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
  12 #include "src/surf/HostImpl.hpp"
  13 #include "simgrid/sg_config.h"
  14 #include "maxmin_private.hpp"
  15
  16 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_network);
  17
  18 static void IB_create_host_callback(simgrid::s4u::Host& host){
  19   using namespace simgrid::surf;
  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==NULL)
  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, NULL);
  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 namespace simgrid::surf;
  39   if(statein!=simgrid::surf::Action::State::running || stateout!=simgrid::surf::Action::State::done)
  40     return;
  41   std::pair<IBNode*,IBNode*> pair = ((NetworkIBModel*)surf_network_model)->active_comms[action];
  42   XBT_DEBUG("IB callback - action %p finished", action);
  43
  44   ((NetworkIBModel*)surf_network_model)->updateIBfactors(action, pair.first, pair.second, 1);
  45
  46   ((NetworkIBModel*)surf_network_model)->active_comms.erase(action);
  47
  48 }
  49
  50
  51 static void IB_action_init_callback(
  52     simgrid::surf::NetworkAction *action, simgrid::surf::NetCard *src, simgrid::surf::NetCard *dst)
  53 {
  54   simgrid::surf::NetworkIBModel* ibModel = (simgrid::surf::NetworkIBModel*)surf_network_model;
  55
  56   simgrid::surf::IBNode* act_src= (simgrid::surf::IBNode*) xbt_dict_get_or_null(ibModel->active_nodes, src->name());
  57   xbt_assert(act_src, "could not find src node active comms !");
  58
  59   simgrid::surf::IBNode* act_dst= (simgrid::surf::IBNode*) xbt_dict_get_or_null(ibModel->active_nodes, dst->name());
  60   xbt_assert(act_dst, "could not find dst node active comms !");
  61
  62   ibModel->active_comms[action]=std::make_pair(act_src, act_dst);
  63
  64   ibModel->updateIBfactors(action, act_src, act_dst, 0);
  65 }
  66
  67 /*********
  68  * Model *
  69  *********/
  70
  71 /************************************************************************/
  72 /* New model based on MPI contention model for Infiniband platforms */
  73 /************************************************************************/
  74 /* @Inproceedings{mescal_vienne_phd, */
  75 /*  author={Jérôme Vienne}, */
  76 /*  title={prédiction de performances d’applications de calcul haute performance sur réseau Infiniband}, */
  77 /*  address={Grenoble FRANCE}, */
  78 /*  month=june, */
  79 /*  year={2010} */
  80 /*  } */
  81 void surf_network_model_init_IB(void)
  82 {
  83   using simgrid::surf::networkActionStateChangedCallbacks;
  84
  85   if (surf_network_model)
  86     return;
  87
  88   surf_network_model = new simgrid::surf::NetworkIBModel();
  89   xbt_dynar_push(all_existing_models, &surf_network_model);
  90   networkActionStateChangedCallbacks.connect(IB_action_state_changed_callback);
  91   Link::onCommunicate.connect(IB_action_init_callback);
  92   simgrid::s4u::Host::onCreation.connect(IB_create_host_callback);
  93   xbt_cfg_setdefault_double("network/weight-S", 8775);
  94
  95 }
  96
  97 #include "src/surf/xml/platf.hpp" // FIXME: move that back to the parsing area
  98
  99 namespace simgrid {
 100   namespace surf {
 101
 102     NetworkIBModel::NetworkIBModel()
 103     : NetworkSmpiModel() {
 104       haveGap_=false;
 105       active_nodes=NULL;
 106
 107       const char* IB_factors_string=xbt_cfg_get_string("smpi/IB-penalty-factors");
 108       xbt_dynar_t radical_elements = xbt_str_split(IB_factors_string, ";");
 109
 110       surf_parse_assert(xbt_dynar_length(radical_elements)==3,
 111           "smpi/IB-penalty-factors should be provided and contain 3 elements, semi-colon separated. Example: 0.965;0.925;1.35");
 112
 113       Be = xbt_str_parse_double(xbt_dynar_get_as(radical_elements, 0, char *), "First part of smpi/IB-penalty-factors is not numerical: %s");
 114       Bs = xbt_str_parse_double(xbt_dynar_get_as(radical_elements, 1, char *), "Second part of smpi/IB-penalty-factors is not numerical: %s");
 115       ys = xbt_str_parse_double(xbt_dynar_get_as(radical_elements, 2, char *), "Third part of smpi/IB-penalty-factors is not numerical: %s");
 116
 117       xbt_dynar_free(&radical_elements);
 118     }
 119
 120     NetworkIBModel::~NetworkIBModel()
 121     {
 122       xbt_dict_cursor_t cursor = NULL;
 123       IBNode* instance = NULL;
 124       char *name = NULL;
 125       xbt_dict_foreach(active_nodes, cursor, name, instance)
 126       delete instance;
 127       xbt_dict_free(&active_nodes);
 128     }
 129
 130     void NetworkIBModel::computeIBfactors(IBNode *root) {
 131       double penalized_bw=0.0;
 132       double num_comm_out = (double) root->ActiveCommsUp.size();
 133       double max_penalty_out=0.0;
 134       //first, compute all outbound penalties to get their max
 135       for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 136         double my_penalty_out = 1.0;
 137
 138         if(num_comm_out!=1){
 139           if((*it)->destination->nbActiveCommsDown > 2)//number of comms sent to the receiving node
 140             my_penalty_out = num_comm_out * Bs * ys;
 141           else
 142             my_penalty_out = num_comm_out * Bs;
 143         }
 144
 145         max_penalty_out = std::max(max_penalty_out,my_penalty_out);
 146       }
 147
 148       for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 149
 150         //compute inbound penalty
 151         double my_penalty_in = 1.0;
 152         int nb_comms = (*it)->destination->nbActiveCommsDown;//total number of incoming comms
 153         if(nb_comms!=1)
 154           my_penalty_in = ((*it)->destination->ActiveCommsDown)[root] //number of comm sent to dest by root node
 155                                                                 * Be
 156                                                                 * (*it)->destination->ActiveCommsDown.size();//number of different nodes sending to dest
 157
 158         double penalty = std::max(my_penalty_in,max_penalty_out);
 159
 160         double rate_before_update = (*it)->action->getBound();
 161         //save initial rate of the action
 162         if((*it)->init_rate==-1)
 163           (*it)->init_rate= rate_before_update;
 164
 165         penalized_bw= ! num_comm_out ? (*it)->init_rate : (*it)->init_rate /penalty;
 166
 167         if (!double_equals(penalized_bw, rate_before_update, sg_surf_precision)){
 168           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 );
 169           lmm_update_variable_bound(maxminSystem_, (*it)->action->getVariable(), penalized_bw);
 170         }else{
 171           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 );
 172         }
 173
 174       }
 175       XBT_DEBUG("Finished computing IB penalties");
 176     }
 177
 178     void NetworkIBModel::updateIBfactors_rec(IBNode *root, bool* updatedlist) {
 179       if(updatedlist[root->id]==0){
 180         XBT_DEBUG("IB - Updating rec %d", root->id);
 181         computeIBfactors(root);
 182         updatedlist[root->id]=1;
 183         for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 184           if(updatedlist[(*it)->destination->id]!=1)
 185             updateIBfactors_rec((*it)->destination, updatedlist);
 186         }
 187         for (std::map<IBNode*, int>::iterator it= root->ActiveCommsDown.begin(); it != root->ActiveCommsDown.end(); ++it) {
 188           if(updatedlist[it->first->id]!=1)
 189             updateIBfactors_rec(it->first, updatedlist);
 190         }
 191       }
 192     }
 193
 194
 195     void NetworkIBModel::updateIBfactors(NetworkAction *action, IBNode *from, IBNode * to, int remove) {
 196       if (from == to)//disregard local comms (should use loopback)
 197         return;
 198
 199       bool* updated=(bool*)xbt_malloc0(xbt_dict_size(active_nodes)*sizeof(bool));
 200       ActiveComm* comm=NULL;
 201       if(remove){
 202         if(to->ActiveCommsDown[from]==1)
 203           to->ActiveCommsDown.erase(from);
 204         else
 205           to->ActiveCommsDown[from]-=1;
 206
 207         to->nbActiveCommsDown--;
 208         for (std::vector<ActiveComm*>::iterator it= from->ActiveCommsUp.begin();
 209             it != from->ActiveCommsUp.end(); ++it) {
 210           if((*it)->action==action){
 211             comm=(*it);
 212             from->ActiveCommsUp.erase(it);
 213             break;
 214           }
 215         }
 216         action->unref();
 217
 218       }else{
 219         action->ref();
 220         ActiveComm* comm=new ActiveComm();
 221         comm->action=action;
 222         comm->destination=to;
 223         from->ActiveCommsUp.push_back(comm);
 224
 225         to->ActiveCommsDown[from]+=1;
 226         to->nbActiveCommsDown++;
 227       }
 228       XBT_DEBUG("IB - Updating %d", from->id);
 229       updateIBfactors_rec(from, updated);
 230       XBT_DEBUG("IB - Finished updating %d", from->id);
 231       if(comm)
 232         delete comm;
 233       xbt_free(updated);
 234     }
 235
 236   }
 237 }