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
|
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2 Final//EN">
<!--Converted with LaTeX2HTML 2019.2 (Released June 5, 2019) -->
<HTML lang="EN">
<HEAD>
<TITLE>5 Parallelism</TITLE>
<META NAME="description" CONTENT="5 Parallelism">
<META NAME="keywords" CONTENT="user_guide">
<META NAME="resource-type" CONTENT="document">
<META NAME="distribution" CONTENT="global">
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=utf-8">
<META NAME="viewport" CONTENT="width=device-width, initial-scale=1.0">
<META NAME="Generator" CONTENT="LaTeX2HTML v2019.2">
<LINK REL="STYLESHEET" HREF="user_guide.css">
<LINK REL="next" HREF="node17.html">
<LINK REL="previous" HREF="node6.html">
<LINK REL="next" HREF="node17.html">
</HEAD>
<BODY >
<!--Navigation Panel-->
<A
HREF="node17.html">
<IMG WIDTH="37" HEIGHT="24" ALT="next" SRC="next.png"></A>
<A
HREF="user_guide.html">
<IMG WIDTH="26" HEIGHT="24" ALT="up" SRC="up.png"></A>
<A
HREF="node15.html">
<IMG WIDTH="63" HEIGHT="24" ALT="previous" SRC="prev.png"></A>
<A ID="tex2html85"
HREF="node1.html">
<IMG WIDTH="65" HEIGHT="24" ALT="contents" SRC="contents.png"></A>
<BR>
<B> Next:</B> <A
HREF="node17.html">6 Troubleshooting</A>
<B> Up:</B> <A
HREF="user_guide.html">User's Guide for the</A>
<B> Previous:</B> <A
HREF="node15.html">4.9 Calculation of phonon-renormalization</A>
<B> <A ID="tex2html86"
HREF="node1.html">Contents</A></B>
<BR>
<BR>
<!--End of Navigation Panel-->
<H1><A ID="SECTION00060000000000000000"></A>
<A ID="Sec:para"></A>
<BR>
5 Parallelism
</H1>
<P>
We refer to the corresponding section of the <TT>PWscf</TT> guide for
an explanation of how parallelism works.
<P>
<TT>ph.x</TT> may take advantage of MPI parallelization on images, plane waves (PW)
and on <B>k</B>-points (``pools''). Currently all other MPI and explicit
OpenMP parallelizations have very limited to nonexistent implementation.
<TT>phcg.x</TT> implements only PW parallelization.
All other codes may be launched in parallel, but will execute
on a single processor.
<P>
In ``image'' parallelization, processors can be divided into different
``images", corresponding to one (or more than one) ``irrep'' or <B>q</B>
vectors. Images are loosely coupled: processors communicate
between different images only once in a while, so image parallelization
is suitable for cheap communication hardware (e.g. Gigabit Ethernet).
Image parallelization is activated by specifying the option
<TT>-nimage N</TT> to <TT>ph.x</TT>. Inside an image, PW and <B>k</B>-point
parallelization can be performed: for instance,
<PRE>
mpirun -np 64 ph.x -ni 8 -nk 2 ...
</PRE>
will run 8 images on 8 processors each, subdivided into 2 pools
of 4 processors for <B>k</B>-point parallelization. In order
to run the <TT>ph.x</TT> code with these flags the <TT>pw.x</TT> run has to be run with:
<PRE>
mpirun -np 8 pw.x -nk 2 ...
</PRE>
without any <TT>-nimage</TT> flag.
After the phonon calculation with images the dynmical matrices of
<B>q</B>-vectors calculated in different images are not present in the
working directory. To obtain them you need to run
<TT>ph.x</TT> again with:
<PRE>
mpirun -np 8 ph.x -nk 2 ...
</PRE>
and the <TT>recover=.true.</TT> flag. This scheme is quite automatic and
does not require any additional work by the user, but it wastes some
CPU time because all images stops when the image that requires the
largest amount of time finishes the calculation. Load balancing
between images is still at
an experimental stage. You can look into the routine <TT>image_q_irr</TT>
inside <TT>PHonon/PH/check_initial_status</TT> to see the present
algorithm for work distribution and modify it if you think that
you can improve the load balancing.
<P>
A different paradigm is the usage of the GRID concept, instead of MPI,
to achieve parallelization over irreps and <B>q</B> vectors.
Complete phonon dispersion calculation can be quite long and
expensive, but it can be split into a number of semi-independent
calculations, using options <TT>start_q</TT>, <TT>last_q</TT>,
<TT>start_irr</TT>, <TT>last_irr</TT>. An example on how to
distribute the calculations and collect the results can be found
in <TT>examples/GRID_example</TT>. Reference:
<BR><I>Calculation of Phonon Dispersions on the GRID using Quantum
ESPRESSO</I>,
R. di Meo, A. Dal Corso, P. Giannozzi, and S. Cozzini, in
<I>Chemistry and Material Science Applications on Grid Infrastructures</I>,
editors: S. Cozzini, A. Laganà , ICTP Lecture Notes Series,
Vol. 24, pp.165-183 (2009).
<P>
<HR>
<!--Navigation Panel-->
<A
HREF="node17.html">
<IMG WIDTH="37" HEIGHT="24" ALT="next" SRC="next.png"></A>
<A
HREF="user_guide.html">
<IMG WIDTH="26" HEIGHT="24" ALT="up" SRC="up.png"></A>
<A
HREF="node15.html">
<IMG WIDTH="63" HEIGHT="24" ALT="previous" SRC="prev.png"></A>
<A ID="tex2html85"
HREF="node1.html">
<IMG WIDTH="65" HEIGHT="24" ALT="contents" SRC="contents.png"></A>
<BR>
<B> Next:</B> <A
HREF="node17.html">6 Troubleshooting</A>
<B> Up:</B> <A
HREF="user_guide.html">User's Guide for the</A>
<B> Previous:</B> <A
HREF="node15.html">4.9 Calculation of phonon-renormalization</A>
<B> <A ID="tex2html86"
HREF="node1.html">Contents</A></B>
<!--End of Navigation Panel-->
</BODY>
</HTML>
|
