src/kernel/routing/DragonflyZone.cpp

   1 /* Copyright (c) 2014-2019. The SimGrid Team. All rights reserved.          */
   2
   3 /* This program is free software; you can redistribute it and/or modify it
   4  * under the terms of the license (GNU LGPL) which comes with this package. */
   5
   6 #include "simgrid/kernel/routing/DragonflyZone.hpp"
   7 #include "simgrid/kernel/routing/NetPoint.hpp"
   8 #include "src/surf/network_interface.hpp"
   9 #include "src/surf/xml/platf_private.hpp"
  10
  11 #include <boost/algorithm/string/classification.hpp>
  12 #include <boost/algorithm/string/split.hpp>
  13 #include <string>
  14
  15 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_route_cluster_dragonfly, surf_route_cluster, "Dragonfly Routing part of surf");
  16
  17 namespace simgrid {
  18 namespace kernel {
  19 namespace routing {
  20
  21 DragonflyZone::DragonflyZone(NetZoneImpl* father, std::string name, resource::NetworkModel* netmodel)
  22     : ClusterZone(father, std::move(name), netmodel)
  23 {
  24 }
  25
  26 DragonflyZone::~DragonflyZone()
  27 {
  28   if (this->routers_ != nullptr) {
  29     for (unsigned int i = 0; i < this->num_groups_ * this->num_chassis_per_group_ * this->num_blades_per_chassis_; i++)
  30       delete routers_[i];
  31     delete[] routers_;
  32   }
  33 }
  34
  35 void DragonflyZone::rankId_to_coords(int rankId, unsigned int coords[4])
  36 {
  37   // coords : group, chassis, blade, node
  38   coords[0] = rankId / (num_chassis_per_group_ * num_blades_per_chassis_ * num_nodes_per_blade_);
  39   rankId    = rankId % (num_chassis_per_group_ * num_blades_per_chassis_ * num_nodes_per_blade_);
  40   coords[1] = rankId / (num_blades_per_chassis_ * num_nodes_per_blade_);
  41   rankId    = rankId % (num_blades_per_chassis_ * num_nodes_per_blade_);
  42   coords[2] = rankId / num_nodes_per_blade_;
  43   coords[3] = rankId % num_nodes_per_blade_;
  44 }
  45
  46 void DragonflyZone::parse_specific_arguments(ClusterCreationArgs* cluster)
  47 {
  48   std::vector<std::string> parameters;
  49   std::vector<std::string> tmp;
  50   boost::split(parameters, cluster->topo_parameters, boost::is_any_of(";"));
  51
  52   if (parameters.size() != 4 || parameters.empty()) {
  53     surf_parse_error(
  54         "Dragonfly are defined by the number of groups, chassis per groups, blades per chassis, nodes per blade");
  55   }
  56
  57   // Blue network : number of groups, number of links between each group
  58   boost::split(tmp, parameters[0], boost::is_any_of(","));
  59   if (tmp.size() != 2) {
  60     surf_parse_error("Dragonfly topologies are defined by 3 levels with 2 elements each, and one with one element");
  61   }
  62
  63   try {
  64     this->num_groups_ = std::stoi(tmp[0]);
  65   } catch (std::invalid_argument& ia) {
  66     throw std::invalid_argument(std::string("Invalid number of groups:") + tmp[0]);
  67   }
  68
  69   try {
  70     this->num_links_blue_ = std::stoi(tmp[1]);
  71   } catch (std::invalid_argument& ia) {
  72     throw std::invalid_argument(std::string("Invalid number of links for the blue level:") + tmp[1]);
  73   }
  74   // Black network : number of chassis/group, number of links between each router on the black network
  75   boost::split(tmp, parameters[1], boost::is_any_of(","));
  76   if (tmp.size() != 2) {
  77     surf_parse_error("Dragonfly topologies are defined by 3 levels with 2 elements each, and one with one element");
  78   }
  79
  80   try {
  81     this->num_chassis_per_group_ = std::stoi(tmp[0]);
  82   } catch (std::invalid_argument& ia) {
  83     throw std::invalid_argument(std::string("Invalid number of groups:") + tmp[0]);
  84   }
  85
  86   try {
  87     this->num_links_black_ = std::stoi(tmp[1]);
  88   } catch (std::invalid_argument& ia) {
  89     throw std::invalid_argument(std::string("Invalid number of links for the black level:") + tmp[1]);
  90   }
  91
  92   // Green network : number of blades/chassis, number of links between each router on the green network
  93   boost::split(tmp, parameters[2], boost::is_any_of(","));
  94   if (tmp.size() != 2) {
  95     surf_parse_error("Dragonfly topologies are defined by 3 levels with 2 elements each, and one with one element");
  96   }
  97
  98   try {
  99     this->num_blades_per_chassis_ = std::stoi(tmp[0]);
 100   } catch (std::invalid_argument& ia) {
 101     throw std::invalid_argument(std::string("Invalid number of groups:") + tmp[0]);
 102   }
 103
 104   try {
 105     this->num_links_green_ = std::stoi(tmp[1]);
 106   } catch (std::invalid_argument& ia) {
 107     throw std::invalid_argument(std::string("Invalid number of links for the green level:") + tmp[1]);
 108   }
 109
 110   // The last part of topo_parameters should be the number of nodes per blade
 111   try {
 112     this->num_nodes_per_blade_ = std::stoi(parameters[3]);
 113   } catch (std::invalid_argument& ia) {
 114     throw std::invalid_argument(std::string("Last parameter is not the amount of nodes per blade:") + parameters[3]);
 115   }
 116
 117   this->sharing_policy_ = cluster->sharing_policy;
 118   if (cluster->sharing_policy == s4u::Link::SharingPolicy::SPLITDUPLEX)
 119     this->num_links_per_link_ = 2;
 120   this->bw_  = cluster->bw;
 121   this->lat_ = cluster->lat;
 122 }
 123
 124 /* Generate the cluster once every node is created */
 125 void DragonflyZone::seal()
 126 {
 127   if (this->num_nodes_per_blade_ == 0) {
 128     return;
 129   }
 130
 131   this->generate_routers();
 132   this->generate_links();
 133 }
 134
 135 DragonflyRouter::DragonflyRouter(int group, int chassis, int blade) : group_(group), chassis_(chassis), blade_(blade)
 136 {
 137 }
 138
 139 DragonflyRouter::~DragonflyRouter()
 140 {
 141   delete[] my_nodes_;
 142   delete[] green_links_;
 143   delete[] black_links_;
 144   delete blue_links_;
 145 }
 146
 147 void DragonflyZone::generate_routers()
 148 {
 149   this->routers_ =
 150       new DragonflyRouter*[this->num_groups_ * this->num_chassis_per_group_ * this->num_blades_per_chassis_];
 151
 152   for (unsigned int i = 0; i < this->num_groups_; i++) {
 153     for (unsigned int j = 0; j < this->num_chassis_per_group_; j++) {
 154       for (unsigned int k = 0; k < this->num_blades_per_chassis_; k++) {
 155         DragonflyRouter* router = new DragonflyRouter(i, j, k);
 156         this->routers_[i * this->num_chassis_per_group_ * this->num_blades_per_chassis_ +
 157                        j * this->num_blades_per_chassis_ + k] = router;
 158       }
 159     }
 160   }
 161 }
 162
 163 void DragonflyZone::create_link(std::string id, int numlinks, resource::LinkImpl** linkup,
 164                                 resource::LinkImpl** linkdown)
 165 {
 166   *linkup   = nullptr;
 167   *linkdown = nullptr;
 168   LinkCreationArgs linkTemplate;
 169   linkTemplate.bandwidth = this->bw_ * numlinks;
 170   linkTemplate.latency   = this->lat_;
 171   linkTemplate.policy    = this->sharing_policy_;
 172   linkTemplate.id        = std::move(id);
 173   sg_platf_new_link(&linkTemplate);
 174   XBT_DEBUG("Generating link %s", linkTemplate.id.c_str());
 175   resource::LinkImpl* link;
 176   if (this->sharing_policy_ == s4u::Link::SharingPolicy::SPLITDUPLEX) {
 177     *linkup   = s4u::Link::by_name(linkTemplate.id + "_UP")->get_impl();   // check link?
 178     *linkdown = s4u::Link::by_name(linkTemplate.id + "_DOWN")->get_impl(); // check link ?
 179   } else {
 180     link      = s4u::Link::by_name(linkTemplate.id)->get_impl();
 181     *linkup   = link;
 182     *linkdown = link;
 183   }
 184 }
 185
 186 void DragonflyZone::generate_links()
 187 {
 188   static int uniqueId = 0;
 189   resource::LinkImpl* linkup;
 190   resource::LinkImpl* linkdown;
 191
 192   unsigned int numRouters = this->num_groups_ * this->num_chassis_per_group_ * this->num_blades_per_chassis_;
 193
 194   // Links from routers to their local nodes.
 195   for (unsigned int i = 0; i < numRouters; i++) {
 196     // allocate structures
 197     this->routers_[i]->my_nodes_    = new resource::LinkImpl*[num_links_per_link_ * this->num_nodes_per_blade_];
 198     this->routers_[i]->green_links_ = new resource::LinkImpl*[this->num_blades_per_chassis_];
 199     this->routers_[i]->black_links_ = new resource::LinkImpl*[this->num_chassis_per_group_];
 200
 201     for (unsigned int j = 0; j < num_links_per_link_ * this->num_nodes_per_blade_; j += num_links_per_link_) {
 202       std::string id = "local_link_from_router_" + std::to_string(i) + "_to_node_" +
 203                        std::to_string(j / num_links_per_link_) + "_" + std::to_string(uniqueId);
 204       this->create_link(id, 1, &linkup, &linkdown);
 205
 206       this->routers_[i]->my_nodes_[j] = linkup;
 207       if (this->sharing_policy_ == s4u::Link::SharingPolicy::SPLITDUPLEX)
 208         this->routers_[i]->my_nodes_[j + 1] = linkdown;
 209
 210       uniqueId++;
 211     }
 212   }
 213
 214   // Green links from routers to same chassis routers - alltoall
 215   for (unsigned int i = 0; i < this->num_groups_ * this->num_chassis_per_group_; i++) {
 216     for (unsigned int j = 0; j < this->num_blades_per_chassis_; j++) {
 217       for (unsigned int k = j + 1; k < this->num_blades_per_chassis_; k++) {
 218         std::string id = "green_link_in_chassis_" + std::to_string(i % num_chassis_per_group_) + "_between_routers_" +
 219                          std::to_string(j) + "_and_" + std::to_string(k) + "_" + std::to_string(uniqueId);
 220         this->create_link(id, this->num_links_green_, &linkup, &linkdown);
 221
 222         this->routers_[i * num_blades_per_chassis_ + j]->green_links_[k] = linkup;
 223         this->routers_[i * num_blades_per_chassis_ + k]->green_links_[j] = linkdown;
 224         uniqueId++;
 225       }
 226     }
 227   }
 228
 229   // Black links from routers to same group routers - alltoall
 230   for (unsigned int i = 0; i < this->num_groups_; i++) {
 231     for (unsigned int j = 0; j < this->num_chassis_per_group_; j++) {
 232       for (unsigned int k = j + 1; k < this->num_chassis_per_group_; k++) {
 233         for (unsigned int l = 0; l < this->num_blades_per_chassis_; l++) {
 234           std::string id = "black_link_in_group_" + std::to_string(i) + "_between_chassis_" + std::to_string(j) +
 235               "_and_" + std::to_string(k) +"_blade_" + std::to_string(l) + "_" + std::to_string(uniqueId);
 236           this->create_link(id, this->num_links_black_, &linkup, &linkdown);
 237
 238           this->routers_[i * num_blades_per_chassis_ * num_chassis_per_group_ + j * num_blades_per_chassis_ + l]
 239               ->black_links_[k] = linkup;
 240           this->routers_[i * num_blades_per_chassis_ * num_chassis_per_group_ + k * num_blades_per_chassis_ + l]
 241               ->black_links_[j] = linkdown;
 242           uniqueId++;
 243         }
 244       }
 245     }
 246   }
 247
 248   // Blue links between groups - Not all routers involved, only one per group is linked to others. Let's say router n of
 249   // each group is linked to group n.
 250   // FIXME: in reality blue links may be attached to several different routers
 251   for (unsigned int i = 0; i < this->num_groups_; i++) {
 252     for (unsigned int j = i + 1; j < this->num_groups_; j++) {
 253       unsigned int routernumi                 = i * num_blades_per_chassis_ * num_chassis_per_group_ + j;
 254       unsigned int routernumj                 = j * num_blades_per_chassis_ * num_chassis_per_group_ + i;
 255       this->routers_[routernumi]->blue_links_ = new resource::LinkImpl*;
 256       this->routers_[routernumj]->blue_links_ = new resource::LinkImpl*;
 257       std::string id = "blue_link_between_group_"+ std::to_string(i) +"_and_" + std::to_string(j) +"_routers_" +
 258           std::to_string(routernumi) + "_and_" + std::to_string(routernumj) + "_" + std::to_string(uniqueId);
 259       this->create_link(id, this->num_links_blue_, &linkup, &linkdown);
 260
 261       this->routers_[routernumi]->blue_links_[0] = linkup;
 262       this->routers_[routernumj]->blue_links_[0] = linkdown;
 263       uniqueId++;
 264     }
 265   }
 266 }
 267
 268 void DragonflyZone::get_local_route(NetPoint* src, NetPoint* dst, RouteCreationArgs* route, double* latency)
 269 {
 270   // Minimal routing version.
 271   // TODO : non-minimal random one, and adaptive ?
 272
 273   if (dst->is_router() || src->is_router())
 274     return;
 275
 276   XBT_VERB("dragonfly getLocalRoute from '%s'[%u] to '%s'[%u]", src->get_cname(), src->id(), dst->get_cname(),
 277            dst->id());
 278
 279   if ((src->id() == dst->id()) && has_loopback_) {
 280     std::pair<resource::LinkImpl*, resource::LinkImpl*> info = private_links_.at(node_pos(src->id()));
 281
 282     route->link_list.push_back(info.first);
 283     if (latency)
 284       *latency += info.first->get_latency();
 285     return;
 286   }
 287
 288   unsigned int myCoords[4];
 289   rankId_to_coords(src->id(), myCoords);
 290   unsigned int targetCoords[4];
 291   rankId_to_coords(dst->id(), targetCoords);
 292   XBT_DEBUG("src : %u group, %u chassis, %u blade, %u node", myCoords[0], myCoords[1], myCoords[2], myCoords[3]);
 293   XBT_DEBUG("dst : %u group, %u chassis, %u blade, %u node", targetCoords[0], targetCoords[1], targetCoords[2],
 294             targetCoords[3]);
 295
 296   DragonflyRouter* myRouter      = routers_[myCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) +
 297                                        myCoords[1] * num_blades_per_chassis_ + myCoords[2]];
 298   DragonflyRouter* targetRouter  = routers_[targetCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) +
 299                                            targetCoords[1] * num_blades_per_chassis_ + targetCoords[2]];
 300   DragonflyRouter* currentRouter = myRouter;
 301
 302   // node->router local link
 303   route->link_list.push_back(myRouter->my_nodes_[myCoords[3] * num_links_per_link_]);
 304   if (latency)
 305     *latency += myRouter->my_nodes_[myCoords[3] * num_links_per_link_]->get_latency();
 306
 307   if (has_limiter_) { // limiter for sender
 308     std::pair<resource::LinkImpl*, resource::LinkImpl*> info = private_links_.at(node_pos_with_loopback(src->id()));
 309     route->link_list.push_back(info.first);
 310   }
 311
 312   if (targetRouter != myRouter) {
 313
 314     // are we on a different group ?
 315     if (targetRouter->group_ != currentRouter->group_) {
 316       // go to the router of our group connected to this one.
 317       if (currentRouter->blade_ != targetCoords[0]) {
 318         // go to the nth router in our chassis
 319         route->link_list.push_back(currentRouter->green_links_[targetCoords[0]]);
 320         if (latency)
 321           *latency += currentRouter->green_links_[targetCoords[0]]->get_latency();
 322         currentRouter = routers_[myCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) +
 323                                  myCoords[1] * num_blades_per_chassis_ + targetCoords[0]];
 324       }
 325
 326       if (currentRouter->chassis_ != 0) {
 327         // go to the first chassis of our group
 328         route->link_list.push_back(currentRouter->black_links_[0]);
 329         if (latency)
 330           *latency += currentRouter->black_links_[0]->get_latency();
 331         currentRouter = routers_[myCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) + targetCoords[0]];
 332       }
 333
 334       // go to destination group - the only optical hop
 335       route->link_list.push_back(currentRouter->blue_links_[0]);
 336       if (latency)
 337         *latency += currentRouter->blue_links_[0]->get_latency();
 338       currentRouter = routers_[targetCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) + myCoords[0]];
 339     }
 340
 341     // same group, but same blade ?
 342     if (targetRouter->blade_ != currentRouter->blade_) {
 343       route->link_list.push_back(currentRouter->green_links_[targetCoords[2]]);
 344       if (latency)
 345         *latency += currentRouter->green_links_[targetCoords[2]]->get_latency();
 346       currentRouter = routers_[targetCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) + targetCoords[2]];
 347     }
 348
 349     // same blade, but same chassis ?
 350     if (targetRouter->chassis_ != currentRouter->chassis_) {
 351       route->link_list.push_back(currentRouter->black_links_[targetCoords[1]]);
 352       if (latency)
 353         *latency += currentRouter->black_links_[targetCoords[1]]->get_latency();
 354     }
 355   }
 356
 357   if (has_limiter_) { // limiter for receiver
 358     std::pair<resource::LinkImpl*, resource::LinkImpl*> info = private_links_.at(node_pos_with_loopback(dst->id()));
 359     route->link_list.push_back(info.first);
 360   }
 361
 362   // router->node local link
 363   route->link_list.push_back(targetRouter->my_nodes_[targetCoords[3] * num_links_per_link_ + num_links_per_link_ - 1]);
 364   if (latency)
 365     *latency += targetRouter->my_nodes_[targetCoords[3] * num_links_per_link_ + num_links_per_link_ - 1]->get_latency();
 366 }
 367 }
 368 }
 369 } // namespace