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();
  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
  53 static void IB_action_init_callback(
  54     simgrid::surf::NetworkAction *action, simgrid::routing::NetCard *src, simgrid::routing::NetCard *dst)
  55 {
  56   simgrid::surf::NetworkIBModel* ibModel = (simgrid::surf::NetworkIBModel*)surf_network_model;
  57
  58   simgrid::surf::IBNode* act_src= (simgrid::surf::IBNode*) xbt_dict_get_or_null(ibModel->active_nodes, src->name());
  59   xbt_assert(act_src, "could not find src node active comms !");
  60
  61   simgrid::surf::IBNode* act_dst= (simgrid::surf::IBNode*) xbt_dict_get_or_null(ibModel->active_nodes, dst->name());
  62   xbt_assert(act_dst, "could not find dst node active comms !");
  63
  64   ibModel->active_comms[action]=std::make_pair(act_src, act_dst);
  65
  66   ibModel->updateIBfactors(action, act_src, act_dst, 0);
  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   using simgrid::surf::networkActionStateChangedCallbacks;
  86
  87   if (surf_network_model)
  88     return;
  89
  90   surf_network_model = new simgrid::surf::NetworkIBModel();
  91   xbt_dynar_push(all_existing_models, &surf_network_model);
  92   networkActionStateChangedCallbacks.connect(IB_action_state_changed_callback);
  93   Link::onCommunicate.connect(IB_action_init_callback);
  94   simgrid::s4u::Host::onCreation.connect(IB_create_host_callback);
  95   xbt_cfg_setdefault_double("network/weight-S", 8775);
  96
  97 }
  98
  99 #include "src/surf/xml/platf.hpp" // FIXME: move that back to the parsing area
 100
 101 namespace simgrid {
 102   namespace surf {
 103
 104     NetworkIBModel::NetworkIBModel()
 105     : NetworkSmpiModel() {
 106       haveGap_=false;
 107       active_nodes=nullptr;
 108
 109       const char* IB_factors_string=xbt_cfg_get_string("smpi/IB-penalty-factors");
 110       xbt_dynar_t radical_elements = xbt_str_split(IB_factors_string, ";");
 111
 112       surf_parse_assert(xbt_dynar_length(radical_elements)==3,
 113           "smpi/IB-penalty-factors should be provided and contain 3 elements, semi-colon separated. Example: 0.965;0.925;1.35");
 114
 115       Be = xbt_str_parse_double(xbt_dynar_get_as(radical_elements, 0, char *), "First part of smpi/IB-penalty-factors is not numerical: %s");
 116       Bs = xbt_str_parse_double(xbt_dynar_get_as(radical_elements, 1, char *), "Second part of smpi/IB-penalty-factors is not numerical: %s");
 117       ys = xbt_str_parse_double(xbt_dynar_get_as(radical_elements, 2, char *), "Third part of smpi/IB-penalty-factors is not numerical: %s");
 118
 119       xbt_dynar_free(&radical_elements);
 120     }
 121
 122     NetworkIBModel::~NetworkIBModel()
 123     {
 124       xbt_dict_cursor_t cursor = nullptr;
 125       IBNode* instance = nullptr;
 126       char *name = nullptr;
 127       xbt_dict_foreach(active_nodes, cursor, name, instance)
 128       delete instance;
 129       xbt_dict_free(&active_nodes);
 130     }
 131
 132     void NetworkIBModel::computeIBfactors(IBNode *root) {
 133       double penalized_bw=0.0;
 134       double num_comm_out = (double) root->ActiveCommsUp.size();
 135       double max_penalty_out=0.0;
 136       //first, compute all outbound penalties to get their max
 137       for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 138         double my_penalty_out = 1.0;
 139
 140         if(num_comm_out!=1){
 141           if((*it)->destination->nbActiveCommsDown > 2)//number of comms sent to the receiving node
 142             my_penalty_out = num_comm_out * Bs * ys;
 143           else
 144             my_penalty_out = num_comm_out * Bs;
 145         }
 146
 147         max_penalty_out = std::max(max_penalty_out,my_penalty_out);
 148       }
 149
 150       for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 151
 152         //compute inbound penalty
 153         double my_penalty_in = 1.0;
 154         int nb_comms = (*it)->destination->nbActiveCommsDown;//total number of incoming comms
 155         if(nb_comms!=1)
 156           my_penalty_in = ((*it)->destination->ActiveCommsDown)[root] //number of comm sent to dest by root node
 157                                                                 * Be
 158                                                                 * (*it)->destination->ActiveCommsDown.size();//number of different nodes sending to dest
 159
 160         double penalty = std::max(my_penalty_in,max_penalty_out);
 161
 162         double rate_before_update = (*it)->action->getBound();
 163         //save initial rate of the action
 164         if((*it)->init_rate==-1)
 165           (*it)->init_rate= rate_before_update;
 166
 167         penalized_bw= ! num_comm_out ? (*it)->init_rate : (*it)->init_rate /penalty;
 168
 169         if (!double_equals(penalized_bw, rate_before_update, sg_surf_precision)){
 170           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 );
 171           lmm_update_variable_bound(maxminSystem_, (*it)->action->getVariable(), penalized_bw);
 172         }else{
 173           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 );
 174         }
 175
 176       }
 177       XBT_DEBUG("Finished computing IB penalties");
 178     }
 179
 180     void NetworkIBModel::updateIBfactors_rec(IBNode *root, bool* updatedlist) {
 181       if(updatedlist[root->id]==0){
 182         XBT_DEBUG("IB - Updating rec %d", root->id);
 183         computeIBfactors(root);
 184         updatedlist[root->id]=1;
 185         for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 186           if(updatedlist[(*it)->destination->id]!=1)
 187             updateIBfactors_rec((*it)->destination, updatedlist);
 188         }
 189         for (std::map<IBNode*, int>::iterator it= root->ActiveCommsDown.begin(); it != root->ActiveCommsDown.end(); ++it) {
 190           if(updatedlist[it->first->id]!=1)
 191             updateIBfactors_rec(it->first, updatedlist);
 192         }
 193       }
 194     }
 195
 196
 197     void NetworkIBModel::updateIBfactors(NetworkAction *action, IBNode *from, IBNode * to, int remove) {
 198       if (from == to)//disregard local comms (should use loopback)
 199         return;
 200
 201       bool* updated=(bool*)xbt_malloc0(xbt_dict_size(active_nodes)*sizeof(bool));
 202       ActiveComm* comm=nullptr;
 203       if(remove){
 204         if(to->ActiveCommsDown[from]==1)
 205           to->ActiveCommsDown.erase(from);
 206         else
 207           to->ActiveCommsDown[from]-=1;
 208
 209         to->nbActiveCommsDown--;
 210         for (std::vector<ActiveComm*>::iterator it= from->ActiveCommsUp.begin();
 211             it != from->ActiveCommsUp.end(); ++it) {
 212           if((*it)->action==action){
 213             comm=(*it);
 214             from->ActiveCommsUp.erase(it);
 215             break;
 216           }
 217         }
 218         action->unref();
 219
 220       }else{
 221         action->ref();
 222         ActiveComm* comm=new ActiveComm();
 223         comm->action=action;
 224         comm->destination=to;
 225         from->ActiveCommsUp.push_back(comm);
 226
 227         to->ActiveCommsDown[from]+=1;
 228         to->nbActiveCommsDown++;
 229       }
 230       XBT_DEBUG("IB - Updating %d", from->id);
 231       updateIBfactors_rec(from, updated);
 232       XBT_DEBUG("IB - Finished updating %d", from->id);
 233       if(comm)
 234         delete comm;
 235       xbt_free(updated);
 236     }
 237
 238   }
 239 }