X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/a95577c265a80a040b40683383ef495fd7c8d9dc..164e34e55e2c4633aac09c2706589fb3a0c40dcf:/examples/s4u/exec-ptask/s4u-exec-ptask.cpp diff --git a/examples/s4u/exec-ptask/s4u-exec-ptask.cpp b/examples/s4u/exec-ptask/s4u-exec-ptask.cpp index 7a9a31ff94..a8811184fe 100644 --- a/examples/s4u/exec-ptask/s4u-exec-ptask.cpp +++ b/examples/s4u/exec-ptask/s4u-exec-ptask.cpp @@ -1,11 +1,11 @@ -/* Copyright (c) 2017. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2017-2020. The SimGrid Team. All rights reserved. */ /* This program is free software; you can redistribute it and/or modify it * under the terms of the license (GNU LGPL) which comes with this package. */ -/* Parallel tasks are convenient abstractions of parallel computational kernels that span over several machines. +/* Parallel activities are convenient abstractions of parallel computational kernels that span over several machines. * To create a new one, you have to provide several things: - * - a vector of hosts on which the task will execute + * - a vector of hosts on which the activity will execute * - a vector of values, the amount of computation for each of the hosts (in flops) * - a matrix of values, the amount of communication between each pair of hosts (in bytes) * @@ -18,70 +18,113 @@ * Please note that you must have the LV07 platform model enabled to use such constructs. */ -#include "simgrid/plugins/energy.h" #include -#include -#include -XBT_LOG_NEW_DEFAULT_CATEGORY(s4u_energyptask, "Messages specific for this s4u example"); +XBT_LOG_NEW_DEFAULT_CATEGORY(s4u_ptask, "Messages specific for this s4u example"); static void runner() { /* Retrieve the list of all hosts as an array of hosts */ - std::vector hosts; - simgrid::s4u::Engine::getInstance()->getHostList(&hosts); - int hosts_count = hosts.size(); + auto hosts = simgrid::s4u::Engine::get_instance()->get_all_hosts(); + size_t hosts_count = hosts.size(); - XBT_INFO("First, build a classical parallel task, with 1 Gflop to execute on each node, " - "and 10MB to exchange between each pair"); - double* computation_amounts = new double[hosts_count](); - double* communication_amounts = new double[hosts_count * hosts_count](); + std::vector computation_amounts; + std::vector communication_amounts; - for (int i = 0; i < hosts_count; i++) - computation_amounts[i] = 1e9; // 1 Gflop + /* ------[ test 1 ]----------------- */ + XBT_INFO("First, build a classical parallel activity, with 1 Gflop to execute on each node, " + "and 10MB to exchange between each pair"); - for (int i = 0; i < hosts_count; i++) - for (int j = i + 1; j < hosts_count; j++) + computation_amounts.assign(hosts_count, 1e9 /*1Gflop*/); + communication_amounts.assign(hosts_count * hosts_count, 0); + for (size_t i = 0; i < hosts_count; i++) + for (size_t j = i + 1; j < hosts_count; j++) communication_amounts[i * hosts_count + j] = 1e7; // 10 MB - simgrid::s4u::this_actor::parallel_execute(hosts_count, hosts.data(), computation_amounts, communication_amounts); - - XBT_INFO("We can do the same with a timeout of one second enabled."); - computation_amounts = new double[hosts_count](); - communication_amounts = new double[hosts_count * hosts_count](); + simgrid::s4u::this_actor::parallel_execute(hosts, computation_amounts, communication_amounts); - for (int i = 0; i < hosts_count; i++) - computation_amounts[i] = 1e9; // 1 Gflop - - for (int i = 0; i < hosts_count; i++) - for (int j = i + 1; j < hosts_count; j++) + /* ------[ test 2 ]----------------- */ + XBT_INFO("We can do the same with a timeout of 10 seconds enabled."); + computation_amounts.assign(hosts_count, 1e9 /*1Gflop*/); + communication_amounts.assign(hosts_count * hosts_count, 0); + for (size_t i = 0; i < hosts_count; i++) + for (size_t j = i + 1; j < hosts_count; j++) communication_amounts[i * hosts_count + j] = 1e7; // 10 MB try { - simgrid::s4u::this_actor::parallel_execute(hosts_count, hosts.data(), computation_amounts, communication_amounts, - 1.0 /* timeout (in seconds)*/); - XBT_WARN("Woops, this did not timeout as expected... Please report that bug."); - } catch (xbt_ex& e) { - /* Do nothing this exception on timeout was expected */ - XBT_DEBUG("Caught expected exception: %s", e.what()); + simgrid::s4u::this_actor::exec_init(hosts, computation_amounts, communication_amounts) + ->wait_for(10.0 /* timeout (in seconds)*/); + xbt_die("Woops, this did not timeout as expected... Please report that bug."); + } catch (const simgrid::TimeoutException&) { + XBT_INFO("Caught the expected timeout exception."); } - XBT_INFO("Then, build a parallel task involving only computations and no communication (1 Gflop per node)"); - computation_amounts = new double[hosts_count](); - for (int i = 0; i < hosts_count; i++) - computation_amounts[i] = 1e9; // 1 Gflop - simgrid::s4u::this_actor::parallel_execute(hosts_count, hosts.data(), computation_amounts, nullptr /* no comm */); - - XBT_INFO("Then, build a parallel task with no computation nor communication (synchro only)"); - computation_amounts = new double[hosts_count](); - communication_amounts = new double[hosts_count * hosts_count](); - simgrid::s4u::this_actor::parallel_execute(hosts_count, hosts.data(), computation_amounts, communication_amounts); + /* ------[ test 3 ]----------------- */ + XBT_INFO("Then, build a parallel activity involving only computations (of different amounts) and no communication"); + computation_amounts = {3e8, 6e8, 1e9}; // 300Mflop, 600Mflop, 1Gflop + communication_amounts.clear(); // no comm + simgrid::s4u::this_actor::parallel_execute(hosts, computation_amounts, communication_amounts); + + /* ------[ test 4 ]----------------- */ + XBT_INFO("Then, build a parallel activity with no computation nor communication (synchro only)"); + computation_amounts.clear(); + communication_amounts.clear(); + simgrid::s4u::this_actor::parallel_execute(hosts, computation_amounts, communication_amounts); + + /* ------[ test 5 ]----------------- */ + XBT_INFO("Then, Monitor the execution of a parallel activity"); + computation_amounts.assign(hosts_count, 1e6 /*1Mflop*/); + communication_amounts = {0, 1e6, 0, 0, 0, 1e6, 1e6, 0, 0}; + simgrid::s4u::ExecPtr activity = + simgrid::s4u::this_actor::exec_init(hosts, computation_amounts, communication_amounts); + activity->start(); + + while (not activity->test()) { + XBT_INFO("Remaining flop ratio: %.0f%%", 100 * activity->get_remaining_ratio()); + simgrid::s4u::this_actor::sleep_for(5); + } + activity->wait(); + + /* ------[ test 6 ]----------------- */ + XBT_INFO("Finally, simulate a malleable task (a parallel execution that gets reconfigured after its start)."); + XBT_INFO(" - Start a regular parallel execution, with both comm and computation"); + computation_amounts.assign(hosts_count, 1e6 /*1Mflop*/); + communication_amounts = {0, 1e6, 0, 0, 1e6, 0, 1e6, 0, 0}; + activity = simgrid::s4u::this_actor::exec_init(hosts, computation_amounts, communication_amounts); + activity->start(); + + simgrid::s4u::this_actor::sleep_for(10); + double remaining_ratio = activity->get_remaining_ratio(); + XBT_INFO(" - After 10 seconds, %.2f%% remains to be done. Change it from 3 hosts to 2 hosts only.", + remaining_ratio * 100); + XBT_INFO(" Let's first suspend the task."); + activity->suspend(); + + XBT_INFO(" - Now, simulate the reconfiguration (modeled as a comm from the removed host to the remaining ones)."); + std::vector rescheduling_comp{0, 0, 0}; + std::vector rescheduling_comm{0, 0, 0, 0, 0, 0, 25000, 25000, 0}; + simgrid::s4u::this_actor::parallel_execute(hosts, rescheduling_comp, rescheduling_comm); + + XBT_INFO(" - Now, let's cancel the old task and create a new task with modified comm and computation vectors:"); + XBT_INFO(" What was already done is removed, and the load of the removed host is shared between remaining ones."); + for (int i = 0; i < 2; i++) { + // remove what we've done so far, for both comm and compute load + computation_amounts[i] *= remaining_ratio; + communication_amounts[i] *= remaining_ratio; + // The work from 1 must be shared between 2 remaining ones. 1/2=50% of extra work for each + computation_amounts[i] *= 1.5; + communication_amounts[i] *= 1.5; + } + hosts.resize(2); + computation_amounts.resize(2); + double remaining_comm = communication_amounts[1]; + communication_amounts = {0, remaining_comm, remaining_comm, 0}; // Resizing a linearized matrix is hairly - XBT_INFO("Finally, trick the ptask to do a 'remote execution', on host %s", hosts[1]->getCname()); - computation_amounts = new double[1]{1e9}; + activity->cancel(); + activity = simgrid::s4u::this_actor::exec_init(hosts, computation_amounts, communication_amounts); - simgrid::s4u::Host* remote[] = {hosts[1]}; - simgrid::s4u::this_actor::parallel_execute(1, remote, computation_amounts, nullptr); + XBT_INFO(" - Done, let's wait for the task completion"); + activity->wait(); XBT_INFO("Goodbye now!"); } @@ -90,16 +133,10 @@ int main(int argc, char* argv[]) { simgrid::s4u::Engine e(&argc, argv); - xbt_assert(argc <= 3, "1Usage: %s [--energy]", argv[0]); - xbt_assert(argc >= 2, "2Usage: %s [--energy]", argv[0]); - - if (argc == 3 && argv[2][2] == 'e') - sg_host_energy_plugin_init(); - - e.loadPlatform(argv[1]); + xbt_assert(argc == 2, "Usage: %s ", argv[0]); - /* Pick a process, no matter which, from the platform file */ - simgrid::s4u::Actor::createActor("test", simgrid::s4u::Host::by_name("MyHost1"), runner); + e.load_platform(argv[1]); + simgrid::s4u::Actor::create("test", simgrid::s4u::Host::by_name("MyHost1"), runner); e.run(); XBT_INFO("Simulation done.");