maps.
256 .PP
257 .SH INPUT FILE FORMAT
258 The optional ASCII input file closely mimics the command line
259 options. The file consists of a sequence of keys, each with a tab or
260 ``='' separated value. The order of the keys is not significant and a
261 line beginning with a ``#'' is considered to be a comment. The
262 following keys are recognized by
263 .I nem_slice
264 :
265 .PP
266 Input ExodusII File
267 .br
268 Output NemesisI File
269 .br
270 Graph Type
271 .br
272 Decomposition Method
273 .br
274 Solver Specifications
275 .br
276 Machine Description
277 .br
278 Weighting Specifications
279 .br
280 Misc Options
281 .PP
282 The case of the words forming a key is not significant.
283 .SH COMMAND LINE EXAMPLES
284 To obtain a nodal load balance for a 2D mesh parallel computer with
285 processors arrayed in a 10x20 grid using multi-level decomposition:
286 .PP
287 .B nem_slice -n -m mesh=10x20 -l multikl geom.exoII
288 .PP
289 For the same problem to be run on a 5 dimensional hypercube use:
290 .PP
291 .B nem_slice -n -m hcube=5 -l multikl geom.exoII
292 .PP
293 To generate an elemental based load balance using a spectral method,
294 with KL refinement, for a 3D mesh architecture, with the RQI
295 eigensolver and quadrisection:
296 .PP
297 .B nem_slice -e -m mesh=10x5x2 -l spectral,kl,num_sects=1 -s use_rqi
298 .PP
299 To generate a nodal based load balance with Inertial+KL with weights
300 read from nodal results contained in an ExodusII file use:
301 .PP
302 .B nem_slice -n -m mesh=10x5x2 -l inertial,kl -w read=weights.exoII,vindx=1,time=2
303 .PP
304 .SH ASCII INPUT FILE EXAMPLE
305 (See the file "input-ldbl" distributed with the executable)
306 .SH SEE ALSO
307 .B The Chaco User's Guide Version 1.0;
308 B. Hendrickson, R. Leland; Sandia Report SAND93-2339, Nov. 1993
309 .PP
310 .B NemesisI: A Set of Functions for Describing Unstructured Finite
311 .B Element Data on Parallel Computers;
312 G. Hennigan, J. Shadid
313 .SH NOTES
314 .B nem_slice
315 attempts to be smart about parsing command line, or ASCII input file,
316 options. However, some errors may sneak through, in which case
317 .B Chaco
318 is relied upon for error detection.
319 .SH AUTHORS
320 Gary L. Hennigan, Sandia National Labs, Dept. 9221
321 .br
322 Matthew M. St. John, Sandia National Labs, Dept. 9221
323 .br
324 John N. Shadid, Sandia National Labs, Dept. 9221