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 == SURF_LINK_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, surf::LinkImpl** linkup, surf::LinkImpl** linkdown)
 163 {
 164   *linkup   = nullptr;
 165   *linkdown = nullptr;
 166   LinkCreationArgs linkTemplate;
 167   linkTemplate.bandwidth = this->bw_ * numlinks;
 168   linkTemplate.latency   = this->lat_;
 169   linkTemplate.policy    = this->sharing_policy_;
 170   linkTemplate.id        = id;
 171   sg_platf_new_link(&linkTemplate);
 172   XBT_DEBUG("Generating link %s", id.c_str());
 173   surf::LinkImpl* link;
 174   if (this->sharing_policy_ == SURF_LINK_SPLITDUPLEX) {
 175     *linkup   = surf::LinkImpl::byName(linkTemplate.id + "_UP");   // check link?
 176     *linkdown = surf::LinkImpl::byName(linkTemplate.id + "_DOWN"); // check link ?
 177   } else {
 178     link      = surf::LinkImpl::byName(linkTemplate.id);
 179     *linkup   = link;
 180     *linkdown = link;
 181   }
 182 }
 183
 184 void DragonflyZone::generateLinks()
 185 {
 186   static int uniqueId = 0;
 187   surf::LinkImpl* linkup;
 188   surf::LinkImpl* linkdown;
 189
 190   unsigned int numRouters = this->num_groups_ * this->num_chassis_per_group_ * this->num_blades_per_chassis_;
 191
 192   // Links from routers to their local nodes.
 193   for (unsigned int i = 0; i < numRouters; i++) {
 194     // allocate structures
 195     this->routers_[i]->my_nodes_    = new surf::LinkImpl*[num_links_per_link_ * this->num_nodes_per_blade_];
 196     this->routers_[i]->green_links_ = new surf::LinkImpl*[this->num_blades_per_chassis_];
 197     this->routers_[i]->black_links_ = new surf::LinkImpl*[this->num_chassis_per_group_];
 198
 199     for (unsigned int j = 0; j < num_links_per_link_ * this->num_nodes_per_blade_; j += num_links_per_link_) {
 200       std::string id = "local_link_from_router_" + std::to_string(i) + "_to_node_" +
 201                        std::to_string(j / num_links_per_link_) + "_" + std::to_string(uniqueId);
 202       this->createLink(id, 1, &linkup, &linkdown);
 203
 204       this->routers_[i]->my_nodes_[j] = linkup;
 205       if (this->sharing_policy_ == SURF_LINK_SPLITDUPLEX)
 206         this->routers_[i]->my_nodes_[j + 1] = linkdown;
 207
 208       uniqueId++;
 209     }
 210   }
 211
 212   // Green links from routers to same chassis routers - alltoall
 213   for (unsigned int i = 0; i < this->num_groups_ * this->num_chassis_per_group_; i++) {
 214     for (unsigned int j = 0; j < this->num_blades_per_chassis_; j++) {
 215       for (unsigned int k = j + 1; k < this->num_blades_per_chassis_; k++) {
 216         std::string id = "green_link_in_chassis_" + std::to_string(i % num_chassis_per_group_) + "_between_routers_" +
 217                          std::to_string(j) + "_and_" + std::to_string(k) + "_" + std::to_string(uniqueId);
 218         this->createLink(id, this->num_links_green_, &linkup, &linkdown);
 219
 220         this->routers_[i * num_blades_per_chassis_ + j]->green_links_[k] = linkup;
 221         this->routers_[i * num_blades_per_chassis_ + k]->green_links_[j] = linkdown;
 222         uniqueId++;
 223       }
 224     }
 225   }
 226
 227   // Black links from routers to same group routers - alltoall
 228   for (unsigned int i = 0; i < this->num_groups_; i++) {
 229     for (unsigned int j = 0; j < this->num_chassis_per_group_; j++) {
 230       for (unsigned int k = j + 1; k < this->num_chassis_per_group_; k++) {
 231         for (unsigned int l = 0; l < this->num_blades_per_chassis_; l++) {
 232           std::string id = "black_link_in_group_" + std::to_string(i) + "_between_chassis_" + std::to_string(j) +
 233               "_and_" + std::to_string(k) +"_blade_" + std::to_string(l) + "_" + std::to_string(uniqueId);
 234           this->createLink(id, this->num_links_black_, &linkup, &linkdown);
 235
 236           this->routers_[i * num_blades_per_chassis_ * num_chassis_per_group_ + j * num_blades_per_chassis_ + l]
 237               ->black_links_[k] = linkup;
 238           this->routers_[i * num_blades_per_chassis_ * num_chassis_per_group_ + k * num_blades_per_chassis_ + l]
 239               ->black_links_[j] = linkdown;
 240           uniqueId++;
 241         }
 242       }
 243     }
 244   }
 245
 246   // Blue links between groups - Not all routers involved, only one per group is linked to others. Let's say router n of
 247   // each group is linked to group n.
 248   // FIXME: in reality blue links may be attached to several different routers
 249   for (unsigned int i = 0; i < this->num_groups_; i++) {
 250     for (unsigned int j = i + 1; j < this->num_groups_; j++) {
 251       unsigned int routernumi                 = i * num_blades_per_chassis_ * num_chassis_per_group_ + j;
 252       unsigned int routernumj                 = j * num_blades_per_chassis_ * num_chassis_per_group_ + i;
 253       this->routers_[routernumi]->blue_links_ = new surf::LinkImpl*;
 254       this->routers_[routernumj]->blue_links_ = new surf::LinkImpl*;
 255       std::string id = "blue_link_between_group_"+ std::to_string(i) +"_and_" + std::to_string(j) +"_routers_" +
 256           std::to_string(routernumi) + "_and_" + std::to_string(routernumj) + "_" + std::to_string(uniqueId);
 257       this->createLink(id, this->num_links_blue_, &linkup, &linkdown);
 258
 259       this->routers_[routernumi]->blue_links_[0] = linkup;
 260       this->routers_[routernumj]->blue_links_[0] = linkdown;
 261       uniqueId++;
 262     }
 263   }
 264 }
 265
 266 void DragonflyZone::get_local_route(NetPoint* src, NetPoint* dst, RouteCreationArgs* route, double* latency)
 267 {
 268   // Minimal routing version.
 269   // TODO : non-minimal random one, and adaptive ?
 270
 271   if (dst->is_router() || src->is_router())
 272     return;
 273
 274   XBT_VERB("dragonfly getLocalRoute from '%s'[%u] to '%s'[%u]", src->get_cname(), src->id(), dst->get_cname(),
 275            dst->id());
 276
 277   if ((src->id() == dst->id()) && has_loopback_) {
 278     std::pair<surf::LinkImpl*, surf::LinkImpl*> info = private_links_.at(node_pos(src->id()));
 279
 280     route->link_list.push_back(info.first);
 281     if (latency)
 282       *latency += info.first->latency();
 283     return;
 284   }
 285
 286   unsigned int myCoords[4];
 287   rankId_to_coords(src->id(), &myCoords);
 288   unsigned int targetCoords[4];
 289   rankId_to_coords(dst->id(), &targetCoords);
 290   XBT_DEBUG("src : %u group, %u chassis, %u blade, %u node", myCoords[0], myCoords[1], myCoords[2], myCoords[3]);
 291   XBT_DEBUG("dst : %u group, %u chassis, %u blade, %u node", targetCoords[0], targetCoords[1], targetCoords[2],
 292             targetCoords[3]);
 293
 294   DragonflyRouter* myRouter      = routers_[myCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) +
 295                                        myCoords[1] * num_blades_per_chassis_ + myCoords[2]];
 296   DragonflyRouter* targetRouter  = routers_[targetCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) +
 297                                            targetCoords[1] * num_blades_per_chassis_ + targetCoords[2]];
 298   DragonflyRouter* currentRouter = myRouter;
 299
 300   // node->router local link
 301   route->link_list.push_back(myRouter->my_nodes_[myCoords[3] * num_links_per_link_]);
 302   if (latency)
 303     *latency += myRouter->my_nodes_[myCoords[3] * num_links_per_link_]->latency();
 304
 305   if (has_limiter_) { // limiter for sender
 306     std::pair<surf::LinkImpl*, surf::LinkImpl*> info = private_links_.at(node_pos_with_loopback(src->id()));
 307     route->link_list.push_back(info.first);
 308   }
 309
 310   if (targetRouter != myRouter) {
 311
 312     // are we on a different group ?
 313     if (targetRouter->group_ != currentRouter->group_) {
 314       // go to the router of our group connected to this one.
 315       if (currentRouter->blade_ != targetCoords[0]) {
 316         // go to the nth router in our chassis
 317         route->link_list.push_back(currentRouter->green_links_[targetCoords[0]]);
 318         if (latency)
 319           *latency += currentRouter->green_links_[targetCoords[0]]->latency();
 320         currentRouter = routers_[myCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) +
 321                                  myCoords[1] * num_blades_per_chassis_ + targetCoords[0]];
 322       }
 323
 324       if (currentRouter->chassis_ != 0) {
 325         // go to the first chassis of our group
 326         route->link_list.push_back(currentRouter->black_links_[0]);
 327         if (latency)
 328           *latency += currentRouter->black_links_[0]->latency();
 329         currentRouter = routers_[myCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) + targetCoords[0]];
 330       }
 331
 332       // go to destination group - the only optical hop
 333       route->link_list.push_back(currentRouter->blue_links_[0]);
 334       if (latency)
 335         *latency += currentRouter->blue_links_[0]->latency();
 336       currentRouter = routers_[targetCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) + myCoords[0]];
 337     }
 338
 339     // same group, but same blade ?
 340     if (targetRouter->blade_ != currentRouter->blade_) {
 341       route->link_list.push_back(currentRouter->green_links_[targetCoords[2]]);
 342       if (latency)
 343         *latency += currentRouter->green_links_[targetCoords[2]]->latency();
 344       currentRouter = routers_[targetCoords[0] * (num_chassis_per_group_ * num_blades_per_chassis_) + targetCoords[2]];
 345     }
 346
 347     // same blade, but same chassis ?
 348     if (targetRouter->chassis_ != currentRouter->chassis_) {
 349       route->link_list.push_back(currentRouter->black_links_[targetCoords[1]]);
 350       if (latency)
 351         *latency += currentRouter->black_links_[targetCoords[1]]->latency();
 352     }
 353   }
 354
 355   if (has_limiter_) { // limiter for receiver
 356     std::pair<surf::LinkImpl*, surf::LinkImpl*> info = private_links_.at(node_pos_with_loopback(dst->id()));
 357     route->link_list.push_back(info.first);
 358   }
 359
 360   // router->node local link
 361   route->link_list.push_back(targetRouter->my_nodes_[targetCoords[3] * num_links_per_link_ + num_links_per_link_ - 1]);
 362   if (latency)
 363     *latency += targetRouter->my_nodes_[targetCoords[3] * num_links_per_link_ + num_links_per_link_ - 1]->latency();
 364 }
 365 }
 366 }
 367 } // namespace