src/surf/network_ib.cpp

   1 /* Copyright (c) 2014-2021. 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/kernel/routing/NetPoint.hpp"
   8 #include "simgrid/sg_config.hpp"
   9 #include "src/kernel/EngineImpl.hpp"
  10 #include "src/surf/HostImpl.hpp"
  11 #include "src/surf/xml/platf.hpp"
  12 #include "surf/surf.hpp"
  13
  14 #include <boost/algorithm/string/classification.hpp>
  15 #include <boost/algorithm/string/split.hpp>
  16
  17 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(res_network);
  18
  19 static void IB_create_host_callback(simgrid::s4u::Host const& host)
  20 {
  21   using simgrid::kernel::resource::IBNode;
  22   using simgrid::kernel::resource::NetworkIBModel;
  23
  24   static int id = 0;
  25   auto* ibModel = static_cast<NetworkIBModel*>(host.get_netpoint()->get_englobing_zone()->get_network_model().get());
  26   ibModel->active_nodes.emplace(host.get_name(), IBNode(id));
  27   id++;
  28 }
  29
  30 static void IB_action_state_changed_callback(simgrid::kernel::resource::NetworkAction& action,
  31                                              simgrid::kernel::resource::Action::State /*previous*/)
  32 {
  33   using simgrid::kernel::resource::IBNode;
  34
  35   if (action.get_state() != simgrid::kernel::resource::Action::State::FINISHED)
  36     return;
  37   auto* ibModel                    = static_cast<simgrid::kernel::resource::NetworkIBModel*>(action.get_model());
  38   std::pair<IBNode*, IBNode*> pair = ibModel->active_comms[&action];
  39   XBT_DEBUG("IB callback - action %p finished", &action);
  40
  41   ibModel->updateIBfactors(&action, pair.first, pair.second, 1);
  42
  43   ibModel->active_comms.erase(&action);
  44 }
  45
  46 static void IB_action_init_callback(simgrid::kernel::resource::NetworkAction& action)
  47 {
  48   auto* ibModel = static_cast<simgrid::kernel::resource::NetworkIBModel*>(action.get_model());
  49   auto* act_src = &ibModel->active_nodes.at(action.get_src().get_name());
  50   auto* act_dst = &ibModel->active_nodes.at(action.get_dst().get_name());
  51
  52   ibModel->active_comms[&action] = std::make_pair(act_src, act_dst);
  53   ibModel->updateIBfactors(&action, act_src, act_dst, 0);
  54 }
  55
  56 /*********
  57  * Model *
  58  *********/
  59
  60 /************************************************************************/
  61 /* New model based on MPI contention model for Infiniband platforms */
  62 /************************************************************************/
  63 /* @Inproceedings{mescal_vienne_phd, */
  64 /*  author={Jérôme Vienne}, */
  65 /*  title={prédiction de performances d’applications de calcul haute performance sur réseau Infiniband}, */
  66 /*  address={Grenoble FRANCE}, */
  67 /*  month=june, */
  68 /*  year={2010} */
  69 /*  } */
  70 void surf_network_model_init_IB()
  71 {
  72   auto net_model = std::make_shared<simgrid::kernel::resource::NetworkIBModel>();
  73   simgrid::kernel::EngineImpl::get_instance()->add_model(simgrid::kernel::resource::Model::Type::NETWORK, net_model,
  74                                                          true);
  75   simgrid::s4u::Engine::get_instance()->get_netzone_root()->get_impl()->set_network_model(net_model);
  76
  77   simgrid::s4u::Link::on_communication_state_change.connect(IB_action_state_changed_callback);
  78   simgrid::s4u::Link::on_communicate.connect(IB_action_init_callback);
  79   simgrid::s4u::Host::on_creation.connect(IB_create_host_callback);
  80   simgrid::config::set_default<double>("network/weight-S", 8775);
  81 }
  82
  83 namespace simgrid {
  84 namespace kernel {
  85 namespace resource {
  86
  87 NetworkIBModel::NetworkIBModel() : NetworkSmpiModel()
  88 {
  89   std::string IB_factors_string = config::get_value<std::string>("smpi/IB-penalty-factors");
  90   std::vector<std::string> radical_elements;
  91   boost::split(radical_elements, IB_factors_string, boost::is_any_of(";"));
  92
  93   surf_parse_assert(radical_elements.size() == 3, "smpi/IB-penalty-factors should be provided and contain 3 "
  94                                                   "elements, semi-colon separated. Example: 0.965;0.925;1.35");
  95
  96   try {
  97     Be = std::stod(radical_elements.front());
  98   } catch (const std::invalid_argument& ia) {
  99     throw std::invalid_argument(std::string("First part of smpi/IB-penalty-factors is not numerical:") + ia.what());
 100   }
 101
 102   try {
 103     Bs = std::stod(radical_elements.at(1));
 104   } catch (const std::invalid_argument& ia) {
 105     throw std::invalid_argument(std::string("Second part of smpi/IB-penalty-factors is not numerical:") + ia.what());
 106   }
 107
 108   try {
 109     ys = std::stod(radical_elements.back());
 110   } catch (const std::invalid_argument& ia) {
 111     throw std::invalid_argument(std::string("Third part of smpi/IB-penalty-factors is not numerical:") + ia.what());
 112   }
 113 }
 114
 115 void NetworkIBModel::computeIBfactors(IBNode* root) const
 116 {
 117   double num_comm_out    = root->ActiveCommsUp.size();
 118   double max_penalty_out = 0.0;
 119   // first, compute all outbound penalties to get their max
 120   for (ActiveComm const* comm : root->ActiveCommsUp) {
 121     double my_penalty_out = 1.0;
 122
 123     if (num_comm_out != 1) {
 124       if (comm->destination->nbActiveCommsDown > 2) // number of comms sent to the receiving node
 125         my_penalty_out = num_comm_out * Bs * ys;
 126       else
 127         my_penalty_out = num_comm_out * Bs;
 128     }
 129
 130     max_penalty_out = std::max(max_penalty_out, my_penalty_out);
 131   }
 132
 133   for (ActiveComm* comm : root->ActiveCommsUp) {
 134     // compute inbound penalty
 135     double my_penalty_in = 1.0;
 136     int nb_comms         = comm->destination->nbActiveCommsDown; // total number of incoming comms
 137     if (nb_comms != 1)
 138       my_penalty_in = (comm->destination->ActiveCommsDown)[root]        // number of comm sent to dest by root node
 139                       * Be * comm->destination->ActiveCommsDown.size(); // number of different nodes sending to dest
 140
 141     double penalty = std::max(my_penalty_in, max_penalty_out);
 142
 143     double rate_before_update = comm->action->get_bound();
 144     // save initial rate of the action
 145     if (comm->init_rate == -1)
 146       comm->init_rate = rate_before_update;
 147
 148     double penalized_bw = num_comm_out ? comm->init_rate / penalty : comm->init_rate;
 149
 150     if (not double_equals(penalized_bw, rate_before_update, sg_surf_precision)) {
 151       XBT_DEBUG("%d->%d action %p penalty updated : bw now %f, before %f , initial rate %f", root->id,
 152                 comm->destination->id, comm->action, penalized_bw, comm->action->get_bound(), comm->init_rate);
 153       get_maxmin_system()->update_variable_bound(comm->action->get_variable(), penalized_bw);
 154     } else {
 155       XBT_DEBUG("%d->%d action %p penalty not updated : bw %f, initial rate %f", root->id, comm->destination->id,
 156                 comm->action, penalized_bw, comm->init_rate);
 157     }
 158   }
 159   XBT_DEBUG("Finished computing IB penalties");
 160 }
 161
 162 void NetworkIBModel::updateIBfactors_rec(IBNode* root, std::vector<bool>& updatedlist) const
 163 {
 164   if (not updatedlist[root->id]) {
 165     XBT_DEBUG("IB - Updating rec %d", root->id);
 166     computeIBfactors(root);
 167     updatedlist[root->id] = true;
 168     for (ActiveComm const* comm : root->ActiveCommsUp) {
 169       if (not updatedlist[comm->destination->id])
 170         updateIBfactors_rec(comm->destination, updatedlist);
 171     }
 172     for (std::map<IBNode*, int>::value_type const& comm : root->ActiveCommsDown) {
 173       if (not updatedlist[comm.first->id])
 174         updateIBfactors_rec(comm.first, updatedlist);
 175     }
 176   }
 177 }
 178
 179 void NetworkIBModel::updateIBfactors(NetworkAction* action, IBNode* from, IBNode* to, int remove) const
 180 {
 181   if (from == to) // disregard local comms (should use loopback)
 182     return;
 183
 184   if (remove) {
 185     if (to->ActiveCommsDown[from] == 1)
 186       to->ActiveCommsDown.erase(from);
 187     else
 188       to->ActiveCommsDown[from] -= 1;
 189
 190     to->nbActiveCommsDown--;
 191     auto it = std::find_if(begin(from->ActiveCommsUp), end(from->ActiveCommsUp),
 192                            [action](const ActiveComm* comm) { return comm->action == action; });
 193     if (it != std::end(from->ActiveCommsUp)) {
 194       delete *it;
 195       from->ActiveCommsUp.erase(it);
 196     }
 197     action->unref();
 198   } else {
 199     action->ref();
 200     auto* 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