src/surf/network_ib.cpp

   1 /* Copyright (c) 2014. 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 "network_ib.hpp"
   8 #include "simgrid/sg_config.h"
   9 #include "maxmin_private.hpp"
  10
  11 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_network);
  12
  13 static void IB_create_host_callback(sg_platf_host_cbarg_t t){
  14
  15   static int id=0;
  16 // pour t->id -> rajouter une nouvelle struct dans le dict, pour stocker les comms actives
  17   if(((NetworkIBModel*)surf_network_model)->active_nodes==NULL)
  18     ((NetworkIBModel*)surf_network_model)->active_nodes=xbt_dict_new();
  19
  20   IBNode* act = new IBNode(id);
  21
  22   id++;
  23   xbt_dict_set(((NetworkIBModel*)surf_network_model)->active_nodes, t->id, act, NULL);
  24
  25 }
  26
  27 static void IB_action_state_changed_callback(NetworkAction *action, e_surf_action_state_t statein, e_surf_action_state_t stateout){
  28  if(statein!=SURF_ACTION_RUNNING|| stateout!=SURF_ACTION_DONE)
  29     return;
  30   std::pair<IBNode*,IBNode*> pair = ((NetworkIBModel*)surf_network_model)->active_comms[action];
  31   XBT_DEBUG("IB callback - action %p finished", action);
  32
  33  ((NetworkIBModel*)surf_network_model)->updateIBfactors(action, pair.first, pair.second, 1);
  34
  35   ((NetworkIBModel*)surf_network_model)->active_comms.erase(action);
  36
  37 }
  38
  39
  40 static void IB_action_init_callback(NetworkAction *action,RoutingEdge *src, RoutingEdge *dst, double size, double rate){
  41   if(((NetworkIBModel*)surf_network_model)->active_nodes==NULL)
  42     xbt_die("IB comm added, without any node connected !");
  43
  44   IBNode* act_src= (IBNode*) xbt_dict_get_or_null(((NetworkIBModel*)surf_network_model)->active_nodes, src->getName());
  45   if(act_src==NULL)
  46     xbt_die("could not find src node active comms !");
  47   //act_src->rate=rate;
  48
  49   IBNode* act_dst= (IBNode*) xbt_dict_get_or_null(((NetworkIBModel*)surf_network_model)->active_nodes, dst->getName());
  50   if(act_dst==NULL)
  51     xbt_die("could not find dst node active comms !");
  52  // act_dst->rate=rate;
  53
  54   ((NetworkIBModel*)surf_network_model)->active_comms[action]=make_pair(act_src, act_dst);
  55   //post the action in the second dist, to retrieve in the other callback
  56   XBT_DEBUG("IB callback - action %p init", action);
  57
  58   ((NetworkIBModel*)surf_network_model)->updateIBfactors(action, act_src, act_dst, 0);
  59
  60 }
  61
  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(void)
  79 {
  80
  81   if (surf_network_model)
  82     return;
  83   surf_network_model = new NetworkIBModel();
  84   net_define_callbacks();
  85   xbt_dynar_push(model_list, &surf_network_model);
  86   surf_callback_connect(networkActionStateChangedCallbacks, IB_action_state_changed_callback);
  87   surf_callback_connect(networkCommunicateCallbacks, IB_action_init_callback);
  88
  89   sg_platf_host_add_cb(IB_create_host_callback);
  90   xbt_cfg_setdefault_double(_sg_cfg_set, "network/weight_S", 8775);
  91
  92 }
  93
  94 NetworkIBModel::NetworkIBModel()
  95  : NetworkSmpiModel() {
  96   m_haveGap=false;
  97   active_nodes=NULL;
  98
  99   const char* IB_factors_string=sg_cfg_get_string("smpi/IB_penalty_factors");
 100   xbt_dynar_t radical_elements = xbt_str_split(IB_factors_string, ";");
 101
 102   if(xbt_dynar_length(radical_elements)!=3)
 103     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");
 104
 105   Be = atof(xbt_dynar_get_as(radical_elements, 0, char *));
 106   Bs = atof(xbt_dynar_get_as(radical_elements, 1, char *));
 107   ys = atof(xbt_dynar_get_as(radical_elements, 2, char *));
 108
 109   xbt_dynar_free(&radical_elements);
 110 }
 111
 112 NetworkIBModel::~NetworkIBModel()
 113 {
 114   xbt_dict_cursor_t cursor = NULL;
 115   IBNode* instance = NULL;
 116   char *name = NULL;
 117   xbt_dict_foreach(active_nodes, cursor, name, instance)
 118     delete instance;
 119   xbt_dict_free(&active_nodes);
 120 }
 121
 122 void NetworkIBModel::computeIBfactors(IBNode *root) {
 123   double penalized_bw=0.0;
 124   double num_comm_out = (double) root->ActiveCommsUp.size();
 125   double max_penalty_out=0.0;
 126   //first, compute all outbound penalties to get their max
 127   for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 128     double my_penalty_out = 1.0;
 129
 130     if(num_comm_out!=1){
 131       if((*it)->destination->nbActiveCommsDown > 2)//number of comms sent to the receiving node
 132         my_penalty_out = num_comm_out * Bs * ys;
 133       else
 134         my_penalty_out = num_comm_out * Bs;
 135     }
 136
 137     max_penalty_out = max(max_penalty_out,my_penalty_out);
 138   }
 139
 140   for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 141
 142     //compute inbound penalty
 143     double my_penalty_in = 1.0;
 144     int nb_comms = (*it)->destination->nbActiveCommsDown;//total number of incoming comms
 145     if(nb_comms!=1)
 146       my_penalty_in = ((*it)->destination->ActiveCommsDown)[root] //number of comm sent to dest by root node
 147                       * Be
 148                       * (*it)->destination->ActiveCommsDown.size();//number of different nodes sending to dest
 149
 150     double penalty=max(my_penalty_in,max_penalty_out);
 151
 152     double rate_before_update = (*it)->action->getBound();
 153     //save initial rate of the action
 154     if((*it)->init_rate==-1)
 155       (*it)->init_rate= rate_before_update;
 156
 157     penalized_bw= ! num_comm_out ? (*it)->init_rate : (*it)->init_rate /penalty;
 158
 159     if (!double_equals(penalized_bw, rate_before_update, sg_surf_precision)){
 160       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 );
 161       lmm_update_variable_bound(p_maxminSystem, (*it)->action->getVariable(), penalized_bw);
 162     }else{
 163       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 );
 164     }
 165
 166   }
 167   XBT_DEBUG("Finished computing IB penalties");
 168 }
 169
 170 void NetworkIBModel::updateIBfactors_rec(IBNode *root, bool* updatedlist) {
 171   if(updatedlist[root->id]==0){
 172     XBT_DEBUG("IB - Updating rec %d", root->id);
 173     computeIBfactors(root);
 174     updatedlist[root->id]=1;
 175     for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 176         if(updatedlist[(*it)->destination->id]!=1)
 177           updateIBfactors_rec((*it)->destination, updatedlist);
 178     }
 179     for (std::map<IBNode*, int>::iterator it= root->ActiveCommsDown.begin(); it != root->ActiveCommsDown.end(); ++it) {
 180         if(updatedlist[it->first->id]!=1)
 181           updateIBfactors_rec(it->first, updatedlist);
 182     }
 183   }
 184 }
 185
 186
 187 void NetworkIBModel::updateIBfactors(NetworkAction *action, IBNode *from, IBNode * to, int remove) {
 188   if (from == to)//disregard local comms (should use loopback)
 189     return;
 190
 191   bool* updated=(bool*)xbt_malloc0(xbt_dict_size(active_nodes)*sizeof(bool));
 192   ActiveComm* comm=NULL;
 193   if(remove){
 194     if(to->ActiveCommsDown[from]==1)
 195       to->ActiveCommsDown.erase(from);
 196     else
 197       to->ActiveCommsDown[from]-=1;
 198
 199     to->nbActiveCommsDown--;
 200     for (std::vector<ActiveComm*>::iterator it= from->ActiveCommsUp.begin();
 201          it != from->ActiveCommsUp.end(); ++it) {
 202       if((*it)->action==action){
 203         comm=(*it);
 204         from->ActiveCommsUp.erase(it);
 205         break;
 206       }
 207     }
 208     action->unref();
 209
 210   }else{
 211     action->ref();
 212     ActiveComm* comm=new ActiveComm();
 213     comm->action=action;
 214     comm->destination=to;
 215     from->ActiveCommsUp.push_back(comm);
 216
 217     to->ActiveCommsDown[from]+=1;
 218     to->nbActiveCommsDown++;
 219   }
 220   XBT_DEBUG("IB - Updating %d", from->id);
 221   updateIBfactors_rec(from, updated);
 222   XBT_DEBUG("IB - Finished updating %d", from->id);
 223   if(comm)
 224     delete comm;
 225   xbt_free(updated);
 226 }