src/smpi/plugins/sampi_loadbalancer.cpp

   1 /* Copyright (c) 2018.      The SimGrid Team. All rights reserved.          */
   2
   3 /* This program is free software; you can redistribute it and/or modify it
   4  * under the terms of the license (GNU LGPL) which comes with this package. */
   5
   6 #include <simgrid/plugins/load_balancer.h>
   7 #include <simgrid/s4u.hpp>
   8 #include <simgrid/smpi/replay.hpp>
   9 #include <smpi/smpi.h>
  10 #include <src/smpi/include/smpi_comm.hpp>
  11 #include <src/smpi/include/smpi_actor.hpp>
  12 #include <src/smpi/plugins/ampi/instr_ampi.hpp>
  13 #include <src/smpi/plugins/ampi/ampi.hpp>
  14 #include <xbt/replay.hpp>
  15
  16 #include "src/kernel/activity/ExecImpl.hpp"
  17 #include "src/simix/ActorImpl.hpp"
  18 #include "src/smpi/plugins/load_balancer/load_balancer.hpp" // This is not yet ready to be public
  19
  20 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(plugin_load_balancer, smpi, "Logging specific to the SMPI load balancing plugin");
  21
  22 static simgrid::config::Flag<int> cfg_migration_frequency("smpi/plugin/lb/migration-frequency", {"smpi/plugin/lb/migration_frequency"},
  23     "After how many calls to the migration function should the migration be actually executed?", 10,
  24     [](double val){if (val != 10) sg_load_balancer_plugin_init();});
  25
  26 namespace simgrid {
  27 namespace smpi {
  28 namespace plugin {
  29
  30 static simgrid::plugin::loadbalancer::LoadBalancer lb;
  31
  32 class MigrateParser : public simgrid::smpi::replay::ActionArgParser {
  33 public:
  34   double memory_consumption;
  35   void parse(simgrid::xbt::ReplayAction& action, std::string name)
  36   {
  37     // The only parameter is the amount of memory used by the current process.
  38     CHECK_ACTION_PARAMS(action, 1, 0);
  39     memory_consumption = std::stod(action[2]);
  40   }
  41 };
  42
  43 /* This function simulates what happens when the original application calls
  44  * (A)MPI_Migrate. It executes the load balancing heuristics, makes the necessary
  45  * migrations and updates the task mapping in the load balancer.
  46  */
  47 class MigrateAction : public simgrid::smpi::replay::ReplayAction<simgrid::smpi::plugin::MigrateParser> {
  48 public:
  49   explicit MigrateAction() : ReplayAction("Migrate") {}
  50   void kernel(simgrid::xbt::ReplayAction& action)
  51   {
  52     static std::map<simgrid::s4u::ActorPtr, int> migration_call_counter;
  53     static simgrid::s4u::Barrier smpilb_bar(smpi_process_count());
  54     simgrid::s4u::Host* cur_host = simgrid::s4u::this_actor::get_host();
  55     simgrid::s4u::Host* migrate_to_host;
  56
  57     TRACE_migration_call(my_proc_id, NULL);
  58
  59     migration_call_counter[simgrid::s4u::Actor::self()]++;
  60     if ((migration_call_counter[simgrid::s4u::Actor::self()] % simgrid::config::get_value<int>(cfg_migration_frequency.get_name())) != 0) {
  61       return;
  62     }
  63
  64     // TODO cheinrich: Why do we need this barrier?
  65     smpilb_bar.wait();
  66
  67     static bool was_executed = false;
  68     if (not was_executed) {
  69       was_executed = true;
  70       smpi_bench_begin();
  71       XBT_INFO("RUNNING THE LB");
  72       lb.run();
  73       smpi_bench_end();
  74     }
  75
  76     // This barrier is required to ensure that the mapping has been computed and is available
  77     smpilb_bar.wait();
  78     was_executed = false; // Must stay behind this barrier so that all processes have passed the if clause
  79
  80     migrate_to_host = lb.get_mapping();
  81     if (cur_host != migrate_to_host) { // Origin and dest are not the same -> migrate
  82       sg_host_t migration_hosts[2] = {cur_host, migrate_to_host};
  83       // Changing this to double[2] ... will cause trouble with parallel_execute, because that fct is trying to call free().
  84       double* comp_amount  = new double[2]{0, 0};
  85       double* comm_amount  = new double[4]{0, /*must not be 0*/std::max(args.memory_consumption, 1.0), 0, 0};
  86
  87       xbt_os_timer_t timer = smpi_process()->timer();
  88       xbt_os_threadtimer_start(timer);
  89       simgrid::s4u::this_actor::parallel_execute(2, migration_hosts, comp_amount, comm_amount, -1.0);
  90       xbt_os_threadtimer_stop(timer);
  91       smpi_execute(xbt_os_timer_elapsed(timer));
  92
  93       // Update the process and host mapping in SimGrid.
  94       TRACE_smpi_process_change_host(my_proc_id, migrate_to_host);
  95       simgrid::s4u::this_actor::migrate(migrate_to_host);
  96     }
  97
  98     smpilb_bar.wait();
  99
 100     smpi_bench_begin();
 101   }
 102 };
 103
 104 /******************************************************************************
 105  *         Code to include the duration of iterations in the trace.           *
 106  ******************************************************************************/
 107
 108 // FIXME Move declaration
 109 XBT_PRIVATE void action_iteration_in(simgrid::xbt::ReplayAction& action);
 110 void action_iteration_in(simgrid::xbt::ReplayAction& action)
 111 {
 112   CHECK_ACTION_PARAMS(action, 0, 0)
 113   TRACE_Iteration_in(simgrid::s4u::this_actor::get_pid(), nullptr);
 114   simgrid::smpi::plugin::ampi::on_iteration_in(MPI_COMM_WORLD->group()->actor(std::stol(action[0])));
 115 }
 116
 117 XBT_PRIVATE void action_iteration_out(simgrid::xbt::ReplayAction& action);
 118 void action_iteration_out(simgrid::xbt::ReplayAction& action)
 119 {
 120   CHECK_ACTION_PARAMS(action, 0, 0)
 121   TRACE_Iteration_out(simgrid::s4u::this_actor::get_pid(), nullptr);
 122   simgrid::smpi::plugin::ampi::on_iteration_out(MPI_COMM_WORLD->group()->actor(std::stol(action[0])));
 123 }
 124 }
 125 }
 126 }
 127
 128 /** @ingroup plugin_loadbalancer
 129  * @brief Initializes the load balancer plugin
 130  * @details The load balancer plugin supports several AMPI load balancers that move ranks
 131  * around, based on their host's load.
 132  */
 133 void sg_load_balancer_plugin_init()
 134 {
 135   static bool done = false;
 136   if (!done) {
 137     done = true;
 138     simgrid::kernel::activity::ExecImpl::on_completion.connect([](simgrid::kernel::activity::ExecImplPtr activity){
 139         simgrid::smpi::plugin::lb.record_actor_computation(activity->simcalls_.front()->issuer->iface(), activity->surf_action_->get_cost());
 140     });
 141
 142     xbt_replay_action_register(
 143         "migrate", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::plugin::MigrateAction().execute(action); });
 144     xbt_replay_action_register("iteration_in", simgrid::smpi::plugin::action_iteration_in);
 145     xbt_replay_action_register("iteration_out", simgrid::smpi::plugin::action_iteration_out);
 146   }
 147 }