src/surf/network_ib.cpp

   1 /* Copyright (c) 2014-2019. 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, simgrid::s4u::Host* src,
  45                                     simgrid::s4u::Host* dst)
  46 {
  47   simgrid::kernel::resource::NetworkIBModel* ibModel = (simgrid::kernel::resource::NetworkIBModel*)surf_network_model;
  48   simgrid::kernel::resource::IBNode* act_src         = &ibModel->active_nodes.at(src->get_name());
  49   simgrid::kernel::resource::IBNode* act_dst         = &ibModel->active_nodes.at(dst->get_name());
  50
  51   ibModel->active_comms[&action] = std::make_pair(act_src, act_dst);
  52   ibModel->updateIBfactors(&action, act_src, act_dst, 0);
  53 }
  54
  55 /*********
  56  * Model *
  57  *********/
  58
  59 /************************************************************************/
  60 /* New model based on MPI contention model for Infiniband platforms */
  61 /************************************************************************/
  62 /* @Inproceedings{mescal_vienne_phd, */
  63 /*  author={Jérôme Vienne}, */
  64 /*  title={prédiction de performances d’applications de calcul haute performance sur réseau Infiniband}, */
  65 /*  address={Grenoble FRANCE}, */
  66 /*  month=june, */
  67 /*  year={2010} */
  68 /*  } */
  69 void surf_network_model_init_IB()
  70 {
  71   xbt_assert(surf_network_model == nullptr, "Cannot set the network model twice");
  72
  73   surf_network_model = new simgrid::kernel::resource::NetworkIBModel();
  74   simgrid::s4u::Link::on_communication_state_change.connect(IB_action_state_changed_callback);
  75   simgrid::s4u::Link::on_communicate.connect(IB_action_init_callback);
  76   simgrid::s4u::Host::on_creation.connect(IB_create_host_callback);
  77   simgrid::config::set_default<double>("network/weight-S", 8775);
  78 }
  79
  80 namespace simgrid {
  81 namespace kernel {
  82 namespace resource {
  83
  84 NetworkIBModel::NetworkIBModel() : NetworkSmpiModel()
  85 {
  86   /* Do not add this into all_existing_models: our ancestor already does so */
  87
  88   std::string IB_factors_string = config::get_value<std::string>("smpi/IB-penalty-factors");
  89   std::vector<std::string> radical_elements;
  90   boost::split(radical_elements, IB_factors_string, boost::is_any_of(";"));
  91
  92   surf_parse_assert(radical_elements.size() == 3, "smpi/IB-penalty-factors should be provided and contain 3 "
  93                                                   "elements, semi-colon separated. Example: 0.965;0.925;1.35");
  94
  95   try {
  96     Be = std::stod(radical_elements.front());
  97   } catch (const std::invalid_argument& ia) {
  98     throw std::invalid_argument(std::string("First part of smpi/IB-penalty-factors is not numerical:") + ia.what());
  99   }
 100
 101   try {
 102     Bs = std::stod(radical_elements.at(1));
 103   } catch (const std::invalid_argument& ia) {
 104     throw std::invalid_argument(std::string("Second part of smpi/IB-penalty-factors is not numerical:") + ia.what());
 105   }
 106
 107   try {
 108     ys = std::stod(radical_elements.back());
 109   } catch (const std::invalid_argument& ia) {
 110     throw std::invalid_argument(std::string("Third part of smpi/IB-penalty-factors is not numerical:") + ia.what());
 111   }
 112 }
 113
 114 void NetworkIBModel::computeIBfactors(IBNode* root)
 115 {
 116   double num_comm_out    = root->ActiveCommsUp.size();
 117   double max_penalty_out = 0.0;
 118   // first, compute all outbound penalties to get their max
 119   for (std::vector<ActiveComm*>::iterator it = root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 120     double my_penalty_out = 1.0;
 121
 122     if (num_comm_out != 1) {
 123       if ((*it)->destination->nbActiveCommsDown > 2) // number of comms sent to the receiving node
 124         my_penalty_out = num_comm_out * Bs * ys;
 125       else
 126         my_penalty_out = num_comm_out * Bs;
 127     }
 128
 129     max_penalty_out = std::max(max_penalty_out, my_penalty_out);
 130   }
 131
 132   for (std::vector<ActiveComm*>::iterator it = root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 133     // compute inbound penalty
 134     double my_penalty_in = 1.0;
 135     int nb_comms         = (*it)->destination->nbActiveCommsDown; // total number of incoming comms
 136     if (nb_comms != 1)
 137       my_penalty_in = ((*it)->destination->ActiveCommsDown)[root]        // number of comm sent to dest by root node
 138                       * Be * (*it)->destination->ActiveCommsDown.size(); // number of different nodes sending to dest
 139
 140     double penalty = std::max(my_penalty_in, max_penalty_out);
 141
 142     double rate_before_update = (*it)->action->get_bound();
 143     // save initial rate of the action
 144     if ((*it)->init_rate == -1)
 145       (*it)->init_rate = rate_before_update;
 146
 147     double penalized_bw = num_comm_out ? (*it)->init_rate / penalty : (*it)->init_rate;
 148
 149     if (not double_equals(penalized_bw, rate_before_update, sg_surf_precision)) {
 150       XBT_DEBUG("%d->%d action %p penalty updated : bw now %f, before %f , initial rate %f", root->id,
 151                 (*it)->destination->id, (*it)->action, penalized_bw, (*it)->action->get_bound(), (*it)->init_rate);
 152       get_maxmin_system()->update_variable_bound((*it)->action->get_variable(), penalized_bw);
 153     } else {
 154       XBT_DEBUG("%d->%d action %p penalty not updated : bw %f, initial rate %f", root->id, (*it)->destination->id,
 155                 (*it)->action, penalized_bw, (*it)->init_rate);
 156     }
 157   }
 158   XBT_DEBUG("Finished computing IB penalties");
 159 }
 160
 161 void NetworkIBModel::updateIBfactors_rec(IBNode* root, std::vector<bool>& updatedlist)
 162 {
 163   if (not updatedlist[root->id]) {
 164     XBT_DEBUG("IB - Updating rec %d", root->id);
 165     computeIBfactors(root);
 166     updatedlist[root->id] = true;
 167     for (std::vector<ActiveComm*>::iterator it = root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 168       if (not updatedlist[(*it)->destination->id])
 169         updateIBfactors_rec((*it)->destination, updatedlist);
 170     }
 171     for (std::map<IBNode*, int>::iterator it = root->ActiveCommsDown.begin(); it != root->ActiveCommsDown.end(); ++it) {
 172       if (not updatedlist[it->first->id])
 173         updateIBfactors_rec(it->first, updatedlist);
 174     }
 175   }
 176 }
 177
 178 void NetworkIBModel::updateIBfactors(NetworkAction* action, IBNode* from, IBNode* to, int remove)
 179 {
 180   if (from == to) // disregard local comms (should use loopback)
 181     return;
 182
 183   if (remove) {
 184     if (to->ActiveCommsDown[from] == 1)
 185       to->ActiveCommsDown.erase(from);
 186     else
 187       to->ActiveCommsDown[from] -= 1;
 188
 189     to->nbActiveCommsDown--;
 190     for (std::vector<ActiveComm*>::iterator it = from->ActiveCommsUp.begin(); it != from->ActiveCommsUp.end(); ++it) {
 191       if ((*it)->action == action) {
 192         delete *it;
 193         from->ActiveCommsUp.erase(it);
 194         break;
 195       }
 196     }
 197     action->unref();
 198   } else {
 199     action->ref();
 200     ActiveComm* comm  = new ActiveComm();
 201     comm->action      = action;
 202     comm->destination = to;
 203     from->ActiveCommsUp.push_back(comm);
 204
 205     to->ActiveCommsDown[from] += 1;
 206     to->nbActiveCommsDown++;
 207   }
 208   XBT_DEBUG("IB - Updating %d", from->id);
 209   std::vector<bool> updated(active_nodes.size(), false);
 210   updateIBfactors_rec(from, updated);
 211   XBT_DEBUG("IB - Finished updating %d", from->id);
 212 }
 213 } // namespace resource
 214 } // namespace kernel
 215 } // namespace simgrid