1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
|
<?xml version='1.0' encoding="UTF-8"?>
<!DOCTYPE guide SYSTEM "/dtd/guide.dtd">
<guide link="/proj/en/cluster/demo-lwe.xml">
<title>Gentoo Cluster Demo</title>
<author title="Author"><mail
link="jean-francois@adelielinux.com">Jean-Francois
Richard</mail></author>
<abstract>This document was written by people at the <uri
link="http://www.adelielinux.com">Adelie Linux R&D Center</uri>,
it presents the demonstration Gentoo cluster for the Linux World
Expo.</abstract>
<version>1.0</version>
<date>August 5, 2003</date>
<chapter><title>Hardware Platforms</title>
<section>
<body>
<p>Here are the hardware platforms accessible for the demo :</p>
<ul>
<li><p>Galapagos: 5 node cluster featuring gigabit connectivity.</p>
<ul>
<li><p>Master node : AMD Duron(tm) : 1000 MHz</p></li>
<li><p>4 slave nodes : AMD Athlon(tm) : 1250 MHz</p></li>
</ul>
</li>
</ul>
</body>
</section>
</chapter>
<chapter><title>Demo on Galapagos</title>
<section><body>
<p>We installed the following software on Galapagos :</p>
<ul>
<li><p>POVRAY</p></li>
<li><p>MPI-POVRAY</p></li>
<li><p>PVM-POVRAY</p></li>
</ul>
<p>These are raytracing programs that run either on a single cpu or on
the whole cluster, using MPI or PVM for message passing.</p>
<p>An account named "demo@galapagos.cyberlogic.net" was created. In
it, you can find a readme file and several .pov files to render.</p>
<p>You will need a VNC viewer to connect to the demo, with an SSH
tunnel that you can setup this way : </p>
<pre caption="SSH tunnel setup">
ssh -C -g -L 5901:localhost:5901 demo@galapagos.cyberlogic.net
</pre>
<p>Once this is done, you can connect to your
<path>localhost:5901</path> with a VNC viewer, like KDE's "Remote
Desktop Connection" or "vncviewer".</p>
</body></section>
<section>
<title>Sample POVRAY commands</title>
<body>
<p>You can use the following commands from <path>~demo/</path>
to test the different POVRAYs installed :</p>
<p>We suggest running the vanilla POVRAY side-by-side with either the
MPI or PVM version to show the speed difference. Just start two
terminals and run both with the following commands.</p>
<note>You can disable the X display by removing the "+D0 +D1 +p"
options.</note>
<note>You can resize the output by changing the "+wWIDTH +hHEIGHT"
parameters.</note>
<pre caption="POVRAY">
$ x-povray -i blue.pov +v1 -d +ft -x \
+a0.300 +r3 -q9 -w640 -h480 -mv2.0 \
+b1000 +D0 +D1 +p
</pre>
<p>This task takes 48 seconds.</p>
<pre caption="PVM-POVRAY">
$ pvm /etc/pvm_hosts
pvm> conf should show node01-node04
pvm> quit
$ ${PVM_ROOT}/bin/LINUX/x-pvmpov \
-i blue.pov \
+v1 -d +ft -x +a0.300 +r3 -q9 -w640 \
-h480 -mv2.0 +b1000 +D0 +D1 +p
Exit PVM
$ pvm
pvm> halt
</pre>
<p>This task takes 11 seconds.</p>
<pre caption="MPI-POVRAY">
$ lamboot /etc/lam-mpi/lam-bhost.lam
$ mpirun -np 5 `which mpi-x-povray` \
-i blue.pov +v1 -d +ft -x +a0.300 +r3 \
-q9 -w640 -h480 -mv2.0 +b1000 +D0 +D1 +p
$ lamhalt
</pre>
<note>You can change the number of CPUs used to compute the image by
changing the "-np X" option.</note>
<p>This task takes 9 seconds with "-np 5", 13 seconds with "-np 4", 20
seconds with "-np 3", 46 seconds with "-np 2". The following chart
shows these results :</p>
<figure link="demo-lwe.png" short="MPI-POVRAY Results"
caption="MPI-POVRAY Results, showing a relation between the number of
nodes and the total processing time. The red line shows the
non-parallel version of POVRAY."/>
<p>The above chart shows there is no significant gain between the 1
and 2 CPUs dots. This difference is mostly caused by the
communication cost in the MPI version and the implementation
differences between POVRAY and MPI-POVRAY. The comparison between
MPI-POVRAY and POVRAY cannot yield truly valid conclusions.</p>
<p>However, there is a a significant relation between the number of
CPUs and the overall speed, as predicted. Note that the speed gain is
not linear and the speedup tends to zero when adding a lot of nodes.
This depends, again, on implementation and on the parallel algorithms
used.</p>
<p>Several scientific and industrial applications are designed to
scale to a lot more nodes while showing an almost-linear behaviour
regarding the speed gain vs. number of nodes. When using these
applications, clusters are a really interesting and cost-effective
solution to add processing power.</p>
</body>
</section>
</chapter>
<chapter><title>About this Document</title>
<section><body>
<p>The original document is published at the <uri
link="http://www.adelielinux.com">Adelie Linux R&D Centre</uri>
web site, and is reproduced here with the permission of the authors
and <uri link="http://www.cyberlogic.ca">Cyberlogic</uri>'s Adelie
Linux R&D Centre.</p>
</body></section></chapter>
</guide>
|
GitWeb