src/surf/network_ib.cpp

   1 /* Copyright (c) 2014-2023. 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/kernel/routing/NetPoint.hpp>
   7
   8 #include "simgrid/sg_config.hpp"
   9 #include "src/kernel/EngineImpl.hpp"
  10 #include "src/kernel/activity/CommImpl.hpp"
  11 #include "src/surf/HostImpl.hpp"
  12 #include "src/surf/network_ib.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 /*********
  20  * Model *
  21  *********/
  22
  23 /************************************************************************/
  24 /* New model based on MPI contention model for Infiniband platforms */
  25 /************************************************************************/
  26 /* @Inproceedings{mescal_vienne_phd, */
  27 /*  author={Jérôme Vienne}, */
  28 /*  title={prédiction de performances d’applications de calcul haute performance sur réseau Infiniband}, */
  29 /*  address={Grenoble FRANCE}, */
  30 /*  month=june, */
  31 /*  year={2010} */
  32 /*  } */
  33 void surf_network_model_init_IB()
  34 {
  35   using simgrid::kernel::resource::NetworkIBModel;
  36
  37   auto net_model = std::make_shared<NetworkIBModel>("Network_IB");
  38   auto* engine   = simgrid::kernel::EngineImpl::get_instance();
  39   engine->add_model(net_model);
  40   engine->get_netzone_root()->set_network_model(net_model);
  41
  42   simgrid::s4u::Link::on_communication_state_change_cb(NetworkIBModel::IB_action_state_changed_callback);
  43   simgrid::kernel::activity::CommImpl::on_start.connect(NetworkIBModel::IB_comm_start_callback);
  44   simgrid::s4u::Host::on_creation_cb(NetworkIBModel::IB_create_host_callback);
  45   simgrid::config::set_default<double>("network/weight-S", 8775);
  46 }
  47
  48 namespace simgrid::kernel::resource {
  49
  50 void NetworkIBModel::IB_create_host_callback(s4u::Host const& host)
  51 {
  52   static int id = 0;
  53   auto* ibModel = static_cast<NetworkIBModel*>(host.get_netpoint()->get_englobing_zone()->get_network_model().get());
  54   ibModel->active_nodes.try_emplace(host.get_name(), id);
  55   id++;
  56 }
  57
  58 void NetworkIBModel::IB_action_state_changed_callback(NetworkAction& action, Action::State /*previous*/)
  59 {
  60   if (action.get_state() != Action::State::FINISHED)
  61     return;
  62   auto* ibModel   = static_cast<NetworkIBModel*>(action.get_model());
  63   auto [src, dst] = ibModel->active_comms[&action];
  64
  65   XBT_DEBUG("IB callback - action %p finished", &action);
  66   ibModel->update_IB_factors(&action, src, dst, 1);
  67   ibModel->active_comms.erase(&action);
  68 }
  69
  70 void NetworkIBModel::IB_comm_start_callback(const activity::CommImpl& comm)
  71 {
  72   auto* action  = static_cast<NetworkAction*>(comm.surf_action_);
  73   auto* ibModel = static_cast<NetworkIBModel*>(action->get_model());
  74   auto* act_src = &ibModel->active_nodes.at(action->get_src().get_name());
  75   auto* act_dst = &ibModel->active_nodes.at(action->get_dst().get_name());
  76
  77   ibModel->active_comms[action] = std::make_pair(act_src, act_dst);
  78   ibModel->update_IB_factors(action, act_src, act_dst, 0);
  79 }
  80
  81 NetworkIBModel::NetworkIBModel(const std::string& name) : NetworkCm02Model(name)
  82 {
  83   std::string IB_factors_string = config::get_value<std::string>("smpi/IB-penalty-factors");
  84   std::vector<std::string> radical_elements;
  85   boost::split(radical_elements, IB_factors_string, boost::is_any_of(";"));
  86
  87   xbt_assert(radical_elements.size() == 3, "smpi/IB-penalty-factors should be provided and contain 3 "
  88                                            "elements, semi-colon separated. Example: 0.965;0.925;1.35");
  89
  90   try {
  91     Be_ = std::stod(radical_elements.front());
  92   } catch (const std::invalid_argument& ia) {
  93     throw std::invalid_argument(std::string("First part of smpi/IB-penalty-factors is not numerical:") + ia.what());
  94   }
  95
  96   try {
  97     Bs_ = std::stod(radical_elements.at(1));
  98   } catch (const std::invalid_argument& ia) {
  99     throw std::invalid_argument(std::string("Second part of smpi/IB-penalty-factors is not numerical:") + ia.what());
 100   }
 101
 102   try {
 103     ys_ = std::stod(radical_elements.back());
 104   } catch (const std::invalid_argument& ia) {
 105     throw std::invalid_argument(std::string("Third part of smpi/IB-penalty-factors is not numerical:") + ia.what());
 106   }
 107 }
 108
 109 void NetworkIBModel::compute_IB_factors(IBNode* root) const
 110 {
 111   size_t num_comm_out    = root->active_comms_up_.size();
 112   double max_penalty_out = 0.0;
 113   // first, compute all outbound penalties to get their max
 114   for (ActiveComm const* comm : root->active_comms_up_) {
 115     double my_penalty_out = 1.0;
 116
 117     if (num_comm_out != 1) {
 118       if (comm->destination->nb_active_comms_down_ > 2) // number of comms sent to the receiving node
 119         my_penalty_out = num_comm_out * Bs_ * ys_;
 120       else
 121         my_penalty_out = num_comm_out * Bs_;
 122     }
 123
 124     max_penalty_out = std::max(max_penalty_out, my_penalty_out);
 125   }
 126
 127   for (ActiveComm* comm : root->active_comms_up_) {
 128     // compute inbound penalty
 129     double my_penalty_in = 1.0;
 130     if (comm->destination->nb_active_comms_down_ != 1)                      // total number of incoming comms
 131       my_penalty_in = (comm->destination->active_comms_down_)[root]         // number of comm sent to dest by root node
 132                       * Be_ * comm->destination->active_comms_down_.size(); // number of different nodes sending to dest
 133
 134     double penalty = std::max(my_penalty_in, max_penalty_out);
 135
 136     double rate_before_update = comm->action->get_bound();
 137     // save initial rate of the action
 138     if (comm->init_rate == -1)
 139       comm->init_rate = rate_before_update;
 140
 141     double penalized_bw = num_comm_out ? comm->init_rate / penalty : comm->init_rate;
 142
 143     if (not double_equals(penalized_bw, rate_before_update, sg_surf_precision)) {
 144       XBT_DEBUG("%d->%d action %p penalty updated : bw now %f, before %f , initial rate %f", root->id_,
 145                 comm->destination->id_, comm->action, penalized_bw, comm->action->get_bound(), comm->init_rate);
 146       get_maxmin_system()->update_variable_bound(comm->action->get_variable(), penalized_bw);
 147     } else {
 148       XBT_DEBUG("%d->%d action %p penalty not updated : bw %f, initial rate %f", root->id_, comm->destination->id_,
 149                 comm->action, penalized_bw, comm->init_rate);
 150     }
 151   }
 152   XBT_DEBUG("Finished computing IB penalties");
 153 }
 154
 155 void NetworkIBModel::update_IB_factors_rec(IBNode* root, std::vector<bool>& updatedlist) const
 156 {
 157   if (not updatedlist[root->id_]) {
 158     XBT_DEBUG("IB - Updating rec %d", root->id_);
 159     compute_IB_factors(root);
 160     updatedlist[root->id_] = true;
 161     for (ActiveComm const* comm : root->active_comms_up_) {
 162       if (not updatedlist[comm->destination->id_])
 163         update_IB_factors_rec(comm->destination, updatedlist);
 164     }
 165     for (auto const& [comm, _] : root->active_comms_down_) {
 166       if (not updatedlist[comm->id_])
 167         update_IB_factors_rec(comm, updatedlist);
 168     }
 169   }
 170 }
 171
 172 void NetworkIBModel::update_IB_factors(NetworkAction* action, IBNode* from, IBNode* to, int remove) const
 173 {
 174   if (from == to) // disregard local comms (should use loopback)
 175     return;
 176
 177   if (remove) {
 178     if (to->active_comms_down_[from] == 1)
 179       to->active_comms_down_.erase(from);
 180     else
 181       to->active_comms_down_[from] -= 1;
 182
 183     to->nb_active_comms_down_--;
 184     if (auto it = std::find_if(begin(from->active_comms_up_), end(from->active_comms_up_),
 185                                [action](const ActiveComm* comm) { return comm->action == action; });
 186         it != std::end(from->active_comms_up_)) {
 187       delete *it;
 188       from->active_comms_up_.erase(it);
 189     }
 190     action->unref();
 191   } else {
 192     action->ref();
 193     auto* comm        = new ActiveComm();
 194     comm->action      = action;
 195     comm->destination = to;
 196     from->active_comms_up_.push_back(comm);
 197
 198     to->active_comms_down_[from] += 1;
 199     to->nb_active_comms_down_++;
 200   }
 201   XBT_DEBUG("IB - Updating %d", from->id_);
 202   std::vector<bool> updated(active_nodes.size(), false);
 203   update_IB_factors_rec(from, updated);
 204   XBT_DEBUG("IB - Finished updating %d", from->id_);
 205 }
 206 } // namespace simgrid::kernel::resource