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