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