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->update_IB_factors(&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->update_IB_factors(&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>("Network_IB");
  73   simgrid::kernel::EngineImpl::get_instance()->add_model(net_model);
  74   simgrid::s4u::Engine::get_instance()->get_netzone_root()->get_impl()->set_network_model(net_model);
  75
  76   simgrid::s4u::Link::on_communication_state_change.connect(IB_action_state_changed_callback);
  77   simgrid::s4u::Link::on_communicate.connect(IB_action_init_callback);
  78   simgrid::s4u::Host::on_creation.connect(IB_create_host_callback);
  79   simgrid::config::set_default<double>("network/weight-S", 8775);
  80 }
  81
  82 namespace simgrid {
  83 namespace kernel {
  84 namespace resource {
  85
  86 NetworkIBModel::NetworkIBModel(const std::string& name) : NetworkSmpiModel(name)
  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::compute_IB_factors(IBNode* root) const
 115 {
 116   double num_comm_out    = root->active_comms_up_.size();
 117   double max_penalty_out = 0.0;
 118   // first, compute all outbound penalties to get their max
 119   for (ActiveComm const* comm : root->active_comms_up_) {
 120     double my_penalty_out = 1.0;
 121
 122     if (num_comm_out != 1) {
 123       if (comm->destination->nb_active_comms_down_ > 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 (ActiveComm* comm : root->active_comms_up_) {
 133     // compute inbound penalty
 134     double my_penalty_in = 1.0;
 135     int nb_comms         = comm->destination->nb_active_comms_down_; // total number of incoming comms
 136     if (nb_comms != 1)
 137       my_penalty_in = (comm->destination->active_comms_down_)[root]         // number of comm sent to dest by root node
 138                       * Be_ * comm->destination->active_comms_down_.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 = comm->action->get_bound();
 143     // save initial rate of the action
 144     if (comm->init_rate == -1)
 145       comm->init_rate = rate_before_update;
 146
 147     double penalized_bw = num_comm_out ? comm->init_rate / penalty : comm->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                 comm->destination->id_, comm->action, penalized_bw, comm->action->get_bound(), comm->init_rate);
 152       get_maxmin_system()->update_variable_bound(comm->action->get_variable(), penalized_bw);
 153     } else {
 154       XBT_DEBUG("%d->%d action %p penalty not updated : bw %f, initial rate %f", root->id_, comm->destination->id_,
 155                 comm->action, penalized_bw, comm->init_rate);
 156     }
 157   }
 158   XBT_DEBUG("Finished computing IB penalties");
 159 }
 160
 161 void NetworkIBModel::update_IB_factors_rec(IBNode* root, std::vector<bool>& updatedlist) const
 162 {
 163   if (not updatedlist[root->id_]) {
 164     XBT_DEBUG("IB - Updating rec %d", root->id_);
 165     compute_IB_factors(root);
 166     updatedlist[root->id_] = true;
 167     for (ActiveComm const* comm : root->active_comms_up_) {
 168       if (not updatedlist[comm->destination->id_])
 169         update_IB_factors_rec(comm->destination, updatedlist);
 170     }
 171     for (std::map<IBNode*, int>::value_type const& comm : root->active_comms_down_) {
 172       if (not updatedlist[comm.first->id_])
 173         update_IB_factors_rec(comm.first, updatedlist);
 174     }
 175   }
 176 }
 177
 178 void NetworkIBModel::update_IB_factors(NetworkAction* action, IBNode* from, IBNode* to, int remove) const
 179 {
 180   if (from == to) // disregard local comms (should use loopback)
 181     return;
 182
 183   if (remove) {
 184     if (to->active_comms_down_[from] == 1)
 185       to->active_comms_down_.erase(from);
 186     else
 187       to->active_comms_down_[from] -= 1;
 188
 189     to->nb_active_comms_down_--;
 190     auto it = std::find_if(begin(from->active_comms_up_), end(from->active_comms_up_),
 191                            [action](const ActiveComm* comm) { return comm->action == action; });
 192     if (it != std::end(from->active_comms_up_)) {
 193       delete *it;
 194       from->active_comms_up_.erase(it);
 195     }
 196     action->unref();
 197   } else {
 198     action->ref();
 199     auto* comm        = new ActiveComm();
 200     comm->action      = action;
 201     comm->destination = to;
 202     from->active_comms_up_.push_back(comm);
 203
 204     to->active_comms_down_[from] += 1;
 205     to->nb_active_comms_down_++;
 206   }
 207   XBT_DEBUG("IB - Updating %d", from->id_);
 208   std::vector<bool> updated(active_nodes.size(), false);
 209   update_IB_factors_rec(from, updated);
 210   XBT_DEBUG("IB - Finished updating %d", from->id_);
 211 }
 212 } // namespace resource
 213 } // namespace kernel
 214 } // namespace simgrid