src/kernel/routing/DragonflyZone.cpp

   1 /* Copyright (c) 2014-2020. 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, const std::string& name, resource::NetworkModel* netmodel)
  22     : ClusterZone(father, name, netmodel)
  23 {
  24 }
  25
  26 DragonflyZone::Coords DragonflyZone::rankId_to_coords(int rankId) const
  27 {
  28   // coords : group, chassis, blade, node
  29   Coords coords;
  30   coords.group   = rankId / (num_chassis_per_group_ * num_blades_per_chassis_ * num_nodes_per_blade_);
  31   rankId         = rankId % (num_chassis_per_group_ * num_blades_per_chassis_ * num_nodes_per_blade_);
  32   coords.chassis = rankId / (num_blades_per_chassis_ * num_nodes_per_blade_);
  33   rankId         = rankId % (num_blades_per_chassis_ * num_nodes_per_blade_);
  34   coords.blade   = rankId / num_nodes_per_blade_;
  35   coords.node    = rankId % num_nodes_per_blade_;
  36   return coords;
  37 }
  38
  39 void DragonflyZone::rankId_to_coords(int rankId, unsigned int coords[4]) const // XBT_ATTRIB_DEPRECATED_v330
  40 {
  41   const auto s_coords = rankId_to_coords(rankId);
  42   coords[0]           = s_coords.group;
  43   coords[1]           = s_coords.chassis;
  44   coords[2]           = s_coords.blade;
  45   coords[3]           = s_coords.node;
  46 }
  47
  48 void DragonflyZone::parse_specific_arguments(ClusterCreationArgs* cluster)
  49 {
  50   std::vector<std::string> parameters;
  51   std::vector<std::string> tmp;
  52   boost::split(parameters, cluster->topo_parameters, boost::is_any_of(";"));
  53
  54   if (parameters.size() != 4 || parameters.empty()) {
  55     surf_parse_error(
  56         "Dragonfly are defined by the number of groups, chassis per groups, blades per chassis, nodes per blade");
  57   }
  58
  59   // Blue network : number of groups, number of links between each group
  60   boost::split(tmp, parameters[0], boost::is_any_of(","));
  61   if (tmp.size() != 2) {
  62     surf_parse_error("Dragonfly topologies are defined by 3 levels with 2 elements each, and one with one element");
  63   }
  64
  65   try {
  66     this->num_groups_ = std::stoi(tmp[0]);
  67   } catch (const std::invalid_argument&) {
  68     throw std::invalid_argument(std::string("Invalid number of groups:") + tmp[0]);
  69   }
  70
  71   try {
  72     this->num_links_blue_ = std::stoi(tmp[1]);
  73   } catch (const std::invalid_argument&) {
  74     throw std::invalid_argument(std::string("Invalid number of links for the blue level:") + tmp[1]);
  75   }
  76   // Black network : number of chassis/group, number of links between each router on the black network
  77   boost::split(tmp, parameters[1], boost::is_any_of(","));
  78   if (tmp.size() != 2) {
  79     surf_parse_error("Dragonfly topologies are defined by 3 levels with 2 elements each, and one with one element");
  80   }
  81
  82   try {
  83     this->num_chassis_per_group_ = std::stoi(tmp[0]);
  84   } catch (const std::invalid_argument&) {
  85     throw std::invalid_argument(std::string("Invalid number of groups:") + tmp[0]);
  86   }
  87
  88   try {
  89     this->num_links_black_ = std::stoi(tmp[1]);
  90   } catch (const std::invalid_argument&) {
  91     throw std::invalid_argument(std::string("Invalid number of links for the black level:") + tmp[1]);
  92   }
  93
  94   // Green network : number of blades/chassis, number of links between each router on the green network
  95   boost::split(tmp, parameters[2], boost::is_any_of(","));
  96   if (tmp.size() != 2) {
  97     surf_parse_error("Dragonfly topologies are defined by 3 levels with 2 elements each, and one with one element");
  98   }
  99
 100   try {
 101     this->num_blades_per_chassis_ = std::stoi(tmp[0]);
 102   } catch (const std::invalid_argument&) {
 103     throw std::invalid_argument(std::string("Invalid number of groups:") + tmp[0]);
 104   }
 105
 106   try {
 107     this->num_links_green_ = std::stoi(tmp[1]);
 108   } catch (const std::invalid_argument&) {
 109     throw std::invalid_argument(std::string("Invalid number of links for the green level:") + tmp[1]);
 110   }
 111
 112   // The last part of topo_parameters should be the number of nodes per blade
 113   try {
 114     this->num_nodes_per_blade_ = std::stoi(parameters[3]);
 115   } catch (const std::invalid_argument&) {
 116     throw std::invalid_argument(std::string("Last parameter is not the amount of nodes per blade:") + parameters[3]);
 117   }
 118
 119   this->sharing_policy_ = cluster->sharing_policy;
 120   if (cluster->sharing_policy == s4u::Link::SharingPolicy::SPLITDUPLEX)
 121     this->num_links_per_link_ = 2;
 122   this->bw_  = cluster->bw;
 123   this->lat_ = cluster->lat;
 124 }
 125
 126 /* Generate the cluster once every node is created */
 127 void DragonflyZone::seal()
 128 {
 129   if (this->num_nodes_per_blade_ == 0) {
 130     return;
 131   }
 132
 133   this->generate_routers();
 134   this->generate_links();
 135 }
 136
 137 void DragonflyZone::generate_routers()
 138 {
 139   this->routers_.reserve(this->num_groups_ * this->num_chassis_per_group_ * this->num_blades_per_chassis_);
 140   for (unsigned int i = 0; i < this->num_groups_; i++)
 141     for (unsigned int j = 0; j < this->num_chassis_per_group_; j++)
 142       for (unsigned int k = 0; k < this->num_blades_per_chassis_; k++)
 143         this->routers_.emplace_back(i, j, k);
 144 }
 145
 146 void DragonflyZone::generate_link(const std::string& id, int numlinks, resource::LinkImpl** linkup,
 147                                   resource::LinkImpl** linkdown) const
 148 {
 149   *linkup   = nullptr;
 150   *linkdown = nullptr;
 151   LinkCreationArgs linkTemplate;
 152   linkTemplate.bandwidths.push_back(this->bw_ * numlinks);
 153   linkTemplate.latency   = this->lat_;
 154   linkTemplate.policy    = this->sharing_policy_;
 155   linkTemplate.id        = id;
 156   sg_platf_new_link(&linkTemplate);
 157   XBT_DEBUG("Generating link %s", linkTemplate.id.c_str());
 158   resource::LinkImpl* link;
 159   if (this->sharing_policy_ == s4u::Link::SharingPolicy::SPLITDUPLEX) {
 160     *linkup   = s4u::Link::by_name(linkTemplate.id + "_UP")->get_impl();   // check link?
 161     *linkdown = s4u::Link::by_name(linkTemplate.id + "_DOWN")->get_impl(); // check link ?
 162   } else {
 163     link      = s4u::Link::by_name(linkTemplate.id)->get_impl();
 164     *linkup   = link;
 165     *linkdown = link;
 166   }
 167 }
 168
 169 void DragonflyZone::generate_links()
 170 {
 171   static int uniqueId = 0;
 172   resource::LinkImpl* linkup;
 173   resource::LinkImpl* linkdown;
 174
 175   unsigned int numRouters = this->num_groups_ * this->num_chassis_per_group_ * this->num_blades_per_chassis_;
 176
 177   // Links from routers to their local nodes.
 178   for (unsigned int i = 0; i < numRouters; i++) {
 179     // allocate structures
 180     this->routers_[i].my_nodes_.resize(num_links_per_link_ * this->num_nodes_per_blade_);
 181     this->routers_[i].green_links_.resize(this->num_blades_per_chassis_);
 182     this->routers_[i].black_links_.resize(this->num_chassis_per_group_);
 183
 184     for (unsigned int j = 0; j < num_links_per_link_ * this->num_nodes_per_blade_; j += num_links_per_link_) {
 185       std::string id = "local_link_from_router_" + std::to_string(i) + "_to_node_" +
 186                        std::to_string(j / num_links_per_link_) + "_" + std::to_string(uniqueId);
 187       this->generate_link(id, 1, &linkup, &linkdown);
 188
 189       this->routers_[i].my_nodes_[j] = linkup;
 190       if (this->sharing_policy_ == s4u::Link::SharingPolicy::SPLITDUPLEX)
 191         this->routers_[i].my_nodes_[j + 1] = linkdown;
 192
 193       uniqueId++;
 194     }
 195   }
 196
 197   // Green links from routers to same chassis routers - alltoall
 198   for (unsigned int i = 0; i < this->num_groups_ * this->num_chassis_per_group_; i++) {
 199     for (unsigned int j = 0; j < this->num_blades_per_chassis_; j++) {
 200       for (unsigned int k = j + 1; k < this->num_blades_per_chassis_; k++) {
 201         std::string id = "green_link_in_chassis_" + std::to_string(i % num_chassis_per_group_) + "_between_routers_" +
 202                          std::to_string(j) + "_and_" + std::to_string(k) + "_" + std::to_string(uniqueId);
 203         this->generate_link(id, this->num_links_green_, &linkup, &linkdown);
 204
 205         this->routers_[i * num_blades_per_chassis_ + j].green_links_[k] = linkup;
 206         this->routers_[i * num_blades_per_chassis_ + k].green_links_[j] = linkdown;
 207         uniqueId++;
 208       }
 209     }
 210   }
 211
 212   // Black links from routers to same group routers - alltoall
 213   for (unsigned int i = 0; i < this->num_groups_; i++) {
 214     for (unsigned int j = 0; j < this->num_chassis_per_group_; j++) {
 215       for (unsigned int k = j + 1; k < this->num_chassis_per_group_; k++) {
 216         for (unsigned int l = 0; l < this->num_blades_per_chassis_; l++) {
 217           std::string id = "black_link_in_group_" + std::to_string(i) + "_between_chassis_" + std::to_string(j) +
 218               "_and_" + std::to_string(k) +"_blade_" + std::to_string(l) + "_" + std::to_string(uniqueId);
 219           this->generate_link(id, this->num_links_black_, &linkup, &linkdown);
 220
 221           this->routers_[i * num_blades_per_chassis_ * num_chassis_per_group_ + j * num_blades_per_chassis_ + l]
 222               .black_links_[k] = linkup;
 223           this->routers_[i * num_blades_per_chassis_ * num_chassis_per_group_ + k * num_blades_per_chassis_ + l]
 224               .black_links_[j] = linkdown;
 225           uniqueId++;
 226         }
 227       }
 228     }
 229   }
 230
 231   // Blue links between groups - Not all routers involved, only one per group is linked to others. Let's say router n of
 232   // each group is linked to group n.
 233   // FIXME: in reality blue links may be attached to several different routers
 234   for (unsigned int i = 0; i < this->num_groups_; i++) {
 235     for (unsigned int j = i + 1; j < this->num_groups_; j++) {
 236       unsigned int routernumi                 = i * num_blades_per_chassis_ * num_chassis_per_group_ + j;
 237       unsigned int routernumj                 = j * num_blades_per_chassis_ * num_chassis_per_group_ + i;
 238       std::string id = "blue_link_between_group_"+ std::to_string(i) +"_and_" + std::to_string(j) +"_routers_" +
 239           std::to_string(routernumi) + "_and_" + std::to_string(routernumj) + "_" + std::to_string(uniqueId);
 240       this->generate_link(id, this->num_links_blue_, &linkup, &linkdown);
 241
 242       this->routers_[routernumi].blue_link_ = linkup;
 243       this->routers_[routernumj].blue_link_ = linkdown;
 244       uniqueId++;
 245     }
 246   }
 247 }
 248
 249 void DragonflyZone::get_local_route(NetPoint* src, NetPoint* dst, RouteCreationArgs* route, double* latency)
 250 {
 251   // Minimal routing version.
 252   // TODO : non-minimal random one, and adaptive ?
 253
 254   if (dst->is_router() || src->is_router())
 255     return;
 256
 257   XBT_VERB("dragonfly getLocalRoute from '%s'[%u] to '%s'[%u]", src->get_cname(), src->id(), dst->get_cname(),
 258            dst->id());
 259
 260   if ((src->id() == dst->id()) && has_loopback_) {
 261     std::pair<resource::LinkImpl*, resource::LinkImpl*> info = private_links_.at(node_pos(src->id()));
 262
 263     route->link_list.push_back(info.first);
 264     if (latency)
 265       *latency += info.first->get_latency();
 266     return;
 267   }
 268
 269   const auto myCoords     = rankId_to_coords(src->id());
 270   const auto targetCoords = rankId_to_coords(dst->id());
 271   XBT_DEBUG("src : %u group, %u chassis, %u blade, %u node", myCoords.group, myCoords.chassis, myCoords.blade,
 272             myCoords.node);
 273   XBT_DEBUG("dst : %u group, %u chassis, %u blade, %u node", targetCoords.group, targetCoords.chassis,
 274             targetCoords.blade, targetCoords.node);
 275
 276   DragonflyRouter* myRouter = &routers_[myCoords.group * (num_chassis_per_group_ * num_blades_per_chassis_) +
 277                                         myCoords.chassis * num_blades_per_chassis_ + myCoords.blade];
 278   DragonflyRouter* targetRouter = &routers_[targetCoords.group * (num_chassis_per_group_ * num_blades_per_chassis_) +
 279                                             targetCoords.chassis * num_blades_per_chassis_ + targetCoords.blade];
 280   DragonflyRouter* currentRouter = myRouter;
 281
 282   // node->router local link
 283   route->link_list.push_back(myRouter->my_nodes_[myCoords.node * num_links_per_link_]);
 284   if (latency)
 285     *latency += myRouter->my_nodes_[myCoords.node * num_links_per_link_]->get_latency();
 286
 287   if (has_limiter_) { // limiter for sender
 288     std::pair<resource::LinkImpl*, resource::LinkImpl*> info = private_links_.at(node_pos_with_loopback(src->id()));
 289     route->link_list.push_back(info.first);
 290   }
 291
 292   if (targetRouter != myRouter) {
 293     // are we on a different group ?
 294     if (targetRouter->group_ != currentRouter->group_) {
 295       // go to the router of our group connected to this one.
 296       if (currentRouter->blade_ != targetCoords.group) {
 297         // go to the nth router in our chassis
 298         route->link_list.push_back(currentRouter->green_links_[targetCoords.group]);
 299         if (latency)
 300           *latency += currentRouter->green_links_[targetCoords.group]->get_latency();
 301         currentRouter = &routers_[myCoords.group * (num_chassis_per_group_ * num_blades_per_chassis_) +
 302                                   myCoords.chassis * num_blades_per_chassis_ + targetCoords.group];
 303       }
 304
 305       if (currentRouter->chassis_ != 0) {
 306         // go to the first chassis of our group
 307         route->link_list.push_back(currentRouter->black_links_[0]);
 308         if (latency)
 309           *latency += currentRouter->black_links_[0]->get_latency();
 310         currentRouter =
 311             &routers_[myCoords.group * (num_chassis_per_group_ * num_blades_per_chassis_) + targetCoords.group];
 312       }
 313
 314       // go to destination group - the only optical hop
 315       route->link_list.push_back(currentRouter->blue_link_);
 316       if (latency)
 317         *latency += currentRouter->blue_link_->get_latency();
 318       currentRouter =
 319           &routers_[targetCoords.group * (num_chassis_per_group_ * num_blades_per_chassis_) + myCoords.group];
 320     }
 321
 322     // same group, but same blade ?
 323     if (targetRouter->blade_ != currentRouter->blade_) {
 324       route->link_list.push_back(currentRouter->green_links_[targetCoords.blade]);
 325       if (latency)
 326         *latency += currentRouter->green_links_[targetCoords.blade]->get_latency();
 327       currentRouter =
 328           &routers_[targetCoords.group * (num_chassis_per_group_ * num_blades_per_chassis_) + targetCoords.blade];
 329     }
 330
 331     // same blade, but same chassis ?
 332     if (targetRouter->chassis_ != currentRouter->chassis_) {
 333       route->link_list.push_back(currentRouter->black_links_[targetCoords.chassis]);
 334       if (latency)
 335         *latency += currentRouter->black_links_[targetCoords.chassis]->get_latency();
 336     }
 337   }
 338
 339   if (has_limiter_) { // limiter for receiver
 340     std::pair<resource::LinkImpl*, resource::LinkImpl*> info = private_links_.at(node_pos_with_loopback(dst->id()));
 341     route->link_list.push_back(info.first);
 342   }
 343
 344   // router->node local link
 345   route->link_list.push_back(
 346       targetRouter->my_nodes_[targetCoords.node * num_links_per_link_ + num_links_per_link_ - 1]);
 347   if (latency)
 348     *latency +=
 349         targetRouter->my_nodes_[targetCoords.node * num_links_per_link_ + num_links_per_link_ - 1]->get_latency();
 350 }
 351 }
 352 }
 353 } // namespace