src/surf/network_ib.cpp

   1 /* Copyright (c) 2014-2015. The SimGrid Team.
   2 *All rights reserved.                                                     */
   3
   4 /* This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the license (GNU LGPL) which comes with this package. */
   6
   7 #include <utility>
   8
   9 #include "simgrid/sg_config.h"
  10 #include "src/surf/HostImpl.hpp"
  11 #include "src/surf/maxmin_private.hpp"
  12 #include "src/surf/network_ib.hpp"
  13 #include "src/surf/xml/platf.hpp"
  14 #include <boost/algorithm/string/classification.hpp>
  15 #include <boost/algorithm/string/split.hpp>
  16
  17 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_network);
  18
  19 static void IB_create_host_callback(simgrid::s4u::Host& host){
  20   using simgrid::surf::NetworkIBModel;
  21   using simgrid::surf::IBNode;
  22
  23   static int id=0;
  24   // pour t->id -> rajouter une nouvelle struct dans le dict, pour stocker les comms actives
  25   if(((NetworkIBModel*)surf_network_model)->active_nodes==nullptr)
  26     ((NetworkIBModel*)surf_network_model)->active_nodes = xbt_dict_new_homogeneous(nullptr);
  27
  28   IBNode* act = new IBNode(id);
  29
  30   id++;
  31   xbt_dict_set(((NetworkIBModel*)surf_network_model)->active_nodes, host.cname(), act, nullptr);
  32 }
  33
  34 static void IB_action_state_changed_callback(simgrid::surf::NetworkAction* action)
  35 {
  36   using simgrid::surf::NetworkIBModel;
  37   using simgrid::surf::IBNode;
  38
  39   if (action->getState() != simgrid::surf::Action::State::done)
  40     return;
  41   std::pair<IBNode*,IBNode*> pair = ((NetworkIBModel*)surf_network_model)->active_comms[action];
  42   XBT_DEBUG("IB callback - action %p finished", action);
  43
  44   ((NetworkIBModel*)surf_network_model)->updateIBfactors(action, pair.first, pair.second, 1);
  45
  46   ((NetworkIBModel*)surf_network_model)->active_comms.erase(action);
  47
  48 }
  49
  50 static void IB_action_init_callback(simgrid::surf::NetworkAction* action, simgrid::s4u::Host* src,
  51                                     simgrid::s4u::Host* dst)
  52 {
  53   simgrid::surf::NetworkIBModel* ibModel = (simgrid::surf::NetworkIBModel*)surf_network_model;
  54
  55   simgrid::surf::IBNode* act_src = (simgrid::surf::IBNode*)xbt_dict_get_or_null(ibModel->active_nodes, src->cname());
  56   xbt_assert(act_src, "could not find src node active comms !");
  57
  58   simgrid::surf::IBNode* act_dst = (simgrid::surf::IBNode*)xbt_dict_get_or_null(ibModel->active_nodes, dst->cname());
  59   xbt_assert(act_dst, "could not find dst node active comms !");
  60
  61   ibModel->active_comms[action]=std::make_pair(act_src, act_dst);
  62
  63   ibModel->updateIBfactors(action, act_src, act_dst, 0);
  64 }
  65
  66 /*********
  67  * Model *
  68  *********/
  69
  70 /************************************************************************/
  71 /* New model based on MPI contention model for Infiniband platforms */
  72 /************************************************************************/
  73 /* @Inproceedings{mescal_vienne_phd, */
  74 /*  author={Jérôme Vienne}, */
  75 /*  title={prédiction de performances d’applications de calcul haute performance sur réseau Infiniband}, */
  76 /*  address={Grenoble FRANCE}, */
  77 /*  month=june, */
  78 /*  year={2010} */
  79 /*  } */
  80 void surf_network_model_init_IB()
  81 {
  82   if (surf_network_model)
  83     return;
  84
  85   surf_network_model = new simgrid::surf::NetworkIBModel();
  86   all_existing_models->push_back(surf_network_model);
  87   simgrid::s4u::Link::onCommunicationStateChange.connect(IB_action_state_changed_callback);
  88   simgrid::s4u::Link::onCommunicate.connect(IB_action_init_callback);
  89   simgrid::s4u::Host::onCreation.connect(IB_create_host_callback);
  90   xbt_cfg_setdefault_double("network/weight-S", 8775);
  91
  92 }
  93
  94 namespace simgrid {
  95   namespace surf {
  96
  97     NetworkIBModel::NetworkIBModel()
  98     : NetworkSmpiModel() {
  99       haveGap_=false;
 100       active_nodes=nullptr;
 101
 102       const char* IB_factors_string=xbt_cfg_get_string("smpi/IB-penalty-factors");
 103       std::vector<std::string> radical_elements;
 104       boost::split(radical_elements, IB_factors_string, boost::is_any_of(";"));
 105
 106       surf_parse_assert(radical_elements.size() == 3, "smpi/IB-penalty-factors should be provided and contain 3 "
 107                                                       "elements, semi-colon separated. Example: 0.965;0.925;1.35");
 108
 109       Be = xbt_str_parse_double(radical_elements.front().c_str(),
 110                                 "First part of smpi/IB-penalty-factors is not numerical: %s");
 111       Bs = xbt_str_parse_double((radical_elements.at(1)).c_str(),
 112                                 "Second part of smpi/IB-penalty-factors is not numerical: %s");
 113       ys = xbt_str_parse_double(radical_elements.back().c_str(),
 114                                 "Third part of smpi/IB-penalty-factors is not numerical: %s");
 115     }
 116
 117     NetworkIBModel::~NetworkIBModel()
 118     {
 119       xbt_dict_cursor_t cursor = nullptr;
 120       IBNode* instance = nullptr;
 121       char *name = nullptr;
 122       xbt_dict_foreach(active_nodes, cursor, name, instance)
 123       delete instance;
 124       xbt_dict_free(&active_nodes);
 125     }
 126
 127     void NetworkIBModel::computeIBfactors(IBNode *root) {
 128       double num_comm_out = (double) root->ActiveCommsUp.size();
 129       double max_penalty_out=0.0;
 130       //first, compute all outbound penalties to get their max
 131       for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 132         double my_penalty_out = 1.0;
 133
 134         if(num_comm_out!=1){
 135           if((*it)->destination->nbActiveCommsDown > 2)//number of comms sent to the receiving node
 136             my_penalty_out = num_comm_out * Bs * ys;
 137           else
 138             my_penalty_out = num_comm_out * Bs;
 139         }
 140
 141         max_penalty_out = std::max(max_penalty_out,my_penalty_out);
 142       }
 143
 144       for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 145
 146         //compute inbound penalty
 147         double my_penalty_in = 1.0;
 148         int nb_comms = (*it)->destination->nbActiveCommsDown;//total number of incoming comms
 149         if(nb_comms!=1)
 150           my_penalty_in = ((*it)->destination->ActiveCommsDown)[root] //number of comm sent to dest by root node
 151                                                                 * Be
 152                                                                 * (*it)->destination->ActiveCommsDown.size();//number of different nodes sending to dest
 153
 154         double penalty = std::max(my_penalty_in,max_penalty_out);
 155
 156         double rate_before_update = (*it)->action->getBound();
 157         //save initial rate of the action
 158         if((*it)->init_rate==-1)
 159           (*it)->init_rate= rate_before_update;
 160
 161         double penalized_bw = num_comm_out ? (*it)->init_rate / penalty : (*it)->init_rate;
 162
 163         if (not double_equals(penalized_bw, rate_before_update, sg_surf_precision)) {
 164           XBT_DEBUG("%d->%d action %p penalty updated : bw now %f, before %f , initial rate %f", root->id,(*it)->destination->id,(*it)->action,penalized_bw, (*it)->action->getBound(), (*it)->init_rate );
 165           lmm_update_variable_bound(maxminSystem_, (*it)->action->getVariable(), penalized_bw);
 166         }else{
 167           XBT_DEBUG("%d->%d action %p penalty not updated : bw %f, initial rate %f", root->id,(*it)->destination->id,(*it)->action,penalized_bw, (*it)->init_rate );
 168         }
 169
 170       }
 171       XBT_DEBUG("Finished computing IB penalties");
 172     }
 173
 174     void NetworkIBModel::updateIBfactors_rec(IBNode *root, bool* updatedlist) {
 175       if(updatedlist[root->id]==0){
 176         XBT_DEBUG("IB - Updating rec %d", root->id);
 177         computeIBfactors(root);
 178         updatedlist[root->id]=1;
 179         for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 180           if(updatedlist[(*it)->destination->id]!=1)
 181             updateIBfactors_rec((*it)->destination, updatedlist);
 182         }
 183         for (std::map<IBNode*, int>::iterator it= root->ActiveCommsDown.begin(); it != root->ActiveCommsDown.end(); ++it) {
 184           if(updatedlist[it->first->id]!=1)
 185             updateIBfactors_rec(it->first, updatedlist);
 186         }
 187       }
 188     }
 189
 190
 191     void NetworkIBModel::updateIBfactors(NetworkAction *action, IBNode *from, IBNode * to, int remove) {
 192       if (from == to)//disregard local comms (should use loopback)
 193         return;
 194
 195       bool* updated=(bool*)xbt_malloc0(xbt_dict_size(active_nodes)*sizeof(bool));
 196       ActiveComm* comm=nullptr;
 197       if(remove){
 198         if(to->ActiveCommsDown[from]==1)
 199           to->ActiveCommsDown.erase(from);
 200         else
 201           to->ActiveCommsDown[from]-=1;
 202
 203         to->nbActiveCommsDown--;
 204         for (std::vector<ActiveComm*>::iterator it= from->ActiveCommsUp.begin();
 205             it != from->ActiveCommsUp.end(); ++it) {
 206           if((*it)->action==action){
 207             comm=(*it);
 208             from->ActiveCommsUp.erase(it);
 209             break;
 210           }
 211         }
 212         action->unref();
 213
 214       }else{
 215         action->ref();
 216         ActiveComm* comm=new ActiveComm();
 217         comm->action=action;
 218         comm->destination=to;
 219         from->ActiveCommsUp.push_back(comm);
 220
 221         to->ActiveCommsDown[from]+=1;
 222         to->nbActiveCommsDown++;
 223       }
 224       XBT_DEBUG("IB - Updating %d", from->id);
 225       updateIBfactors_rec(from, updated);
 226       XBT_DEBUG("IB - Finished updating %d", from->id);
 227       if(comm)
 228         delete comm;
 229       xbt_free(updated);
 230     }
 231
 232   }
 233 }