src/kernel/routing/DragonflyZone.cpp

   1 /* Copyright (c) 2014-2021. 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(const std::string& name) : ClusterZone(name) {}
  22
  23 DragonflyZone::Coords DragonflyZone::rankId_to_coords(int rankId) const
  24 {
  25   // coords : group, chassis, blade, node
  26   Coords coords;
  27   coords.group   = rankId / (num_chassis_per_group_ * num_blades_per_chassis_ * num_nodes_per_blade_);
  28   rankId         = rankId % (num_chassis_per_group_ * num_blades_per_chassis_ * num_nodes_per_blade_);
  29   coords.chassis = rankId / (num_blades_per_chassis_ * num_nodes_per_blade_);
  30   rankId         = rankId % (num_blades_per_chassis_ * num_nodes_per_blade_);
  31   coords.blade   = rankId / num_nodes_per_blade_;
  32   coords.node    = rankId % num_nodes_per_blade_;
  33   return coords;
  34 }
  35
  36 void DragonflyZone::rankId_to_coords(int rankId, unsigned int coords[4]) const // XBT_ATTRIB_DEPRECATED_v330
  37 {
  38   const auto s_coords = rankId_to_coords(rankId);
  39   coords[0]           = s_coords.group;
  40   coords[1]           = s_coords.chassis;
  41   coords[2]           = s_coords.blade;
  42   coords[3]           = s_coords.node;
  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) {
  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 (const std::invalid_argument&) {
  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 (const std::invalid_argument&) {
  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 (const std::invalid_argument&) {
  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 (const std::invalid_argument&) {
  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 (const std::invalid_argument&) {
 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 (const std::invalid_argument&) {
 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 (const std::invalid_argument&) {
 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::do_seal()
 125 {
 126   if (this->num_nodes_per_blade_ == 0) {
 127     return;
 128   }
 129
 130   this->generate_routers();
 131   this->generate_links();
 132 }
 133
 134 void DragonflyZone::generate_routers()
 135 {
 136   this->routers_.reserve(this->num_groups_ * this->num_chassis_per_group_ * this->num_blades_per_chassis_);
 137   for (unsigned int i = 0; i < this->num_groups_; i++)
 138     for (unsigned int j = 0; j < this->num_chassis_per_group_; j++)
 139       for (unsigned int k = 0; k < this->num_blades_per_chassis_; k++)
 140         this->routers_.emplace_back(i, j, k);
 141 }
 142
 143 void DragonflyZone::generate_link(const std::string& id, int numlinks, resource::LinkImpl** linkup,
 144                                   resource::LinkImpl** linkdown) const
 145 {
 146   *linkup   = nullptr;
 147   *linkdown = nullptr;
 148   LinkCreationArgs linkTemplate;
 149   linkTemplate.bandwidths.push_back(this->bw_ * numlinks);
 150   linkTemplate.latency = this->lat_;
 151   linkTemplate.policy  = this->sharing_policy_;
 152   linkTemplate.id      = id;
 153   sg_platf_new_link(&linkTemplate);
 154   XBT_DEBUG("Generating link %s", linkTemplate.id.c_str());
 155   resource::LinkImpl* link;
 156   if (this->sharing_policy_ == s4u::Link::SharingPolicy::SPLITDUPLEX) {
 157     *linkup   = s4u::Link::by_name(linkTemplate.id + "_UP")->get_impl();   // check link?
 158     *linkdown = s4u::Link::by_name(linkTemplate.id + "_DOWN")->get_impl(); // check link ?
 159   } else {
 160     link      = s4u::Link::by_name(linkTemplate.id)->get_impl();
 161     *linkup   = link;
 162     *linkdown = link;
 163   }
 164 }
 165
 166 void DragonflyZone::generate_links()
 167 {
 168   static int uniqueId = 0;
 169   resource::LinkImpl* linkup;
 170   resource::LinkImpl* linkdown;
 171
 172   unsigned int numRouters = this->num_groups_ * this->num_chassis_per_group_ * this->num_blades_per_chassis_;
 173
 174   // Links from routers to their local nodes.
 175   for (unsigned int i = 0; i < numRouters; i++) {
 176     // allocate structures
 177     this->routers_[i].my_nodes_.resize(num_links_per_link_ * this->num_nodes_per_blade_);
 178     this->routers_[i].green_links_.resize(this->num_blades_per_chassis_);
 179     this->routers_[i].black_links_.resize(this->num_chassis_per_group_);
 180
 181     for (unsigned int j = 0; j < num_links_per_link_ * this->num_nodes_per_blade_; j += num_links_per_link_) {
 182       std::string id = "local_link_from_router_" + std::to_string(i) + "_to_node_" +
 183                        std::to_string(j / num_links_per_link_) + "_" + std::to_string(uniqueId);
 184       this->generate_link(id, 1, &linkup, &linkdown);
 185
 186       this->routers_[i].my_nodes_[j] = linkup;
 187       if (this->sharing_policy_ == s4u::Link::SharingPolicy::SPLITDUPLEX)
 188         this->routers_[i].my_nodes_[j + 1] = linkdown;
 189
 190       uniqueId++;
 191     }
 192   }
 193
 194   // Green links from routers to same chassis routers - alltoall
 195   for (unsigned int i = 0; i < this->num_groups_ * this->num_chassis_per_group_; i++) {
 196     for (unsigned int j = 0; j < this->num_blades_per_chassis_; j++) {
 197       for (unsigned int k = j + 1; k < this->num_blades_per_chassis_; k++) {
 198         std::string id = "green_link_in_chassis_" + std::to_string(i % num_chassis_per_group_) + "_between_routers_" +
 199                          std::to_string(j) + "_and_" + std::to_string(k) + "_" + std::to_string(uniqueId);
 200         this->generate_link(id, this->num_links_green_, &linkup, &linkdown);
 201
 202         this->routers_[i * num_blades_per_chassis_ + j].green_links_[k] = linkup;
 203         this->routers_[i * num_blades_per_chassis_ + k].green_links_[j] = linkdown;
 204         uniqueId++;
 205       }
 206     }
 207   }
 208
 209   // Black links from routers to same group routers - alltoall
 210   for (unsigned int i = 0; i < this->num_groups_; i++) {
 211     for (unsigned int j = 0; j < this->num_chassis_per_group_; j++) {
 212       for (unsigned int k = j + 1; k < this->num_chassis_per_group_; k++) {
 213         for (unsigned int l = 0; l < this->num_blades_per_chassis_; l++) {
 214           std::string id = "black_link_in_group_" + std::to_string(i) + "_between_chassis_" + std::to_string(j) +
 215                            "_and_" + std::to_string(k) + "_blade_" + std::to_string(l) + "_" + std::to_string(uniqueId);
 216           this->generate_link(id, this->num_links_black_, &linkup, &linkdown);
 217
 218           this->routers_[i * num_blades_per_chassis_ * num_chassis_per_group_ + j * num_blades_per_chassis_ + l]
 219               .black_links_[k] = linkup;
 220           this->routers_[i * num_blades_per_chassis_ * num_chassis_per_group_ + k * num_blades_per_chassis_ + l]
 221               .black_links_[j] = linkdown;
 222           uniqueId++;
 223         }
 224       }
 225     }
 226   }
 227
 228   // Blue links between groups - Not all routers involved, only one per group is linked to others. Let's say router n of
 229   // each group is linked to group n.
 230   // FIXME: in reality blue links may be attached to several different routers
 231   for (unsigned int i = 0; i < this->num_groups_; i++) {
 232     for (unsigned int j = i + 1; j < this->num_groups_; j++) {
 233       unsigned int routernumi = i * num_blades_per_chassis_ * num_chassis_per_group_ + j;
 234       unsigned int routernumj = j * num_blades_per_chassis_ * num_chassis_per_group_ + i;
 235       std::string id = "blue_link_between_group_" + std::to_string(i) + "_and_" + std::to_string(j) + "_routers_" +
 236                        std::to_string(routernumi) + "_and_" + std::to_string(routernumj) + "_" +
 237                        std::to_string(uniqueId);
 238       this->generate_link(id, this->num_links_blue_, &linkup, &linkdown);
 239
 240       this->routers_[routernumi].blue_link_ = linkup;
 241       this->routers_[routernumj].blue_link_ = linkdown;
 242       uniqueId++;
 243     }
 244   }
 245 }
 246
 247 void DragonflyZone::get_local_route(NetPoint* src, NetPoint* dst, RouteCreationArgs* route, double* latency)
 248 {
 249   // Minimal routing version.
 250   // TODO : non-minimal random one, and adaptive ?
 251
 252   if (dst->is_router() || src->is_router())
 253     return;
 254
 255   XBT_VERB("dragonfly getLocalRoute from '%s'[%u] to '%s'[%u]", src->get_cname(), src->id(), dst->get_cname(),
 256            dst->id());
 257
 258   if ((src->id() == dst->id()) && has_loopback()) {
 259     resource::LinkImpl* uplink = get_uplink_from(node_pos(src->id()));
 260
 261     route->link_list.push_back(uplink);
 262     if (latency)
 263       *latency += uplink->get_latency();
 264     return;
 265   }
 266
 267   const auto myCoords     = rankId_to_coords(src->id());
 268   const auto targetCoords = rankId_to_coords(dst->id());
 269   XBT_DEBUG("src : %u group, %u chassis, %u blade, %u node", myCoords.group, myCoords.chassis, myCoords.blade,
 270             myCoords.node);
 271   XBT_DEBUG("dst : %u group, %u chassis, %u blade, %u node", targetCoords.group, targetCoords.chassis,
 272             targetCoords.blade, targetCoords.node);
 273
 274   DragonflyRouter* myRouter      = &routers_[myCoords.group * (num_chassis_per_group_ * num_blades_per_chassis_) +
 275                                         myCoords.chassis * num_blades_per_chassis_ + myCoords.blade];
 276   DragonflyRouter* targetRouter  = &routers_[targetCoords.group * (num_chassis_per_group_ * num_blades_per_chassis_) +
 277                                             targetCoords.chassis * num_blades_per_chassis_ + targetCoords.blade];
 278   DragonflyRouter* currentRouter = myRouter;
 279
 280   // node->router local link
 281   route->link_list.push_back(myRouter->my_nodes_[myCoords.node * num_links_per_link_]);
 282   if (latency)
 283     *latency += myRouter->my_nodes_[myCoords.node * num_links_per_link_]->get_latency();
 284
 285   if (has_limiter()) { // limiter for sender
 286     route->link_list.push_back(get_uplink_from(node_pos_with_loopback(src->id())));
 287   }
 288
 289   if (targetRouter != myRouter) {
 290     // are we on a different group ?
 291     if (targetRouter->group_ != currentRouter->group_) {
 292       // go to the router of our group connected to this one.
 293       if (currentRouter->blade_ != targetCoords.group) {
 294         // go to the nth router in our chassis
 295         route->link_list.push_back(currentRouter->green_links_[targetCoords.group]);
 296         if (latency)
 297           *latency += currentRouter->green_links_[targetCoords.group]->get_latency();
 298         currentRouter = &routers_[myCoords.group * (num_chassis_per_group_ * num_blades_per_chassis_) +
 299                                   myCoords.chassis * num_blades_per_chassis_ + targetCoords.group];
 300       }
 301
 302       if (currentRouter->chassis_ != 0) {
 303         // go to the first chassis of our group
 304         route->link_list.push_back(currentRouter->black_links_[0]);
 305         if (latency)
 306           *latency += currentRouter->black_links_[0]->get_latency();
 307         currentRouter =
 308             &routers_[myCoords.group * (num_chassis_per_group_ * num_blades_per_chassis_) + targetCoords.group];
 309       }
 310
 311       // go to destination group - the only optical hop
 312       route->link_list.push_back(currentRouter->blue_link_);
 313       if (latency)
 314         *latency += currentRouter->blue_link_->get_latency();
 315       currentRouter =
 316           &routers_[targetCoords.group * (num_chassis_per_group_ * num_blades_per_chassis_) + myCoords.group];
 317     }
 318
 319     // same group, but same blade ?
 320     if (targetRouter->blade_ != currentRouter->blade_) {
 321       route->link_list.push_back(currentRouter->green_links_[targetCoords.blade]);
 322       if (latency)
 323         *latency += currentRouter->green_links_[targetCoords.blade]->get_latency();
 324       currentRouter =
 325           &routers_[targetCoords.group * (num_chassis_per_group_ * num_blades_per_chassis_) + targetCoords.blade];
 326     }
 327
 328     // same blade, but same chassis ?
 329     if (targetRouter->chassis_ != currentRouter->chassis_) {
 330       route->link_list.push_back(currentRouter->black_links_[targetCoords.chassis]);
 331       if (latency)
 332         *latency += currentRouter->black_links_[targetCoords.chassis]->get_latency();
 333     }
 334   }
 335
 336   if (has_limiter()) { // limiter for receiver
 337     route->link_list.push_back(get_downlink_to(node_pos_with_loopback(dst->id())));
 338   }
 339
 340   // router->node local link
 341   route->link_list.push_back(
 342       targetRouter->my_nodes_[targetCoords.node * num_links_per_link_ + num_links_per_link_ - 1]);
 343   if (latency)
 344     *latency +=
 345         targetRouter->my_nodes_[targetCoords.node * num_links_per_link_ + num_links_per_link_ - 1]->get_latency();
 346 }
 347 } // namespace routing
 348 } // namespace kernel
 349
 350 namespace s4u {
 351 NetZone* create_dragonfly_zone(const std::string& name)
 352 {
 353   return (new kernel::routing::DragonflyZone(name))->get_iface();
 354 }
 355 } // namespace s4u
 356
 357 } // namespace simgrid