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     double size, double rate)
  54 {
  55   using namespace simgrid::surf;
  56   if(((NetworkIBModel*)surf_network_model)->active_nodes==NULL)
  57     xbt_die("IB comm added, without any node connected !");
  58
  59   IBNode* act_src= (IBNode*) xbt_dict_get_or_null(((NetworkIBModel*)surf_network_model)->active_nodes, src->name());
  60   if(act_src==NULL)
  61     xbt_die("could not find src node active comms !");
  62   //act_src->rate=rate;
  63
  64   IBNode* act_dst= (IBNode*) xbt_dict_get_or_null(((NetworkIBModel*)surf_network_model)->active_nodes, dst->name());
  65   if(act_dst==NULL)
  66     xbt_die("could not find dst node active comms !");
  67   // act_dst->rate=rate;
  68
  69   ((NetworkIBModel*)surf_network_model)->active_comms[action]=std::make_pair(act_src, act_dst);
  70   //post the action in the second dist, to retrieve in the other callback
  71   XBT_DEBUG("IB callback - action %p init", action);
  72
  73   ((NetworkIBModel*)surf_network_model)->updateIBfactors(action, act_src, act_dst, 0);
  74
  75 }
  76
  77 /*********
  78  * Model *
  79  *********/
  80
  81 /************************************************************************/
  82 /* New model based on MPI contention model for Infiniband platforms */
  83 /************************************************************************/
  84 /* @Inproceedings{mescal_vienne_phd, */
  85 /*  author={Jérôme Vienne}, */
  86 /*  title={prédiction de performances d’applications de calcul haute performance sur réseau Infiniband}, */
  87 /*  address={Grenoble FRANCE}, */
  88 /*  month=june, */
  89 /*  year={2010} */
  90 /*  } */
  91 void surf_network_model_init_IB(void)
  92 {
  93   using simgrid::surf::networkActionStateChangedCallbacks;
  94   using simgrid::surf::networkCommunicateCallbacks;
  95
  96   if (surf_network_model)
  97     return;
  98
  99   simgrid::surf::on_link.connect(netlink_parse_init);
 100   surf_network_model = new simgrid::surf::NetworkIBModel();
 101   xbt_dynar_push(all_existing_models, &surf_network_model);
 102   networkActionStateChangedCallbacks.connect(IB_action_state_changed_callback);
 103   networkCommunicateCallbacks.connect(IB_action_init_callback);
 104   simgrid::s4u::Host::onCreation.connect(IB_create_host_callback);
 105   xbt_cfg_setdefault_double(_sg_cfg_set, "network/weight_S", 8775);
 106
 107 }
 108
 109 #include "src/surf/xml/platf.hpp" // FIXME: move that back to the parsing area
 110
 111 namespace simgrid {
 112   namespace surf {
 113
 114     NetworkIBModel::NetworkIBModel()
 115     : NetworkSmpiModel() {
 116       haveGap_=false;
 117       active_nodes=NULL;
 118
 119       const char* IB_factors_string=sg_cfg_get_string("smpi/IB_penalty_factors");
 120       xbt_dynar_t radical_elements = xbt_str_split(IB_factors_string, ";");
 121
 122       surf_parse_assert(xbt_dynar_length(radical_elements)==3,
 123           "smpi/IB_penalty_factors should be provided and contain 3 elements, semi-colon separated : for example 0.965;0.925;1.35");
 124
 125       Be = xbt_str_parse_double(xbt_dynar_get_as(radical_elements, 0, char *), "First part of smpi/IB_penalty_factors is not numerical: %s");
 126       Bs = xbt_str_parse_double(xbt_dynar_get_as(radical_elements, 1, char *), "Second part of smpi/IB_penalty_factors is not numerical: %s");
 127       ys = xbt_str_parse_double(xbt_dynar_get_as(radical_elements, 2, char *), "Third part of smpi/IB_penalty_factors is not numerical: %s");
 128
 129       xbt_dynar_free(&radical_elements);
 130     }
 131
 132     NetworkIBModel::~NetworkIBModel()
 133     {
 134       xbt_dict_cursor_t cursor = NULL;
 135       IBNode* instance = NULL;
 136       char *name = NULL;
 137       xbt_dict_foreach(active_nodes, cursor, name, instance)
 138       delete instance;
 139       xbt_dict_free(&active_nodes);
 140     }
 141
 142     void NetworkIBModel::computeIBfactors(IBNode *root) {
 143       double penalized_bw=0.0;
 144       double num_comm_out = (double) root->ActiveCommsUp.size();
 145       double max_penalty_out=0.0;
 146       //first, compute all outbound penalties to get their max
 147       for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 148         double my_penalty_out = 1.0;
 149
 150         if(num_comm_out!=1){
 151           if((*it)->destination->nbActiveCommsDown > 2)//number of comms sent to the receiving node
 152             my_penalty_out = num_comm_out * Bs * ys;
 153           else
 154             my_penalty_out = num_comm_out * Bs;
 155         }
 156
 157         max_penalty_out = std::max(max_penalty_out,my_penalty_out);
 158       }
 159
 160       for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 161
 162         //compute inbound penalty
 163         double my_penalty_in = 1.0;
 164         int nb_comms = (*it)->destination->nbActiveCommsDown;//total number of incoming comms
 165         if(nb_comms!=1)
 166           my_penalty_in = ((*it)->destination->ActiveCommsDown)[root] //number of comm sent to dest by root node
 167                                                                 * Be
 168                                                                 * (*it)->destination->ActiveCommsDown.size();//number of different nodes sending to dest
 169
 170         double penalty = std::max(my_penalty_in,max_penalty_out);
 171
 172         double rate_before_update = (*it)->action->getBound();
 173         //save initial rate of the action
 174         if((*it)->init_rate==-1)
 175           (*it)->init_rate= rate_before_update;
 176
 177         penalized_bw= ! num_comm_out ? (*it)->init_rate : (*it)->init_rate /penalty;
 178
 179         if (!double_equals(penalized_bw, rate_before_update, sg_surf_precision)){
 180           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 );
 181           lmm_update_variable_bound(maxminSystem_, (*it)->action->getVariable(), penalized_bw);
 182         }else{
 183           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 );
 184         }
 185
 186       }
 187       XBT_DEBUG("Finished computing IB penalties");
 188     }
 189
 190     void NetworkIBModel::updateIBfactors_rec(IBNode *root, bool* updatedlist) {
 191       if(updatedlist[root->id]==0){
 192         XBT_DEBUG("IB - Updating rec %d", root->id);
 193         computeIBfactors(root);
 194         updatedlist[root->id]=1;
 195         for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
 196           if(updatedlist[(*it)->destination->id]!=1)
 197             updateIBfactors_rec((*it)->destination, updatedlist);
 198         }
 199         for (std::map<IBNode*, int>::iterator it= root->ActiveCommsDown.begin(); it != root->ActiveCommsDown.end(); ++it) {
 200           if(updatedlist[it->first->id]!=1)
 201             updateIBfactors_rec(it->first, updatedlist);
 202         }
 203       }
 204     }
 205
 206
 207     void NetworkIBModel::updateIBfactors(NetworkAction *action, IBNode *from, IBNode * to, int remove) {
 208       if (from == to)//disregard local comms (should use loopback)
 209         return;
 210
 211       bool* updated=(bool*)xbt_malloc0(xbt_dict_size(active_nodes)*sizeof(bool));
 212       ActiveComm* comm=NULL;
 213       if(remove){
 214         if(to->ActiveCommsDown[from]==1)
 215           to->ActiveCommsDown.erase(from);
 216         else
 217           to->ActiveCommsDown[from]-=1;
 218
 219         to->nbActiveCommsDown--;
 220         for (std::vector<ActiveComm*>::iterator it= from->ActiveCommsUp.begin();
 221             it != from->ActiveCommsUp.end(); ++it) {
 222           if((*it)->action==action){
 223             comm=(*it);
 224             from->ActiveCommsUp.erase(it);
 225             break;
 226           }
 227         }
 228         action->unref();
 229
 230       }else{
 231         action->ref();
 232         ActiveComm* comm=new ActiveComm();
 233         comm->action=action;
 234         comm->destination=to;
 235         from->ActiveCommsUp.push_back(comm);
 236
 237         to->ActiveCommsDown[from]+=1;
 238         to->nbActiveCommsDown++;
 239       }
 240       XBT_DEBUG("IB - Updating %d", from->id);
 241       updateIBfactors_rec(from, updated);
 242       XBT_DEBUG("IB - Finished updating %d", from->id);
 243       if(comm)
 244         delete comm;
 245       xbt_free(updated);
 246     }
 247
 248   }
 249 }