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