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 <algorithm>
   8 #include <utility>
   9
  10 #include "network_ib.hpp"
  11 #include "simgrid/sg_config.h"
  12 #include "maxmin_private.hpp"
  13 #include "src/surf/host_interface.hpp"
  14
  15 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_network);
  16
  17 static void IB_create_host_callback(simgrid::surf::Host* host){
  18   using namespace simgrid::surf;
  19
  20   static int id=0;
  21 // pour t->id -> rajouter une nouvelle struct dans le dict, pour stocker les comms actives
  22   if(((NetworkIBModel*)surf_network_model)->active_nodes==NULL)
  23     ((NetworkIBModel*)surf_network_model)->active_nodes=xbt_dict_new();
  24
  25   IBNode* act = new IBNode(id);
  26
  27   id++;
  28   xbt_dict_set(((NetworkIBModel*)surf_network_model)->active_nodes,
  29     host->getName(), act, NULL);
  30
  31 }
  32
  33 static void IB_action_state_changed_callback(
  34   simgrid::surf::NetworkAction *action,
  35   e_surf_action_state_t statein, e_surf_action_state_t stateout)
  36 {
  37  using namespace simgrid::surf;
  38  if(statein!=SURF_ACTION_RUNNING|| stateout!=SURF_ACTION_DONE)
  39     return;
  40   std::pair<IBNode*,IBNode*> pair = ((NetworkIBModel*)surf_network_model)->active_comms[action];
  41   XBT_DEBUG("IB callback - action %p finished", action);
  42
  43  ((NetworkIBModel*)surf_network_model)->updateIBfactors(action, pair.first, pair.second, 1);
  44
  45   ((NetworkIBModel*)surf_network_model)->active_comms.erase(action);
  46
  47 }
  48
  49
  50 static void IB_action_init_callback(
  51   simgrid::surf::NetworkAction *action, simgrid::surf::RoutingEdge *src, simgrid::surf::RoutingEdge *dst,
  52   double size, double rate)
  53 {
  54   using namespace simgrid::surf;
  55   if(((NetworkIBModel*)surf_network_model)->active_nodes==NULL)
  56     xbt_die("IB comm added, without any node connected !");
  57
  58   IBNode* act_src= (IBNode*) xbt_dict_get_or_null(((NetworkIBModel*)surf_network_model)->active_nodes, src->getName());
  59   if(act_src==NULL)
  60     xbt_die("could not find src node active comms !");
  61   //act_src->rate=rate;
  62
  63   IBNode* act_dst= (IBNode*) xbt_dict_get_or_null(((NetworkIBModel*)surf_network_model)->active_nodes, dst->getName());
  64   if(act_dst==NULL)
  65     xbt_die("could not find dst node active comms !");
  66  // act_dst->rate=rate;
  67
  68   ((NetworkIBModel*)surf_network_model)->active_comms[action]=std::make_pair(act_src, act_dst);
  69   //post the action in the second dist, to retrieve in the other callback
  70   XBT_DEBUG("IB callback - action %p init", action);
  71
  72   ((NetworkIBModel*)surf_network_model)->updateIBfactors(action, act_src, act_dst, 0);
  73
  74 }
  75
  76 /*********
  77  * Model *
  78  *********/
  79
  80 /************************************************************************/
  81 /* New model based on MPI contention model for Infiniband platforms */
  82 /************************************************************************/
  83 /* @Inproceedings{mescal_vienne_phd, */
  84 /*  author={Jérôme Vienne}, */
  85 /*  title={prédiction de performances d’applications de calcul haute performance sur réseau Infiniband}, */
  86 /*  address={Grenoble FRANCE}, */
  87 /*  month=june, */
  88 /*  year={2010} */
  89 /*  } */
  90 void surf_network_model_init_IB(void)
  91 {
  92   using simgrid::surf::networkActionStateChangedCallbacks;
  93   using simgrid::surf::networkCommunicateCallbacks;
  94   using simgrid::surf::hostCreatedCallbacks;
  95
  96   if (surf_network_model)
  97     return;
  98   surf_network_model = new simgrid::surf::NetworkIBModel();
  99   net_define_callbacks();
 100   xbt_dynar_push(all_existing_models, &surf_network_model);
 101   networkActionStateChangedCallbacks.connect(IB_action_state_changed_callback);
 102   networkCommunicateCallbacks.connect(IB_action_init_callback);
 103   hostCreatedCallbacks.connect(IB_create_host_callback);
 104   xbt_cfg_setdefault_double(_sg_cfg_set, "network/weight_S", 8775);
 105
 106 }
 107
 108 namespace simgrid {
 109 namespace surf {
 110
 111 NetworkIBModel::NetworkIBModel()
 112  : NetworkSmpiModel() {
 113   m_haveGap=false;
 114   active_nodes=NULL;
 115
 116   const char* IB_factors_string=sg_cfg_get_string("smpi/IB_penalty_factors");
 117   xbt_dynar_t radical_elements = xbt_str_split(IB_factors_string, ";");
 118
 119   if(xbt_dynar_length(radical_elements)!=3)
 120     surf_parse_error("smpi/IB_penalty_factors should be provided and contain 3 elements, semi-colon separated : for example 0.965;0.925;1.35");
 121
 122   Be = atof(xbt_dynar_get_as(radical_elements, 0, char *));
 123   Bs = atof(xbt_dynar_get_as(radical_elements, 1, char *));
 124   ys = atof(xbt_dynar_get_as(radical_elements, 2, char *));
 125
 126   xbt_dynar_free(&radical_elements);
 127 }
 128
 129 NetworkIBModel::~NetworkIBModel()
 130 {
 131   xbt_dict_cursor_t cursor = NULL;
 132   IBNode* instance = NULL;
 133   char *name = NULL;
 134   xbt_dict_foreach(active_nodes, cursor, name, instance)
 135     delete instance;
 136   xbt_dict_free(&active_nodes);
 137 }
 138
 139 void NetworkIBModel::computeIBfactors(IBNode *root) {
 140   double penalized_bw=0.0;
 141   double num_comm_out = (double) root->ActiveCommsUp.size();
 142   double max_penalty_out=0.0;
 143   //first, compute all outbound penalties to get their max
 144   for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 145     double my_penalty_out = 1.0;
 146
 147     if(num_comm_out!=1){
 148       if((*it)->destination->nbActiveCommsDown > 2)//number of comms sent to the receiving node
 149         my_penalty_out = num_comm_out * Bs * ys;
 150       else
 151         my_penalty_out = num_comm_out * Bs;
 152     }
 153
 154     max_penalty_out = std::max(max_penalty_out,my_penalty_out);
 155   }
 156
 157   for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 158
 159     //compute inbound penalty
 160     double my_penalty_in = 1.0;
 161     int nb_comms = (*it)->destination->nbActiveCommsDown;//total number of incoming comms
 162     if(nb_comms!=1)
 163       my_penalty_in = ((*it)->destination->ActiveCommsDown)[root] //number of comm sent to dest by root node
 164                       * Be
 165                       * (*it)->destination->ActiveCommsDown.size();//number of different nodes sending to dest
 166
 167     double penalty = std::max(my_penalty_in,max_penalty_out);
 168
 169     double rate_before_update = (*it)->action->getBound();
 170     //save initial rate of the action
 171     if((*it)->init_rate==-1)
 172       (*it)->init_rate= rate_before_update;
 173
 174     penalized_bw= ! num_comm_out ? (*it)->init_rate : (*it)->init_rate /penalty;
 175
 176     if (!double_equals(penalized_bw, rate_before_update, sg_surf_precision)){
 177       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 );
 178       lmm_update_variable_bound(p_maxminSystem, (*it)->action->getVariable(), penalized_bw);
 179     }else{
 180       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 );
 181     }
 182
 183   }
 184   XBT_DEBUG("Finished computing IB penalties");
 185 }
 186
 187 void NetworkIBModel::updateIBfactors_rec(IBNode *root, bool* updatedlist) {
 188   if(updatedlist[root->id]==0){
 189     XBT_DEBUG("IB - Updating rec %d", root->id);
 190     computeIBfactors(root);
 191     updatedlist[root->id]=1;
 192     for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 193         if(updatedlist[(*it)->destination->id]!=1)
 194           updateIBfactors_rec((*it)->destination, updatedlist);
 195     }
 196     for (std::map<IBNode*, int>::iterator it= root->ActiveCommsDown.begin(); it != root->ActiveCommsDown.end(); ++it) {
 197         if(updatedlist[it->first->id]!=1)
 198           updateIBfactors_rec(it->first, updatedlist);
 199     }
 200   }
 201 }
 202
 203
 204 void NetworkIBModel::updateIBfactors(NetworkAction *action, IBNode *from, IBNode * to, int remove) {
 205   if (from == to)//disregard local comms (should use loopback)
 206     return;
 207
 208   bool* updated=(bool*)xbt_malloc0(xbt_dict_size(active_nodes)*sizeof(bool));
 209   ActiveComm* comm=NULL;
 210   if(remove){
 211     if(to->ActiveCommsDown[from]==1)
 212       to->ActiveCommsDown.erase(from);
 213     else
 214       to->ActiveCommsDown[from]-=1;
 215
 216     to->nbActiveCommsDown--;
 217     for (std::vector<ActiveComm*>::iterator it= from->ActiveCommsUp.begin();
 218          it != from->ActiveCommsUp.end(); ++it) {
 219       if((*it)->action==action){
 220         comm=(*it);
 221         from->ActiveCommsUp.erase(it);
 222         break;
 223       }
 224     }
 225     action->unref();
 226
 227   }else{
 228     action->ref();
 229     ActiveComm* comm=new ActiveComm();
 230     comm->action=action;
 231     comm->destination=to;
 232     from->ActiveCommsUp.push_back(comm);
 233
 234     to->ActiveCommsDown[from]+=1;
 235     to->nbActiveCommsDown++;
 236   }
 237   XBT_DEBUG("IB - Updating %d", from->id);
 238   updateIBfactors_rec(from, updated);
 239   XBT_DEBUG("IB - Finished updating %d", from->id);
 240   if(comm)
 241     delete comm;
 242   xbt_free(updated);
 243 }
 244
 245 }
 246 }