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   if (cluster->sharing_policy == SURF_LINK_SPLITDUPLEX)
 117     this->num_links_per_link_ = 2;
 118
 119   this->cluster_ = cluster;
 120 }
 121
 122 /* Generate the cluster once every node is created */
 123 void DragonflyZone::seal()
 124 {
 125   if (this->num_nodes_per_blade_ == 0) {
 126     return;
 127   }
 128
 129   this->generateRouters();
 130   this->generateLinks();
 131 }
 132
 133 DragonflyRouter::DragonflyRouter(int group, int chassis, int blade) : group_(group), chassis_(chassis), blade_(blade)
 134 {
 135 }
 136
 137 DragonflyRouter::~DragonflyRouter()
 138 {
 139   delete[] my_nodes_;
 140   delete[] green_links_;
 141   delete[] black_links_;
 142   delete blue_links_;
 143 }
 144
 145 void DragonflyZone::generateRouters()
 146 {
 147   this->routers_ =
 148       new DragonflyRouter*[this->num_groups_ * this->num_chassis_per_group_ * this->num_blades_per_chassis_];
 149
 150   for (unsigned int i = 0; i < this->num_groups_; i++) {
 151     for (unsigned int j = 0; j < this->num_chassis_per_group_; j++) {
 152       for (unsigned int k = 0; k < this->num_blades_per_chassis_; k++) {
 153         DragonflyRouter* router = new DragonflyRouter(i, j, k);
 154         this->routers_[i * this->num_chassis_per_group_ * this->num_blades_per_chassis_ +
 155                        j * this->num_blades_per_chassis_ + k] = router;
 156       }
 157     }
 158   }
 159 }
 160
 161 void DragonflyZone::createLink(const std::string& id, int numlinks, surf::LinkImpl** linkup, surf::LinkImpl** linkdown)
 162 {
 163   *linkup   = nullptr;
 164   *linkdown = nullptr;
 165   LinkCreationArgs linkTemplate;
 166   linkTemplate.bandwidth = this->cluster_->bw * numlinks;
 167   linkTemplate.latency   = this->cluster_->lat;
 168   linkTemplate.policy    = this->cluster_->sharing_policy; // sthg to do with that ?
 169   linkTemplate.id        = id;
 170   sg_platf_new_link(&linkTemplate);
 171   XBT_DEBUG("Generating link %s", id.c_str());
 172   surf::LinkImpl* link;
 173   if (this->cluster_->sharing_policy == SURF_LINK_SPLITDUPLEX) {
 174     *linkup   = surf::LinkImpl::byName(linkTemplate.id + "_UP");   // check link?
 175     *linkdown = surf::LinkImpl::byName(linkTemplate.id + "_DOWN"); // check link ?
 176   } else {
 177     link      = surf::LinkImpl::byName(linkTemplate.id);
 178     *linkup   = link;
 179     *linkdown = link;
 180   }
 181 }
 182
 183 void DragonflyZone::generateLinks()
 184 {
 185   static int uniqueId = 0;
 186   surf::LinkImpl* linkup;
 187   surf::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 surf::LinkImpl*[num_links_per_link_ * this->num_nodes_per_blade_];
 195     this->routers_[i]->green_links_ = new surf::LinkImpl*[this->num_blades_per_chassis_];
 196     this->routers_[i]->black_links_ = new surf::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->createLink(id, 1, &linkup, &linkdown);
 202
 203       this->routers_[i]->my_nodes_[j] = linkup;
 204       if (this->cluster_->sharing_policy == SURF_LINK_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->createLink(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->createLink(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 surf::LinkImpl*;
 253       this->routers_[routernumj]->blue_links_ = new surf::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->createLink(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::getLocalRoute(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->isRouter() || src->isRouter())
 271     return;
 272
 273   XBT_VERB("dragonfly getLocalRoute from '%s'[%u] to '%s'[%u]", src->getCname(), src->id(), dst->getCname(), dst->id());
 274
 275   if ((src->id() == dst->id()) && has_loopback_) {
 276     std::pair<surf::LinkImpl*, surf::LinkImpl*> info = private_links_.at(nodePosition(src->id()));
 277
 278     route->link_list.push_back(info.first);
 279     if (latency)
 280       *latency += info.first->latency();
 281     return;
 282   }
 283
 284   unsigned int myCoords[4];
 285   rankId_to_coords(src->id(), &myCoords);
 286   unsigned int targetCoords[4];
 287   rankId_to_coords(dst->id(), &targetCoords);
 288   XBT_DEBUG("src : %u group, %u chassis, %u blade, %u node", myCoords[0], myCoords[1], myCoords[2], myCoords[3]);
 289   XBT_DEBUG("dst : %u group, %u chassis, %u blade, %u node", targetCoords[0], targetCoords[1], targetCoords[2],
 290             targetCoords[3]);
 291
 292   DragonflyRouter* myRouter      = routers_[myCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) +
 293                                        myCoords[1] * num_blades_per_chassis_ + myCoords[2]];
 294   DragonflyRouter* targetRouter  = routers_[targetCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) +
 295                                            targetCoords[1] * num_blades_per_chassis_ + targetCoords[2]];
 296   DragonflyRouter* currentRouter = myRouter;
 297
 298   // node->router local link
 299   route->link_list.push_back(myRouter->my_nodes_[myCoords[3] * num_links_per_link_]);
 300   if (latency)
 301     *latency += myRouter->my_nodes_[myCoords[3] * num_links_per_link_]->latency();
 302
 303   if (has_limiter_) { // limiter for sender
 304     std::pair<surf::LinkImpl*, surf::LinkImpl*> info = private_links_.at(nodePositionWithLoopback(src->id()));
 305     route->link_list.push_back(info.first);
 306   }
 307
 308   if (targetRouter != myRouter) {
 309
 310     // are we on a different group ?
 311     if (targetRouter->group_ != currentRouter->group_) {
 312       // go to the router of our group connected to this one.
 313       if (currentRouter->blade_ != targetCoords[0]) {
 314         // go to the nth router in our chassis
 315         route->link_list.push_back(currentRouter->green_links_[targetCoords[0]]);
 316         if (latency)
 317           *latency += currentRouter->green_links_[targetCoords[0]]->latency();
 318         currentRouter = routers_[myCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) +
 319                                  myCoords[1] * num_blades_per_chassis_ + targetCoords[0]];
 320       }
 321
 322       if (currentRouter->chassis_ != 0) {
 323         // go to the first chassis of our group
 324         route->link_list.push_back(currentRouter->black_links_[0]);
 325         if (latency)
 326           *latency += currentRouter->black_links_[0]->latency();
 327         currentRouter = routers_[myCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) + targetCoords[0]];
 328       }
 329
 330       // go to destination group - the only optical hop
 331       route->link_list.push_back(currentRouter->blue_links_[0]);
 332       if (latency)
 333         *latency += currentRouter->blue_links_[0]->latency();
 334       currentRouter = routers_[targetCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) + myCoords[0]];
 335     }
 336
 337     // same group, but same blade ?
 338     if (targetRouter->blade_ != currentRouter->blade_) {
 339       route->link_list.push_back(currentRouter->green_links_[targetCoords[2]]);
 340       if (latency)
 341         *latency += currentRouter->green_links_[targetCoords[2]]->latency();
 342       currentRouter = routers_[targetCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) + targetCoords[2]];
 343     }
 344
 345     // same blade, but same chassis ?
 346     if (targetRouter->chassis_ != currentRouter->chassis_) {
 347       route->link_list.push_back(currentRouter->black_links_[targetCoords[1]]);
 348       if (latency)
 349         *latency += currentRouter->black_links_[targetCoords[1]]->latency();
 350     }
 351   }
 352
 353   if (has_limiter_) { // limiter for receiver
 354     std::pair<surf::LinkImpl*, surf::LinkImpl*> info = private_links_.at(nodePositionWithLoopback(dst->id()));
 355     route->link_list.push_back(info.first);
 356   }
 357
 358   // router->node local link
 359   route->link_list.push_back(targetRouter->my_nodes_[targetCoords[3] * num_links_per_link_ + num_links_per_link_ - 1]);
 360   if (latency)
 361     *latency += targetRouter->my_nodes_[targetCoords[3] * num_links_per_link_ + num_links_per_link_ - 1]->latency();
 362 }
 363 }
 364 }
 365 } // namespace