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 "network_ib.hpp"
  10
  11 #include "src/surf/HostImpl.hpp"
  12 #include "simgrid/sg_config.h"
  13 #include "maxmin_private.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,
  30       host.name().c_str(), act, nullptr);
  31
  32 }
  33
  34 static void IB_action_state_changed_callback(
  35     simgrid::surf::NetworkAction *action,
  36     simgrid::surf::Action::State statein, simgrid::surf::Action::State stateout)
  37 {
  38   using simgrid::surf::NetworkIBModel;
  39   using simgrid::surf::IBNode;
  40
  41   if(statein!=simgrid::surf::Action::State::running || stateout!=simgrid::surf::Action::State::done)
  42     return;
  43   std::pair<IBNode*,IBNode*> pair = ((NetworkIBModel*)surf_network_model)->active_comms[action];
  44   XBT_DEBUG("IB callback - action %p finished", action);
  45
  46   ((NetworkIBModel*)surf_network_model)->updateIBfactors(action, pair.first, pair.second, 1);
  47
  48   ((NetworkIBModel*)surf_network_model)->active_comms.erase(action);
  49
  50 }
  51
  52 static void IB_action_init_callback(simgrid::surf::NetworkAction* action, simgrid::s4u::Host* src,
  53                                     simgrid::s4u::Host* dst)
  54 {
  55   simgrid::surf::NetworkIBModel* ibModel = (simgrid::surf::NetworkIBModel*)surf_network_model;
  56
  57   simgrid::surf::IBNode* act_src =
  58       (simgrid::surf::IBNode*)xbt_dict_get_or_null(ibModel->active_nodes, src->name().c_str());
  59   xbt_assert(act_src, "could not find src node active comms !");
  60
  61   simgrid::surf::IBNode* act_dst =
  62       (simgrid::surf::IBNode*)xbt_dict_get_or_null(ibModel->active_nodes, dst->name().c_str());
  63   xbt_assert(act_dst, "could not find dst node active comms !");
  64
  65   ibModel->active_comms[action]=std::make_pair(act_src, act_dst);
  66
  67   ibModel->updateIBfactors(action, act_src, act_dst, 0);
  68 }
  69
  70 /*********
  71  * Model *
  72  *********/
  73
  74 /************************************************************************/
  75 /* New model based on MPI contention model for Infiniband platforms */
  76 /************************************************************************/
  77 /* @Inproceedings{mescal_vienne_phd, */
  78 /*  author={Jérôme Vienne}, */
  79 /*  title={prédiction de performances d’applications de calcul haute performance sur réseau Infiniband}, */
  80 /*  address={Grenoble FRANCE}, */
  81 /*  month=june, */
  82 /*  year={2010} */
  83 /*  } */
  84 void surf_network_model_init_IB()
  85 {
  86   using simgrid::surf::networkActionStateChangedCallbacks;
  87
  88   if (surf_network_model)
  89     return;
  90
  91   surf_network_model = new simgrid::surf::NetworkIBModel();
  92   all_existing_models->push_back(surf_network_model);
  93   networkActionStateChangedCallbacks.connect(IB_action_state_changed_callback);
  94   Link::onCommunicate.connect(IB_action_init_callback);
  95   simgrid::s4u::Host::onCreation.connect(IB_create_host_callback);
  96   xbt_cfg_setdefault_double("network/weight-S", 8775);
  97
  98 }
  99
 100 namespace simgrid {
 101   namespace surf {
 102
 103     NetworkIBModel::NetworkIBModel()
 104     : NetworkSmpiModel() {
 105       haveGap_=false;
 106       active_nodes=nullptr;
 107
 108       const char* IB_factors_string=xbt_cfg_get_string("smpi/IB-penalty-factors");
 109       xbt_dynar_t radical_elements = xbt_str_split(IB_factors_string, ";");
 110
 111       surf_parse_assert(xbt_dynar_length(radical_elements)==3,
 112           "smpi/IB-penalty-factors should be provided and contain 3 elements, semi-colon separated. Example: 0.965;0.925;1.35");
 113
 114       Be = xbt_str_parse_double(xbt_dynar_get_as(radical_elements, 0, char *), "First part of smpi/IB-penalty-factors is not numerical: %s");
 115       Bs = xbt_str_parse_double(xbt_dynar_get_as(radical_elements, 1, char *), "Second part of smpi/IB-penalty-factors is not numerical: %s");
 116       ys = xbt_str_parse_double(xbt_dynar_get_as(radical_elements, 2, char *), "Third part of smpi/IB-penalty-factors is not numerical: %s");
 117
 118       xbt_dynar_free(&radical_elements);
 119     }
 120
 121     NetworkIBModel::~NetworkIBModel()
 122     {
 123       xbt_dict_cursor_t cursor = nullptr;
 124       IBNode* instance = nullptr;
 125       char *name = nullptr;
 126       xbt_dict_foreach(active_nodes, cursor, name, instance)
 127       delete instance;
 128       xbt_dict_free(&active_nodes);
 129     }
 130
 131     void NetworkIBModel::computeIBfactors(IBNode *root) {
 132       double penalized_bw=0.0;
 133       double num_comm_out = (double) root->ActiveCommsUp.size();
 134       double max_penalty_out=0.0;
 135       //first, compute all outbound penalties to get their max
 136       for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 137         double my_penalty_out = 1.0;
 138
 139         if(num_comm_out!=1){
 140           if((*it)->destination->nbActiveCommsDown > 2)//number of comms sent to the receiving node
 141             my_penalty_out = num_comm_out * Bs * ys;
 142           else
 143             my_penalty_out = num_comm_out * Bs;
 144         }
 145
 146         max_penalty_out = std::max(max_penalty_out,my_penalty_out);
 147       }
 148
 149       for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 150
 151         //compute inbound penalty
 152         double my_penalty_in = 1.0;
 153         int nb_comms = (*it)->destination->nbActiveCommsDown;//total number of incoming comms
 154         if(nb_comms!=1)
 155           my_penalty_in = ((*it)->destination->ActiveCommsDown)[root] //number of comm sent to dest by root node
 156                                                                 * Be
 157                                                                 * (*it)->destination->ActiveCommsDown.size();//number of different nodes sending to dest
 158
 159         double penalty = std::max(my_penalty_in,max_penalty_out);
 160
 161         double rate_before_update = (*it)->action->getBound();
 162         //save initial rate of the action
 163         if((*it)->init_rate==-1)
 164           (*it)->init_rate= rate_before_update;
 165
 166         penalized_bw= ! num_comm_out ? (*it)->init_rate : (*it)->init_rate /penalty;
 167
 168         if (!double_equals(penalized_bw, rate_before_update, sg_surf_precision)){
 169           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 );
 170           lmm_update_variable_bound(maxminSystem_, (*it)->action->getVariable(), penalized_bw);
 171         }else{
 172           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 );
 173         }
 174
 175       }
 176       XBT_DEBUG("Finished computing IB penalties");
 177     }
 178
 179     void NetworkIBModel::updateIBfactors_rec(IBNode *root, bool* updatedlist) {
 180       if(updatedlist[root->id]==0){
 181         XBT_DEBUG("IB - Updating rec %d", root->id);
 182         computeIBfactors(root);
 183         updatedlist[root->id]=1;
 184         for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 185           if(updatedlist[(*it)->destination->id]!=1)
 186             updateIBfactors_rec((*it)->destination, updatedlist);
 187         }
 188         for (std::map<IBNode*, int>::iterator it= root->ActiveCommsDown.begin(); it != root->ActiveCommsDown.end(); ++it) {
 189           if(updatedlist[it->first->id]!=1)
 190             updateIBfactors_rec(it->first, updatedlist);
 191         }
 192       }
 193     }
 194
 195
 196     void NetworkIBModel::updateIBfactors(NetworkAction *action, IBNode *from, IBNode * to, int remove) {
 197       if (from == to)//disregard local comms (should use loopback)
 198         return;
 199
 200       bool* updated=(bool*)xbt_malloc0(xbt_dict_size(active_nodes)*sizeof(bool));
 201       ActiveComm* comm=nullptr;
 202       if(remove){
 203         if(to->ActiveCommsDown[from]==1)
 204           to->ActiveCommsDown.erase(from);
 205         else
 206           to->ActiveCommsDown[from]-=1;
 207
 208         to->nbActiveCommsDown--;
 209         for (std::vector<ActiveComm*>::iterator it= from->ActiveCommsUp.begin();
 210             it != from->ActiveCommsUp.end(); ++it) {
 211           if((*it)->action==action){
 212             comm=(*it);
 213             from->ActiveCommsUp.erase(it);
 214             break;
 215           }
 216         }
 217         action->unref();
 218
 219       }else{
 220         action->ref();
 221         ActiveComm* comm=new ActiveComm();
 222         comm->action=action;
 223         comm->destination=to;
 224         from->ActiveCommsUp.push_back(comm);
 225
 226         to->ActiveCommsDown[from]+=1;
 227         to->nbActiveCommsDown++;
 228       }
 229       XBT_DEBUG("IB - Updating %d", from->id);
 230       updateIBfactors_rec(from, updated);
 231       XBT_DEBUG("IB - Finished updating %d", from->id);
 232       if(comm)
 233         delete comm;
 234       xbt_free(updated);
 235     }
 236
 237   }
 238 }