src/surf/network_ib.cpp

   1 /* Copyright (c) 2014-2020. 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 "src/surf/network_ib.hpp"
   7 #include "simgrid/sg_config.hpp"
   8 #include "src/surf/HostImpl.hpp"
   9 #include "src/surf/xml/platf.hpp"
  10 #include "surf/surf.hpp"
  11
  12 #include <boost/algorithm/string/classification.hpp>
  13 #include <boost/algorithm/string/split.hpp>
  14
  15 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_network);
  16
  17 static void IB_create_host_callback(simgrid::s4u::Host const& host)
  18 {
  19   using simgrid::kernel::resource::IBNode;
  20   using simgrid::kernel::resource::NetworkIBModel;
  21
  22   static int id=0;
  23
  24   ((NetworkIBModel*)surf_network_model)->active_nodes.emplace(host.get_name(), IBNode(id));
  25   id++;
  26 }
  27
  28 static void IB_action_state_changed_callback(simgrid::kernel::resource::NetworkAction& action,
  29                                              simgrid::kernel::resource::Action::State /*previous*/)
  30 {
  31   using simgrid::kernel::resource::IBNode;
  32   using simgrid::kernel::resource::NetworkIBModel;
  33
  34   if (action.get_state() != simgrid::kernel::resource::Action::State::FINISHED)
  35     return;
  36   std::pair<IBNode*, IBNode*> pair = ((NetworkIBModel*)surf_network_model)->active_comms[&action];
  37   XBT_DEBUG("IB callback - action %p finished", &action);
  38
  39   ((NetworkIBModel*)surf_network_model)->updateIBfactors(&action, pair.first, pair.second, 1);
  40
  41   ((NetworkIBModel*)surf_network_model)->active_comms.erase(&action);
  42 }
  43
  44 static void IB_action_init_callback(simgrid::kernel::resource::NetworkAction& action)
  45 {
  46   simgrid::kernel::resource::NetworkIBModel* ibModel = (simgrid::kernel::resource::NetworkIBModel*)surf_network_model;
  47   simgrid::kernel::resource::IBNode* act_src         = &ibModel->active_nodes.at(action.get_src().get_name());
  48   simgrid::kernel::resource::IBNode* act_dst         = &ibModel->active_nodes.at(action.get_dst().get_name());
  49
  50   ibModel->active_comms[&action] = std::make_pair(act_src, act_dst);
  51   ibModel->updateIBfactors(&action, act_src, act_dst, 0);
  52 }
  53
  54 /*********
  55  * Model *
  56  *********/
  57
  58 /************************************************************************/
  59 /* New model based on MPI contention model for Infiniband platforms */
  60 /************************************************************************/
  61 /* @Inproceedings{mescal_vienne_phd, */
  62 /*  author={Jérôme Vienne}, */
  63 /*  title={prédiction de performances d’applications de calcul haute performance sur réseau Infiniband}, */
  64 /*  address={Grenoble FRANCE}, */
  65 /*  month=june, */
  66 /*  year={2010} */
  67 /*  } */
  68 void surf_network_model_init_IB()
  69 {
  70   xbt_assert(surf_network_model == nullptr, "Cannot set the network model twice");
  71
  72   surf_network_model = new simgrid::kernel::resource::NetworkIBModel();
  73   simgrid::s4u::Link::on_communication_state_change.connect(IB_action_state_changed_callback);
  74   simgrid::s4u::Link::on_communicate.connect(IB_action_init_callback);
  75   simgrid::s4u::Host::on_creation.connect(IB_create_host_callback);
  76   simgrid::config::set_default<double>("network/weight-S", 8775);
  77 }
  78
  79 namespace simgrid {
  80 namespace kernel {
  81 namespace resource {
  82
  83 NetworkIBModel::NetworkIBModel() : NetworkSmpiModel()
  84 {
  85   /* Do not add this into all_existing_models: our ancestor already does so */
  86
  87   std::string IB_factors_string = config::get_value<std::string>("smpi/IB-penalty-factors");
  88   std::vector<std::string> radical_elements;
  89   boost::split(radical_elements, IB_factors_string, boost::is_any_of(";"));
  90
  91   surf_parse_assert(radical_elements.size() == 3, "smpi/IB-penalty-factors should be provided and contain 3 "
  92                                                   "elements, semi-colon separated. Example: 0.965;0.925;1.35");
  93
  94   try {
  95     Be = std::stod(radical_elements.front());
  96   } catch (const std::invalid_argument& ia) {
  97     throw std::invalid_argument(std::string("First part of smpi/IB-penalty-factors is not numerical:") + ia.what());
  98   }
  99
 100   try {
 101     Bs = std::stod(radical_elements.at(1));
 102   } catch (const std::invalid_argument& ia) {
 103     throw std::invalid_argument(std::string("Second part of smpi/IB-penalty-factors is not numerical:") + ia.what());
 104   }
 105
 106   try {
 107     ys = std::stod(radical_elements.back());
 108   } catch (const std::invalid_argument& ia) {
 109     throw std::invalid_argument(std::string("Third part of smpi/IB-penalty-factors is not numerical:") + ia.what());
 110   }
 111 }
 112
 113 void NetworkIBModel::computeIBfactors(IBNode* root) const
 114 {
 115   double num_comm_out    = root->ActiveCommsUp.size();
 116   double max_penalty_out = 0.0;
 117   // first, compute all outbound penalties to get their max
 118   for (ActiveComm const* comm : root->ActiveCommsUp) {
 119     double my_penalty_out = 1.0;
 120
 121     if (num_comm_out != 1) {
 122       if (comm->destination->nbActiveCommsDown > 2) // number of comms sent to the receiving node
 123         my_penalty_out = num_comm_out * Bs * ys;
 124       else
 125         my_penalty_out = num_comm_out * Bs;
 126     }
 127
 128     max_penalty_out = std::max(max_penalty_out, my_penalty_out);
 129   }
 130
 131   for (ActiveComm* comm : root->ActiveCommsUp) {
 132     // compute inbound penalty
 133     double my_penalty_in = 1.0;
 134     int nb_comms         = comm->destination->nbActiveCommsDown; // total number of incoming comms
 135     if (nb_comms != 1)
 136       my_penalty_in = (comm->destination->ActiveCommsDown)[root]        // number of comm sent to dest by root node
 137                       * Be * comm->destination->ActiveCommsDown.size(); // number of different nodes sending to dest
 138
 139     double penalty = std::max(my_penalty_in, max_penalty_out);
 140
 141     double rate_before_update = comm->action->get_bound();
 142     // save initial rate of the action
 143     if (comm->init_rate == -1)
 144       comm->init_rate = rate_before_update;
 145
 146     double penalized_bw = num_comm_out ? comm->init_rate / penalty : comm->init_rate;
 147
 148     if (not double_equals(penalized_bw, rate_before_update, sg_surf_precision)) {
 149       XBT_DEBUG("%d->%d action %p penalty updated : bw now %f, before %f , initial rate %f", root->id,
 150                 comm->destination->id, comm->action, penalized_bw, comm->action->get_bound(), comm->init_rate);
 151       get_maxmin_system()->update_variable_bound(comm->action->get_variable(), penalized_bw);
 152     } else {
 153       XBT_DEBUG("%d->%d action %p penalty not updated : bw %f, initial rate %f", root->id, comm->destination->id,
 154                 comm->action, penalized_bw, comm->init_rate);
 155     }
 156   }
 157   XBT_DEBUG("Finished computing IB penalties");
 158 }
 159
 160 void NetworkIBModel::updateIBfactors_rec(IBNode* root, std::vector<bool>& updatedlist) const
 161 {
 162   if (not updatedlist[root->id]) {
 163     XBT_DEBUG("IB - Updating rec %d", root->id);
 164     computeIBfactors(root);
 165     updatedlist[root->id] = true;
 166     for (ActiveComm const* comm : root->ActiveCommsUp) {
 167       if (not updatedlist[comm->destination->id])
 168         updateIBfactors_rec(comm->destination, updatedlist);
 169     }
 170     for (std::map<IBNode*, int>::value_type const& comm : root->ActiveCommsDown) {
 171       if (not updatedlist[comm.first->id])
 172         updateIBfactors_rec(comm.first, updatedlist);
 173     }
 174   }
 175 }
 176
 177 void NetworkIBModel::updateIBfactors(NetworkAction* action, IBNode* from, IBNode* to, int remove) const
 178 {
 179   if (from == to) // disregard local comms (should use loopback)
 180     return;
 181
 182   if (remove) {
 183     if (to->ActiveCommsDown[from] == 1)
 184       to->ActiveCommsDown.erase(from);
 185     else
 186       to->ActiveCommsDown[from] -= 1;
 187
 188     to->nbActiveCommsDown--;
 189     std::vector<ActiveComm*>::iterator it =
 190         std::find_if(begin(from->ActiveCommsUp), end(from->ActiveCommsUp),
 191                      [action](const ActiveComm* comm) { return comm->action == action; });
 192     if (it != std::end(from->ActiveCommsUp)) {
 193       delete *it;
 194       from->ActiveCommsUp.erase(it);
 195     }
 196     action->unref();
 197   } else {
 198     action->ref();
 199     ActiveComm* comm  = new ActiveComm();
 200     comm->action      = action;
 201     comm->destination = to;
 202     from->ActiveCommsUp.push_back(comm);
 203
 204     to->ActiveCommsDown[from] += 1;
 205     to->nbActiveCommsDown++;
 206   }
 207   XBT_DEBUG("IB - Updating %d", from->id);
 208   std::vector<bool> updated(active_nodes.size(), false);
 209   updateIBfactors_rec(from, updated);
 210   XBT_DEBUG("IB - Finished updating %d", from->id);
 211 }
 212 } // namespace resource
 213 } // namespace kernel
 214 } // namespace simgrid