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