{
 "metadata": {
  "name": "9932_05_03"
 },
 "nbformat": 3,
 "nbformat_minor": 0,
 "worksheets": [
  {
   "cells": [
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Chapter 5, example 3\n",
      "====================\n",
      "\n",
      "In this example we show how to use MPI from IPython."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "from IPython.parallel import Client"
     ],
     "language": "python",
     "metadata": {},
     "outputs": []
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "We first need to create a MPI profile with:\n",
      "\n",
      "    ipython profile create --parallel --profile=mpi\n",
      "\n",
      "Then specify the following in the cluster configuration file:\n",
      "\n",
      "    c.IPClusterEngines.engine_launcher_class = 'MPIEngineSetLauncher'\n",
      "\n",
      "and run the engines with:\n",
      "\n",
      "    ipcluster start -n 4 --profile=mpi\n",
      "\n",
      "Now, we can create a MPI client."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "c = Client(profile='mpi')"
     ],
     "language": "python",
     "metadata": {},
     "outputs": []
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "view = c[:]"
     ],
     "language": "python",
     "metadata": {},
     "outputs": []
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "view.activate() # enable magics"
     ],
     "language": "python",
     "metadata": {},
     "outputs": []
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "We will compute the sum of all integers between 1 and 16 in parallel over four cores. The script `psum.py` contains the following code:\n",
      "\n",
      "    from mpi4py import MPI\n",
      "    import numpy as np\n",
      "    \n",
      "    # This function will be executed on all processes.\n",
      "    def psum(a):\n",
      "        # \"a\" only contains a subset of all integers.\n",
      "        # They are summed locally on this process. \n",
      "        locsum = np.sum(a)\n",
      "    \n",
      "        # We allocate a variable that will contain the final result, that is,\n",
      "        # the sum of all our integers.\n",
      "        rcvBuf = np.array(0.0,'d')\n",
      "    \n",
      "        # We use a MPI reduce operation:\n",
      "        #   * locsum is combined from all processes\n",
      "        #   * these local sums are summed with the MPI.SUM operation\n",
      "        #   * the result (total sum) is distributed back to all processes in\n",
      "        #     the rcvBuf variable\n",
      "        MPI.COMM_WORLD.Allreduce([locsum, MPI.DOUBLE],\n",
      "            [rcvBuf, MPI.DOUBLE],\n",
      "            op=MPI.SUM)\n",
      "        return rcvBuf"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "We execute this script on all engines so that the function is available everywhere."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "view.run('psum.py') # the script runs on all processes"
     ],
     "language": "python",
     "metadata": {},
     "outputs": []
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Now we distribute the array with the 16 values across the engines (each engine gets a subarray)."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "view.scatter('a', np.arange(16)) # this array is scattered across processes"
     ],
     "language": "python",
     "metadata": {},
     "outputs": []
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "We compute the total sum in parallel using MPI."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "%px totalsum = psum(a) # psum is executed on all processes"
     ],
     "language": "python",
     "metadata": {},
     "outputs": []
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Finally, the result is available on all engines thanks to the MPI `Allreduce` function."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "view['totalsum']"
     ],
     "language": "python",
     "metadata": {},
     "outputs": []
    }
   ],
   "metadata": {}
  }
 ]
}