1<?xml version="1.0" encoding="utf-8" ?> 2<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"> 3<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en" lang="en"> 4<head> 5<meta http-equiv="Content-Type" content="text/html; charset=utf-8" /> 6<meta name="generator" content="Docutils 0.6: http://docutils.sourceforge.net/" /> 7<title>Parallel BGL MPI BSP Process Group</title> 8<link rel="stylesheet" href="../../../../rst.css" type="text/css" /> 9</head> 10<body> 11<div class="document" id="logo-mpi-bsp-process-group"> 12<h1 class="title"><a class="reference external" href="http://www.osl.iu.edu/research/pbgl"><img align="middle" alt="Parallel BGL" class="align-middle" src="pbgl-logo.png" /></a> MPI BSP Process Group</h1> 13 14<!-- Copyright (C) 2004-2009 The Trustees of Indiana University. 15Use, modification and distribution is subject to the Boost Software 16License, Version 1.0. (See accompanying file LICENSE_1_0.txt or copy at 17http://www.boost.org/LICENSE_1_0.txt) --> 18<div class="contents topic" id="contents"> 19<p class="topic-title first">Contents</p> 20<ul class="simple"> 21<li><a class="reference internal" href="#introduction" id="id1">Introduction</a></li> 22<li><a class="reference internal" href="#where-defined" id="id2">Where Defined</a></li> 23<li><a class="reference internal" href="#reference" id="id3">Reference</a></li> 24</ul> 25</div> 26<div class="section" id="introduction"> 27<h1><a class="toc-backref" href="#id1">Introduction</a></h1> 28<p>The MPI <tt class="docutils literal"><span class="pre">mpi_process_group</span></tt> is an implementation of the <a class="reference external" href="process_group.html">process 29group</a> interface using the Message Passing Interface (MPI). It is the 30primary process group used in the Parallel BGL at this time.</p> 31</div> 32<div class="section" id="where-defined"> 33<h1><a class="toc-backref" href="#id2">Where Defined</a></h1> 34<p>Header <tt class="docutils literal"><span class="pre"><boost/graph/distributed/mpi_process_group.hpp></span></tt></p> 35</div> 36<div class="section" id="reference"> 37<h1><a class="toc-backref" href="#id3">Reference</a></h1> 38<pre class="literal-block"> 39namespace boost { namespace graph { namespace distributed { 40 41class mpi_process_group 42{ 43public: 44 typedef boost::mpi::communicator communicator_type; 45 46 // Process group constructors 47 mpi_process_group(communicator_type comm = communicator_type()); 48 mpi_process_group(std::size_t num_headers, std::size_t buffer_size, 49 communicator_type comm = communicator_type()); 50 51 mpi_process_group(); 52 mpi_process_group(const mpi_process_group&, boost::parallel::attach_distributed_object); 53 54 // Triggers 55 template<typename Type, typename Handler> 56 void trigger(int tag, const Handler& handler); 57 58 template<typename Type, typename Handler> 59 void trigger_with_reply(int tag, const Handler& handler); 60 61 trigger_receive_context trigger_context() const; 62 63 // Helper operations 64 void poll(); 65 mpi_process_group base() const; 66}; 67 68// Process query 69int process_id(const mpi_process_group&); 70int num_processes(const mpi_process_group&); 71 72// Message transmission 73template<typename T> 74 void send(const mpi_process_group& pg, int dest, int tag, const T& value); 75 76template<typename T> 77 void receive(const mpi_process_group& pg, int source, int tag, T& value); 78 79optional<std::pair<int, int> > probe(const mpi_process_group& pg); 80 81// Synchronization 82void synchronize(const mpi_process_group& pg); 83 84// Out-of-band communication 85template<typename T> 86 void send_oob(const mpi_process_group& pg, int dest, int tag, const T& value); 87 88template<typename T, typename U> 89 void 90 send_oob_with_reply(const mpi_process_group& pg, int dest, int 91 tag, const T& send_value, U& receive_value); 92 93template<typename T> 94 void receive_oob(const mpi_process_group& pg, int source, int tag, T& value); 95 96} } } 97</pre> 98<p>Since the <tt class="docutils literal"><span class="pre">mpi_process_group</span></tt> is an implementation of the <a class="reference external" href="process_group.html">process 99group</a> interface, we omit the description of most of the functions in 100the prototype. Two constructors need special mentioning:</p> 101<pre class="literal-block"> 102mpi_process_group(communicator_type comm = communicator_type()); 103</pre> 104<p>The constructor can take an optional MPI communicator. As default a communicator 105constructed from MPI_COMM_WORLD is used.</p> 106<pre class="literal-block"> 107mpi_process_group(std::size_t num_headers, std::size_t buffer_size, 108 communicator_type comm = communicator_type()); 109</pre> 110<p>For performance fine tuning the maximum number of headers in a message batch 111(num_headers) and the maximum combined size of batched messages (buffer_size) 112can be specified. The maximum message size of a batch is 11316*num_headers+buffer_size. Sensible default values have been found by optimizing 114a typical application on a cluster with Ethernet network, and are num_header=64k 115and buffer_size=1MB, for a total maximum batches message size of 2MB.</p> 116<hr class="docutils" /> 117<p>Copyright (C) 2007 Douglas Gregor</p> 118<p>Copyright (C) 2007 Matthias Troyer</p> 119</div> 120</div> 121<div class="footer"> 122<hr class="footer" /> 123Generated on: 2009-05-31 00:22 UTC. 124Generated by <a class="reference external" href="http://docutils.sourceforge.net/">Docutils</a> from <a class="reference external" href="http://docutils.sourceforge.net/rst.html">reStructuredText</a> source. 125 126</div> 127</body> 128</html> 129
