Debian Codegraph @

GUI/PWgui/doc/pwdocs/user_guide/node21.html

raw
1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2 Final//EN">
2
3 <!--Converted with LaTeX2HTML 2019.2 (Released June 5, 2019) -->
4 <HTML lang="EN">
5 <HEAD>
6 <TITLE>3.4 Tricks and problems</TITLE>
7 <META NAME="description" CONTENT="3.4 Tricks and problems">
8 <META NAME="keywords" CONTENT="user_guide">
9 <META NAME="resource-type" CONTENT="document">
10 <META NAME="distribution" CONTENT="global">
11
12 <META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=utf-8">
13 <META NAME="viewport" CONTENT="width=device-width, initial-scale=1.0">
14 <META NAME="Generator" CONTENT="LaTeX2HTML v2019.2">
15
16 <LINK REL="STYLESHEET" HREF="user_guide.css">
17
18 <LINK REL="previous" HREF="node20.html">
19 <LINK REL="next" HREF="node22.html">
20 </HEAD>
21
22 <BODY >
23 <!--Navigation Panel-->
24 <A
25 HREF="node22.html">
26 <IMG WIDTH="37" HEIGHT="24" ALT="next" SRC="next.png"></A>
27 <A
28 HREF="node17.html">
29 <IMG WIDTH="26" HEIGHT="24" ALT="up" SRC="up.png"></A>
30 <A
31 HREF="node20.html">
32 <IMG WIDTH="63" HEIGHT="24" ALT="previous" SRC="prev.png"></A>
33 <A ID="tex2html231"
34 HREF="node1.html">
35 <IMG WIDTH="65" HEIGHT="24" ALT="contents" SRC="contents.png"></A>
36 <BR>
37 <B> Next:</B> <A
38 HREF="node22.html">About this document ...</A>
39 <B> Up:</B> <A
40 HREF="node17.html">3 Parallelism</A>
41 <B> Previous:</B> <A
42 HREF="node20.html">3.3 Parallelization levels</A>
43 <B> <A ID="tex2html232"
44 HREF="node1.html">Contents</A></B>
45 <BR>
46 <BR>
47 <!--End of Navigation Panel-->
48 <!--Table of Child-Links-->
49 <A ID="CHILD_LINKS"><STRONG>Subsections</STRONG></A>
50
51 <UL>
52 <LI><UL>
53 <LI><A ID="tex2html233"
54 HREF="node21.html#SECTION00044010000000000000">3.4.0.1 Trouble with input files</A>
55 <LI><A ID="tex2html234"
56 HREF="node21.html#SECTION00044020000000000000">3.4.0.2 Trouble with MKL and MPI parallelization</A>
57 <LI><A ID="tex2html235"
58 HREF="node21.html#SECTION00044030000000000000">3.4.0.3 Trouble with compilers and MPI libraries</A>
59 </UL></UL>
60 <!--End of Table of Child-Links-->
61 <HR>
62
63 <H2><A ID="SECTION00044000000000000000"></A>
64 <A ID="SubSec:badpara"></A>
65 <BR>
66 3.4 Tricks and problems
67 </H2>
68
69 <P>
70 Many problems in parallel execution derive from the mixup of different
71 MPI libraries and runtime environments. There are two major MPI
72 implementations, OpenMPI and MPICH, coming in various versions,
73 not necessarily compatible; plus vendor-specific implementations
74 (e.g. Intel MPI). A parallel machine may have multiple parallel
75 compilers (typically, <TT>mpif90</TT> scripts calling different
76 serial compilers), multiple MPI libraries, multiple launchers
77 for parallel codes (different versions of <TT>mpirun</TT> and/or
78 <TT>mpiexec</TT>). You have to figure out the proper combination
79 of all of the above, which may require using command <TT>module</TT>
80 or manually setting environment variables and execution paths.
81 What exactly has to be done depends upon the configuration of your
82 machine. You should inquire with your system administrator or user
83 support (if available; if not, YOU are the system administrator
84 and user support and YOU have to solve your problems).
85
86 <P>
87 Always verify if your executable is actually compiled for
88 parallel execution or not: it is declared in the first lines
89 of output. Running several instances of a serial code with
90 <TT>mpirun</TT> or <TT>mpiexec</TT> produces strange crashes.
91
92 <P>
93
94 <H4><A ID="SECTION00044010000000000000">
95 3.4.0.1 Trouble with input files</A>
96 </H4>
97 Some implementations of the MPI library have problems with input
98 redirection in parallel. This typically shows up under the form of
99 mysterious errors when reading data. If this happens, use the option
100 <TT>-i</TT> (or <TT>-in</TT>, <TT>-inp</TT>, <TT>-input</TT>),
101 followed by the input file name.
102 Example:
103 <PRE>
104 pw.x -i inputfile -nk 4 &gt; outputfile
105 </PRE>
106 Of course the
107 input file must be accessible by the processor that must read it
108 (only one processor reads the input file and subsequently broadcasts
109 its contents to all other processors).
110
111 <P>
112 Apparently the LSF implementation of MPI libraries manages to ignore or to
113 confuse even the <TT>-i/in/inp/input</TT> mechanism that is present in all
114 Q<SMALL>UANTUM </SMALL>ESPRESSO codes. In this case, use the <TT>-i</TT> option of <TT>mpirun.lsf</TT>
115 to provide an input file.
116
117 <P>
118
119 <H4><A ID="SECTION00044020000000000000">
120 3.4.0.2 Trouble with MKL and MPI parallelization</A>
121 </H4>
122 If you notice very bad parallel performances with MPI and MKL libraries,
123 it is very likely that the OpenMP parallelization performed by the latter
124 is colliding with MPI. Recent versions of MKL enable autoparallelization
125 by default on multicore machines. You must set the environment variable
126 OMP_NUM_THREADS to 1 to disable it.
127 Note that if for some reason the correct setting of variable
128 OMP_NUM_THREADS
129 does not propagate to all processors, you may equally run into trouble.
130 Lorenzo Paulatto (Nov. 2008) suggests to use the <TT>-x</TT> option to <TT>mpirun</TT> to
131 propagate OMP_NUM_THREADS to all processors.
132 Axel Kohlmeyer suggests the following (April 2008):
133 "(I've) found that Intel is now turning on multithreading without any
134 warning and that is for example why their FFT seems faster than
135 FFTW. For serial and OpenMP based runs this makes no difference (in
136 fact the multi-threaded FFT helps), but if you run MPI locally, you
137 actually lose performance. Also if you use the 'numactl' tool on linux
138 to bind a job to a specific cpu core, MKL will still try to use all
139 available cores (and slow down badly). The cleanest way of avoiding
140 this mess is to either link with
141 <BLOCKQUOTE>
142 <TT>-lmkl_intel_lp64 -lmkl_sequential -lmkl_core</TT> (on 64-bit:
143 x86_64, ia64)
144 <BR><TT>-lmkl_intel -lmkl_sequential -lmkl_core</TT> (on 32-bit, i.e. ia32 )
145
146 </BLOCKQUOTE>
147 or edit the <TT>libmkl_'platform'.a</TT> file. I'm using now a file
148 <TT>libmkl10.a</TT> with:
149 <PRE>
150 GROUP (libmkl_intel_lp64.a libmkl_sequential.a libmkl_core.a)
151 </PRE>
152 It works like a charm". UPDATE: Since v.4.2, <TT>configure</TT> links by
153 default MKL without multithreaded support.
154
155 <P>
156
157 <H4><A ID="SECTION00044030000000000000">
158 3.4.0.3 Trouble with compilers and MPI libraries</A>
159 </H4>
160 Many users of Q<SMALL>UANTUM </SMALL>ESPRESSO, in particular those working on PC clusters,
161 have to rely on themselves (or on less-than-adequate system managers) for
162 the correct configuration of software for parallel execution. Mysterious and
163 irreproducible crashes in parallel execution are sometimes due to bugs
164 in Q<SMALL>UANTUM </SMALL>ESPRESSO, but more often than not are a consequence of buggy
165 compilers or of buggy or miscompiled MPI libraries.
166
167 <P>
168 <HR>
169 <!--Navigation Panel-->
170 <A
171 HREF="node22.html">
172 <IMG WIDTH="37" HEIGHT="24" ALT="next" SRC="next.png"></A>
173 <A
174 HREF="node17.html">
175 <IMG WIDTH="26" HEIGHT="24" ALT="up" SRC="up.png"></A>
176 <A
177 HREF="node20.html">
178 <IMG WIDTH="63" HEIGHT="24" ALT="previous" SRC="prev.png"></A>
179 <A ID="tex2html231"
180 HREF="node1.html">
181 <IMG WIDTH="65" HEIGHT="24" ALT="contents" SRC="contents.png"></A>
182 <BR>
183 <B> Next:</B> <A
184 HREF="node22.html">About this document ...</A>
185 <B> Up:</B> <A
186 HREF="node17.html">3 Parallelism</A>
187 <B> Previous:</B> <A
188 HREF="node20.html">3.3 Parallelization levels</A>
189 <B> <A ID="tex2html232"
190 HREF="node1.html">Contents</A></B>
191 <!--End of Navigation Panel-->
192
193 </BODY>
194 </HTML>