source: trunk/documents/UserDoc/UsersGuides/ForApplicationDeveloper/html/Examples/parallel.html@ 1208

Last change on this file since 1208 was 1208, checked in by garnier, 16 years ago

CVS update
File size: 8.1 KB
Line 
1<html>
2<head>
3<title>Parallel Geant4 (ParGeant4)</title>
4</head>
5
6<body>
7<table WIDTH="100%" >
8<tr>
9<TD><a href="../../../../Overview/html/index.html">
10
11</a><a href="index.html">
12<img SRC="../../../../resources/html/IconsGIF/Contents.gif" ALT="Contents" height=16 width=59></a>
13<a href="ExtendedCodes.html">
14<img SRC="../../../../resources/html/IconsGIF/Previous.gif" ALT="Previous" height=16 width=59></a>
15<img SRC="../../../../resources/html/IconsGIF/NextGR.gif" ALT="Next" height=16 width=59></td>
16
17<td ALIGN=RIGHT><b><font color="#238E23"><font size=-1>Geant4 User's Guide</font></font></b>
18<br><b><font color="#238E23"><font size=-1>For Application Developers</font></font></b>
19<br><b><font color="#238E23"><font size=-1>Examples</font></font></b></td>
20</tr>
21</table>
22
23<center>
24<p><b><font color="#238E23"><font size=+3>Parallel Geant4 (ParGeant4)</font></font></b>
25<address>by Gene Cooperman (gene@ccs.neu.edu)<br>
26 and Viet Ha Nguyen (vietha@ccs.neu.edu)
27</address>
28</center>
29<p>
30<hr ALIGN="Center" SIZE="7%">
31<h2>What is ParGeant4 ?</h2>
32 <a target="ext" href="http://www.ccs.neu.edu/home/gene/pargeant4.html">ParGeant4</a>
33 is a parallel version of Geant4 that implements event-level
34 parallelism to simulate separate events on remote processors.
35 Typical simulations demonstrate a nearly linear speedup in running
36 time as the number of remote processors increases. The needed
37 enhancements of Geant4 are included in the
38 examples/extended/parallel directory of the Geant4 distribution.
39</p>
40<p>
41<hr ALIGN="Center" SIZE="7%">
42<h2>Why is ParGeant4 useful?</h2>
43
44 When doing a large Geant4 simulation, one often wishes to run
45 on many processors to reduce the overall time. Traditionally,
46 this has been done by splitting the events into multiple groups,
47 and running Geant4 independently on each processor for its own
48 group of events. This requires restarting a run if a processor
49 goes down. It also requires saving the histogram files from
50 each run, and merging the files prior to using the analysis tool.
51 The human effort in this is considerable.
52<p>
53 ParGeant4 provides a much simpler mechanism. After setting up
54 ParGeant4 one links and runs the sequential Geant4 application
55 exactly as before, but additionally linking with some parallel libraries.
56 Upon execution, ParGeant4 on the console sends out events to
57 slave processes, collects all hits, and calls any analysis tool --
58 exactly as one would do in the sequential case.
59</p><p>
60 There is no need to split events into separate groups, track
61 whether one of the processors crashed, merge histogram files, etc.
62 If a slave processor crashes, ParGeant automatically re-sends the
63 events of that slave processor to a new slave processor for re-execution.
64</p>
65<p>
66<hr ALIGN="Center" SIZE="7%">
67<h2>What is the performance of ParGeant4?</h2>
68
69 As a rule of thumb, speedup will be nearly linear when each event
70 simulation lasts for at least several milliseconds.
71 ParGeant4 has been tested extensively on parallelizations of
72 examples/novice/N02 and of examples/advanced/underground_physics .
73 On N02, we see a speedup of 27 for 50 nodes and a speedup of
74 33 for 100 nodes. When using the <code>--aggregated-tasks=50</code>
75 option (see below)
76 the speedup improves to 35 for 50 nodes and 60 for 100 nodes.
77<p>
78 In tests of underground_physics, events are longer and we see nearly
79 linear speedup (94 times speedup with 100 nodes).
80</p>
81<p>
82<hr ALIGN="Center" SIZE="7%">
83<h2>Getting started </h2>
84
85 Detailed information is under extended/parallel/ParN02/docs/000README.
86 There are four steps:
87 <ol>
88 <li> Install <a target="ext" href="http://www.ccs.neu.edu/home/gene/topc.html">TOP-C</a>.</li>
89 <li> Compile ParN02 by running <TT>gmake</TT>.</li>
90 <li> Make sure the "procgroup" file is correct and copy it to
91 directory of the executable binary file (for example,
92 <TT>$G4BIN/Linux-g++</TT>).</li>
93 <li> Run the parallel binary program.</li>
94 </ol>
95</p>
96<p>
97<hr ALIGN="Center" SIZE="7%">
98<h2>What is involved in setting up ParGeant4?</h2>
99
100 To set up ParGeant4, one needs
101 <a target="ext" href="http://www.ccs.neu.edu/home/gene/topc.html">TOP-C</a> and
102 <a target="ext" href="http://www.ccs.neu.edu/home/gene/marshalgen.html">Marshalgen</a>
103 (free, open source software).
104 If one is parallelizing a new Geant4 application, one must then add/modify
105 approximately 20 lines of annotations (C++ comments
106 to indicate shallow vs. deep copying of pointers, etc.) in the .h files
107 for each hit type being defined
108 by the application. For details of the annotations, refer to the manual of the Marshalgen package.<br>
109
110 Finally, in the main routine of the application,
111 one replaces the call to the <TT>G4RunManager</TT> constructor by a call
112 to the <TT>ParRunManager</TT> constructor. (<TT>ParRunManger</TT> is a derived class
113 of <TT>G4RunManager</TT>.)</p>
114<p>
115 After this, one invokes the already provided
116 GNUMakefile (a slightly modified version of the Geant4 example GNUMakefile)
117 to create the parallel application. Finally, one writes a
118 "procgroup" file, which declares the names of the remote hosts
119 to use in the parallel computation. Optionally, one may also
120 specify filenames (e.g. slave1.out, slave2.out, ...) to store
121 the printout from each slave process. One then calls the
122 ParGeant4 binary exactly as one would call the Geant4 binary,
123 and the results appear as normal, only faster.
124</p>
125<p>
126<hr ALIGN="Center" SIZE="7%">
127<h2>Are there examples of using ParGeant4?</h2>
128 Yes. ParGeant4 includes parallelizations of
129 other examples from the Geant4 distribution. Specifically, ParGeant4
130 includes example parallelizations of novice/N02, novice/N04, and
131 advanced/underground_physics .
132</p>
133<p>
134<hr ALIGN="Center" SIZE="7%">
135<h2>What are some of the features of ParGeant4?</h2>
136 ParGeant4 includes all of the features of TOP-C. In particular,
137 after building a binary, "parMySimulation",
138 one might call:<p>
139</p><dl><dt>
140 ./parMySimulation --TOPC-help
141</dt><dd>
142 Display command options, and then exit.
143<p>
144</p></dd><dt>
145 ./parMySimulation --TOPC-trace=0 <geant4 arguments="">
146</geant4></dt><dd>
147 By default, ParGeant4 traces each time a new event is sent to
148 a task. This turns it off.
149<p>
150</p></dd><dt>
151 ./parMySimulation --TOPC-verbose=0 <geant4 arguments="">
152</geant4></dt><dd>
153 By default, ParGeant4 provides statistics indicating what
154 process was run, when it was run, what machine, the running
155 times and elapsed times of master and the average slave, and
156 other information. This turns off the statistics.
157<p>
158</p></dd><dt>
159 ./parMySimulation --TOPC-aggregated-tasks=10 <geant4 arguments="">
160</geant4></dt><dd>
161 By default, ParGeant4 sends one event to one remote process before
162 turning to the next process. This option sends 10 events
163 to a single remote process in one message. This is useful
164 when events are relatively short, and the network latency
165 of sending a message is starting to dominate the running time.
166<p>
167</p></dd><dt>
168 ./parMySimulation --TOPC-slave-timeout=3600 <geant4 arguments="">
169</geant4></dt><dd>
170 By default, if a remote process has not communicated with the
171 master after 1800 seconds (a half hour), the slave process will
172 kill itself. This prevents runaway processes that may be in an
173 infinite loop for some event, or may have lost their socket to
174 communicate with the master process. In this example, we allow
175 7200 seconds (two hours) because we expect simulation of some
176 events to last up to (but not more than) 7200 seconds.
177</dd></dl>
178<p>
179 By default, ParGeant4 uses its own subset implementation of MPI (MPINU).
180 ParGeant4 adds approximately 50 KB to the "footprint" of the binary
181 executable. By default, ParGeant4 uses "ssh" to set up remote processes.
182 Those who wish to use their own MPI (perhaps if a batch cluster requires
183 a specific MPI) may do so. (See "Configuring a Different `MPI')"
184 in the <a target="ext" href="http://www.ccs.neu.edu/home/gene/topc/topc_toc.html">
185 TOP-C manual</a>.)
186
187</p>
188<hr><i><a href="../../../../Authors/html/subjectsToAuthors.html">About
189the authors</a></i>
190</body>
191</html>
Note: See TracBrowser for help on using the repository browser.