src/kernel/routing/DragonflyZone.cpp

   1 /* Copyright (c) 2014-2018. 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) : ClusterZone(father, name) {}
  22
  23 DragonflyZone::~DragonflyZone()
  24 {
  25   if (this->routers_ != nullptr) {
  26     for (unsigned int i = 0; i < this->num_groups_ * this->num_chassis_per_group_ * this->num_blades_per_chassis_; i++)
  27       delete routers_[i];
  28     delete[] routers_;
  29   }
  30 }
  31
  32 void DragonflyZone::rankId_to_coords(int rankId, unsigned int coords[4])
  33 {
  34   // coords : group, chassis, blade, node
  35   coords[0] = rankId / (num_chassis_per_group_ * num_blades_per_chassis_ * num_nodes_per_blade_);
  36   rankId    = rankId % (num_chassis_per_group_ * num_blades_per_chassis_ * num_nodes_per_blade_);
  37   coords[1] = rankId / (num_blades_per_chassis_ * num_nodes_per_blade_);
  38   rankId    = rankId % (num_blades_per_chassis_ * num_nodes_per_blade_);
  39   coords[2] = rankId / num_nodes_per_blade_;
  40   coords[3] = rankId % num_nodes_per_blade_;
  41 }
  42
  43 void DragonflyZone::parse_specific_arguments(ClusterCreationArgs* cluster)
  44 {
  45   std::vector<std::string> parameters;
  46   std::vector<std::string> tmp;
  47   boost::split(parameters, cluster->topo_parameters, boost::is_any_of(";"));
  48
  49   if (parameters.size() != 4 || parameters.empty()) {
  50     surf_parse_error(
  51         "Dragonfly are defined by the number of groups, chassis per groups, blades per chassis, nodes per blade");
  52   }
  53
  54   // Blue network : number of groups, number of links between each group
  55   boost::split(tmp, parameters[0], boost::is_any_of(","));
  56   if (tmp.size() != 2) {
  57     surf_parse_error("Dragonfly topologies are defined by 3 levels with 2 elements each, and one with one element");
  58   }
  59
  60   try {
  61     this->num_groups_ = std::stoi(tmp[0]);
  62   } catch (std::invalid_argument& ia) {
  63     throw std::invalid_argument(std::string("Invalid number of groups:") + tmp[0]);
  64   }
  65
  66   try {
  67     this->num_links_blue_ = std::stoi(tmp[1]);
  68   } catch (std::invalid_argument& ia) {
  69     throw std::invalid_argument(std::string("Invalid number of links for the blue level:") + tmp[1]);
  70   }
  71   // Black network : number of chassis/group, number of links between each router on the black network
  72   boost::split(tmp, parameters[1], boost::is_any_of(","));
  73   if (tmp.size() != 2) {
  74     surf_parse_error("Dragonfly topologies are defined by 3 levels with 2 elements each, and one with one element");
  75   }
  76
  77   try {
  78     this->num_chassis_per_group_ = std::stoi(tmp[0]);
  79   } catch (std::invalid_argument& ia) {
  80     throw std::invalid_argument(std::string("Invalid number of groups:") + tmp[0]);
  81   }
  82
  83   try {
  84     this->num_links_black_ = std::stoi(tmp[1]);
  85   } catch (std::invalid_argument& ia) {
  86     throw std::invalid_argument(std::string("Invalid number of links for the black level:") + tmp[1]);
  87   }
  88
  89   // Green network : number of blades/chassis, number of links between each router on the green network
  90   boost::split(tmp, parameters[2], boost::is_any_of(","));
  91   if (tmp.size() != 2) {
  92     surf_parse_error("Dragonfly topologies are defined by 3 levels with 2 elements each, and one with one element");
  93   }
  94
  95   try {
  96     this->num_blades_per_chassis_ = std::stoi(tmp[0]);
  97   } catch (std::invalid_argument& ia) {
  98     throw std::invalid_argument(std::string("Invalid number of groups:") + tmp[0]);
  99   }
 100
 101   try {
 102     this->num_links_green_ = std::stoi(tmp[1]);
 103   } catch (std::invalid_argument& ia) {
 104     throw std::invalid_argument(std::string("Invalid number of links for the green level:") + tmp[1]);
 105   }
 106
 107   // The last part of topo_parameters should be the number of nodes per blade
 108   try {
 109     this->num_nodes_per_blade_ = std::stoi(parameters[3]);
 110   } catch (std::invalid_argument& ia) {
 111     throw std::invalid_argument(std::string("Last parameter is not the amount of nodes per blade:") + parameters[3]);
 112   }
 113
 114   this->sharing_policy_ = cluster->sharing_policy;
 115   if (cluster->sharing_policy == s4u::Link::SharingPolicy::SPLITDUPLEX)
 116     this->num_links_per_link_ = 2;
 117   this->bw_  = cluster->bw;
 118   this->lat_ = cluster->lat;
 119 }
 120
 121 /* Generate the cluster once every node is created */
 122 void DragonflyZone::seal()
 123 {
 124   if (this->num_nodes_per_blade_ == 0) {
 125     return;
 126   }
 127
 128   this->generate_routers();
 129   this->generate_links();
 130 }
 131
 132 DragonflyRouter::DragonflyRouter(int group, int chassis, int blade) : group_(group), chassis_(chassis), blade_(blade)
 133 {
 134 }
 135
 136 DragonflyRouter::~DragonflyRouter()
 137 {
 138   delete[] my_nodes_;
 139   delete[] green_links_;
 140   delete[] black_links_;
 141   delete blue_links_;
 142 }
 143
 144 void DragonflyZone::generate_routers()
 145 {
 146   this->routers_ =
 147       new DragonflyRouter*[this->num_groups_ * this->num_chassis_per_group_ * this->num_blades_per_chassis_];
 148
 149   for (unsigned int i = 0; i < this->num_groups_; i++) {
 150     for (unsigned int j = 0; j < this->num_chassis_per_group_; j++) {
 151       for (unsigned int k = 0; k < this->num_blades_per_chassis_; k++) {
 152         DragonflyRouter* router = new DragonflyRouter(i, j, k);
 153         this->routers_[i * this->num_chassis_per_group_ * this->num_blades_per_chassis_ +
 154                        j * this->num_blades_per_chassis_ + k] = router;
 155       }
 156     }
 157   }
 158 }
 159
 160 void DragonflyZone::create_link(const std::string& id, int numlinks, resource::LinkImpl** linkup,
 161                                 resource::LinkImpl** linkdown)
 162 {
 163   *linkup   = nullptr;
 164   *linkdown = nullptr;
 165   LinkCreationArgs linkTemplate;
 166   linkTemplate.bandwidth = this->bw_ * numlinks;
 167   linkTemplate.latency   = this->lat_;
 168   linkTemplate.policy    = this->sharing_policy_;
 169   linkTemplate.id        = id;
 170   sg_platf_new_link(&linkTemplate);
 171   XBT_DEBUG("Generating link %s", id.c_str());
 172   resource::LinkImpl* link;
 173   if (this->sharing_policy_ == s4u::Link::SharingPolicy::SPLITDUPLEX) {
 174     *linkup   = s4u::Link::by_name(linkTemplate.id + "_UP")->get_impl();   // check link?
 175     *linkdown = s4u::Link::by_name(linkTemplate.id + "_DOWN")->get_impl(); // check link ?
 176   } else {
 177     link      = s4u::Link::by_name(linkTemplate.id)->get_impl();
 178     *linkup   = link;
 179     *linkdown = link;
 180   }
 181 }
 182
 183 void DragonflyZone::generate_links()
 184 {
 185   static int uniqueId = 0;
 186   resource::LinkImpl* linkup;
 187   resource::LinkImpl* linkdown;
 188
 189   unsigned int numRouters = this->num_groups_ * this->num_chassis_per_group_ * this->num_blades_per_chassis_;
 190
 191   // Links from routers to their local nodes.
 192   for (unsigned int i = 0; i < numRouters; i++) {
 193     // allocate structures
 194     this->routers_[i]->my_nodes_    = new resource::LinkImpl*[num_links_per_link_ * this->num_nodes_per_blade_];
 195     this->routers_[i]->green_links_ = new resource::LinkImpl*[this->num_blades_per_chassis_];
 196     this->routers_[i]->black_links_ = new resource::LinkImpl*[this->num_chassis_per_group_];
 197
 198     for (unsigned int j = 0; j < num_links_per_link_ * this->num_nodes_per_blade_; j += num_links_per_link_) {
 199       std::string id = "local_link_from_router_" + std::to_string(i) + "_to_node_" +
 200                        std::to_string(j / num_links_per_link_) + "_" + std::to_string(uniqueId);
 201       this->create_link(id, 1, &linkup, &linkdown);
 202
 203       this->routers_[i]->my_nodes_[j] = linkup;
 204       if (this->sharing_policy_ == s4u::Link::SharingPolicy::SPLITDUPLEX)
 205         this->routers_[i]->my_nodes_[j + 1] = linkdown;
 206
 207       uniqueId++;
 208     }
 209   }
 210
 211   // Green links from routers to same chassis routers - alltoall
 212   for (unsigned int i = 0; i < this->num_groups_ * this->num_chassis_per_group_; i++) {
 213     for (unsigned int j = 0; j < this->num_blades_per_chassis_; j++) {
 214       for (unsigned int k = j + 1; k < this->num_blades_per_chassis_; k++) {
 215         std::string id = "green_link_in_chassis_" + std::to_string(i % num_chassis_per_group_) + "_between_routers_" +
 216                          std::to_string(j) + "_and_" + std::to_string(k) + "_" + std::to_string(uniqueId);
 217         this->create_link(id, this->num_links_green_, &linkup, &linkdown);
 218
 219         this->routers_[i * num_blades_per_chassis_ + j]->green_links_[k] = linkup;
 220         this->routers_[i * num_blades_per_chassis_ + k]->green_links_[j] = linkdown;
 221         uniqueId++;
 222       }
 223     }
 224   }
 225
 226   // Black links from routers to same group routers - alltoall
 227   for (unsigned int i = 0; i < this->num_groups_; i++) {
 228     for (unsigned int j = 0; j < this->num_chassis_per_group_; j++) {
 229       for (unsigned int k = j + 1; k < this->num_chassis_per_group_; k++) {
 230         for (unsigned int l = 0; l < this->num_blades_per_chassis_; l++) {
 231           std::string id = "black_link_in_group_" + std::to_string(i) + "_between_chassis_" + std::to_string(j) +
 232               "_and_" + std::to_string(k) +"_blade_" + std::to_string(l) + "_" + std::to_string(uniqueId);
 233           this->create_link(id, this->num_links_black_, &linkup, &linkdown);
 234
 235           this->routers_[i * num_blades_per_chassis_ * num_chassis_per_group_ + j * num_blades_per_chassis_ + l]
 236               ->black_links_[k] = linkup;
 237           this->routers_[i * num_blades_per_chassis_ * num_chassis_per_group_ + k * num_blades_per_chassis_ + l]
 238               ->black_links_[j] = linkdown;
 239           uniqueId++;
 240         }
 241       }
 242     }
 243   }
 244
 245   // Blue links between groups - Not all routers involved, only one per group is linked to others. Let's say router n of
 246   // each group is linked to group n.
 247   // FIXME: in reality blue links may be attached to several different routers
 248   for (unsigned int i = 0; i < this->num_groups_; i++) {
 249     for (unsigned int j = i + 1; j < this->num_groups_; j++) {
 250       unsigned int routernumi                 = i * num_blades_per_chassis_ * num_chassis_per_group_ + j;
 251       unsigned int routernumj                 = j * num_blades_per_chassis_ * num_chassis_per_group_ + i;
 252       this->routers_[routernumi]->blue_links_ = new resource::LinkImpl*;
 253       this->routers_[routernumj]->blue_links_ = new resource::LinkImpl*;
 254       std::string id = "blue_link_between_group_"+ std::to_string(i) +"_and_" + std::to_string(j) +"_routers_" +
 255           std::to_string(routernumi) + "_and_" + std::to_string(routernumj) + "_" + std::to_string(uniqueId);
 256       this->create_link(id, this->num_links_blue_, &linkup, &linkdown);
 257
 258       this->routers_[routernumi]->blue_links_[0] = linkup;
 259       this->routers_[routernumj]->blue_links_[0] = linkdown;
 260       uniqueId++;
 261     }
 262   }
 263 }
 264
 265 void DragonflyZone::get_local_route(NetPoint* src, NetPoint* dst, RouteCreationArgs* route, double* latency)
 266 {
 267   // Minimal routing version.
 268   // TODO : non-minimal random one, and adaptive ?
 269
 270   if (dst->is_router() || src->is_router())
 271     return;
 272
 273   XBT_VERB("dragonfly getLocalRoute from '%s'[%u] to '%s'[%u]", src->get_cname(), src->id(), dst->get_cname(),
 274            dst->id());
 275
 276   if ((src->id() == dst->id()) && has_loopback_) {
 277     std::pair<resource::LinkImpl*, resource::LinkImpl*> info = private_links_.at(node_pos(src->id()));
 278
 279     route->link_list.push_back(info.first);
 280     if (latency)
 281       *latency += info.first->get_latency();
 282     return;
 283   }
 284
 285   unsigned int myCoords[4];
 286   rankId_to_coords(src->id(), myCoords);
 287   unsigned int targetCoords[4];
 288   rankId_to_coords(dst->id(), targetCoords);
 289   XBT_DEBUG("src : %u group, %u chassis, %u blade, %u node", myCoords[0], myCoords[1], myCoords[2], myCoords[3]);
 290   XBT_DEBUG("dst : %u group, %u chassis, %u blade, %u node", targetCoords[0], targetCoords[1], targetCoords[2],
 291             targetCoords[3]);
 292
 293   DragonflyRouter* myRouter      = routers_[myCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) +
 294                                        myCoords[1] * num_blades_per_chassis_ + myCoords[2]];
 295   DragonflyRouter* targetRouter  = routers_[targetCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) +
 296                                            targetCoords[1] * num_blades_per_chassis_ + targetCoords[2]];
 297   DragonflyRouter* currentRouter = myRouter;
 298
 299   // node->router local link
 300   route->link_list.push_back(myRouter->my_nodes_[myCoords[3] * num_links_per_link_]);
 301   if (latency)
 302     *latency += myRouter->my_nodes_[myCoords[3] * num_links_per_link_]->get_latency();
 303
 304   if (has_limiter_) { // limiter for sender
 305     std::pair<resource::LinkImpl*, resource::LinkImpl*> info = private_links_.at(node_pos_with_loopback(src->id()));
 306     route->link_list.push_back(info.first);
 307   }
 308
 309   if (targetRouter != myRouter) {
 310
 311     // are we on a different group ?
 312     if (targetRouter->group_ != currentRouter->group_) {
 313       // go to the router of our group connected to this one.
 314       if (currentRouter->blade_ != targetCoords[0]) {
 315         // go to the nth router in our chassis
 316         route->link_list.push_back(currentRouter->green_links_[targetCoords[0]]);
 317         if (latency)
 318           *latency += currentRouter->green_links_[targetCoords[0]]->get_latency();
 319         currentRouter = routers_[myCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) +
 320                                  myCoords[1] * num_blades_per_chassis_ + targetCoords[0]];
 321       }
 322
 323       if (currentRouter->chassis_ != 0) {
 324         // go to the first chassis of our group
 325         route->link_list.push_back(currentRouter->black_links_[0]);
 326         if (latency)
 327           *latency += currentRouter->black_links_[0]->get_latency();
 328         currentRouter = routers_[myCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) + targetCoords[0]];
 329       }
 330
 331       // go to destination group - the only optical hop
 332       route->link_list.push_back(currentRouter->blue_links_[0]);
 333       if (latency)
 334         *latency += currentRouter->blue_links_[0]->get_latency();
 335       currentRouter = routers_[targetCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) + myCoords[0]];
 336     }
 337
 338     // same group, but same blade ?
 339     if (targetRouter->blade_ != currentRouter->blade_) {
 340       route->link_list.push_back(currentRouter->green_links_[targetCoords[2]]);
 341       if (latency)
 342         *latency += currentRouter->green_links_[targetCoords[2]]->get_latency();
 343       currentRouter = routers_[targetCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) + targetCoords[2]];
 344     }
 345
 346     // same blade, but same chassis ?
 347     if (targetRouter->chassis_ != currentRouter->chassis_) {
 348       route->link_list.push_back(currentRouter->black_links_[targetCoords[1]]);
 349       if (latency)
 350         *latency += currentRouter->black_links_[targetCoords[1]]->get_latency();
 351     }
 352   }
 353
 354   if (has_limiter_) { // limiter for receiver
 355     std::pair<resource::LinkImpl*, resource::LinkImpl*> info = private_links_.at(node_pos_with_loopback(dst->id()));
 356     route->link_list.push_back(info.first);
 357   }
 358
 359   // router->node local link
 360   route->link_list.push_back(targetRouter->my_nodes_[targetCoords[3] * num_links_per_link_ + num_links_per_link_ - 1]);
 361   if (latency)
 362     *latency += targetRouter->my_nodes_[targetCoords[3] * num_links_per_link_ + num_links_per_link_ - 1]->get_latency();
 363 }
 364 }
 365 }
 366 } // namespace