src/surf/network_cm02.cpp

   1 /* Copyright (c) 2013-2018. 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 <algorithm>
   7
   8 #include "network_cm02.hpp"
   9 #include "simgrid/s4u/Host.hpp"
  10 #include "simgrid/sg_config.h"
  11 #include "src/instr/instr_private.hpp" // TRACE_is_enabled(). FIXME: remove by subscribing tracing to the surf signals
  12 #include "src/kernel/lmm/maxmin.hpp"
  13
  14 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_network);
  15
  16 double sg_latency_factor = 1.0; /* default value; can be set by model or from command line */
  17 double sg_bandwidth_factor = 1.0;       /* default value; can be set by model or from command line */
  18 double sg_weight_S_parameter = 0.0;     /* default value; can be set by model or from command line */
  19
  20 double sg_tcp_gamma = 0.0;
  21 int sg_network_crosstraffic = 0;
  22
  23 /************************************************************************/
  24 /* New model based on optimizations discussed during Pedro Velho's thesis*/
  25 /************************************************************************/
  26 /* @techreport{VELHO:2011:HAL-00646896:1, */
  27 /*      url = {http://hal.inria.fr/hal-00646896/en/}, */
  28 /*      title = {{Flow-level network models: have we reached the limits?}}, */
  29 /*      author = {Velho, Pedro and Schnorr, Lucas and Casanova, Henri and Legrand, Arnaud}, */
  30 /*      type = {Rapport de recherche}, */
  31 /*      institution = {INRIA}, */
  32 /*      number = {RR-7821}, */
  33 /*      year = {2011}, */
  34 /*      month = Nov, */
  35 /*      pdf = {http://hal.inria.fr/hal-00646896/PDF/rr-validity.pdf}, */
  36 /*  } */
  37 void surf_network_model_init_LegrandVelho()
  38 {
  39   if (surf_network_model)
  40     return;
  41
  42   surf_network_model = new simgrid::surf::NetworkCm02Model();
  43   all_existing_models->push_back(surf_network_model);
  44
  45   xbt_cfg_setdefault_double("network/latency-factor",      13.01);
  46   xbt_cfg_setdefault_double("network/bandwidth-factor",     0.97);
  47   xbt_cfg_setdefault_double("network/weight-S",         20537);
  48 }
  49
  50 /***************************************************************************/
  51 /* The nice TCP sharing model designed by Loris Marchal and Henri Casanova */
  52 /***************************************************************************/
  53 /* @TechReport{      rr-lip2002-40, */
  54 /*   author        = {Henri Casanova and Loris Marchal}, */
  55 /*   institution   = {LIP}, */
  56 /*   title         = {A Network Model for Simulation of Grid Application}, */
  57 /*   number        = {2002-40}, */
  58 /*   month         = {oct}, */
  59 /*   year          = {2002} */
  60 /* } */
  61 void surf_network_model_init_CM02()
  62 {
  63
  64   if (surf_network_model)
  65     return;
  66
  67   xbt_cfg_setdefault_double("network/latency-factor",   1.0);
  68   xbt_cfg_setdefault_double("network/bandwidth-factor", 1.0);
  69   xbt_cfg_setdefault_double("network/weight-S",         0.0);
  70
  71   surf_network_model = new simgrid::surf::NetworkCm02Model();
  72   all_existing_models->push_back(surf_network_model);
  73 }
  74
  75 /***************************************************************************/
  76 /* The models from Steven H. Low                                           */
  77 /***************************************************************************/
  78 /* @article{Low03,                                                         */
  79 /*   author={Steven H. Low},                                               */
  80 /*   title={A Duality Model of {TCP} and Queue Management Algorithms},     */
  81 /*   year={2003},                                                          */
  82 /*   journal={{IEEE/ACM} Transactions on Networking},                      */
  83 /*    volume={11}, number={4},                                             */
  84 /*  }                                                                      */
  85 void surf_network_model_init_Reno()
  86 {
  87   if (surf_network_model)
  88     return;
  89
  90   set_default_protocol_function(simgrid::kernel::lmm::func_reno_f, simgrid::kernel::lmm::func_reno_fp,
  91                                 simgrid::kernel::lmm::func_reno_fpi);
  92
  93   xbt_cfg_setdefault_double("network/latency-factor", 13.01);
  94   xbt_cfg_setdefault_double("network/bandwidth-factor", 0.97);
  95   xbt_cfg_setdefault_double("network/weight-S", 20537);
  96
  97   surf_network_model = new simgrid::surf::NetworkCm02Model(&simgrid::kernel::lmm::lagrange_solve);
  98   all_existing_models->push_back(surf_network_model);
  99 }
 100
 101
 102 void surf_network_model_init_Reno2()
 103 {
 104   if (surf_network_model)
 105     return;
 106
 107   set_default_protocol_function(simgrid::kernel::lmm::func_reno2_f, simgrid::kernel::lmm::func_reno2_fp,
 108                                 simgrid::kernel::lmm::func_reno2_fpi);
 109
 110   xbt_cfg_setdefault_double("network/latency-factor", 13.01);
 111   xbt_cfg_setdefault_double("network/bandwidth-factor", 0.97);
 112   xbt_cfg_setdefault_double("network/weight-S", 20537);
 113
 114   surf_network_model = new simgrid::surf::NetworkCm02Model(&simgrid::kernel::lmm::lagrange_solve);
 115   all_existing_models->push_back(surf_network_model);
 116 }
 117
 118 void surf_network_model_init_Vegas()
 119 {
 120   if (surf_network_model)
 121     return;
 122
 123   set_default_protocol_function(simgrid::kernel::lmm::func_vegas_f, simgrid::kernel::lmm::func_vegas_fp,
 124                                 simgrid::kernel::lmm::func_vegas_fpi);
 125
 126   xbt_cfg_setdefault_double("network/latency-factor", 13.01);
 127   xbt_cfg_setdefault_double("network/bandwidth-factor", 0.97);
 128   xbt_cfg_setdefault_double("network/weight-S", 20537);
 129
 130   surf_network_model = new simgrid::surf::NetworkCm02Model(&simgrid::kernel::lmm::lagrange_solve);
 131   all_existing_models->push_back(surf_network_model);
 132 }
 133
 134 namespace simgrid {
 135 namespace surf {
 136
 137 NetworkCm02Model::NetworkCm02Model()
 138   :NetworkModel()
 139 {
 140   std::string optim = xbt_cfg_get_string("network/optim");
 141   bool select = xbt_cfg_get_boolean("network/maxmin-selective-update");
 142
 143   if (optim == "Full") {
 144     setUpdateMechanism(UM_FULL);
 145   } else if (optim == "Lazy") {
 146     select = true;
 147     setUpdateMechanism(UM_LAZY);
 148     xbt_assert(select || (xbt_cfg_is_default_value("network/maxmin-selective-update")),
 149                "You cannot disable selective update when using the lazy update mechanism");
 150   } else {
 151     xbt_die("Unsupported optimization (%s) for this model. Accepted: Full, Lazy.", optim.c_str());
 152   }
 153
 154   maxminSystem_ = new simgrid::kernel::lmm::System(select);
 155   loopback_     = NetworkCm02Model::createLink("__loopback__", 498000000, 0.000015, SURF_LINK_FATPIPE);
 156
 157   if (getUpdateMechanism() == UM_LAZY) {
 158     modifiedSet_              = new kernel::resource::ActionLmmList();
 159     maxminSystem_->keep_track = modifiedSet_;
 160   }
 161 }
 162
 163 NetworkCm02Model::NetworkCm02Model(void (*specificSolveFun)(lmm_system_t self)) : NetworkCm02Model()
 164 {
 165   maxminSystem_->solve_fun = specificSolveFun;
 166 }
 167
 168 LinkImpl* NetworkCm02Model::createLink(const std::string& name, double bandwidth, double latency,
 169                                        e_surf_link_sharing_policy_t policy)
 170 {
 171   return new NetworkCm02Link(this, name, bandwidth, latency, policy, maxminSystem_);
 172 }
 173
 174 void NetworkCm02Model::updateActionsStateLazy(double now, double /*delta*/)
 175 {
 176   while (not actionHeapIsEmpty() && double_equals(actionHeapTopDate(), now, sg_surf_precision)) {
 177
 178     NetworkCm02Action* action = static_cast<NetworkCm02Action*>(actionHeapPop());
 179     XBT_DEBUG("Something happened to action %p", action);
 180     if (TRACE_is_enabled()) {
 181       int n = action->getVariable()->get_number_of_constraint();
 182
 183       for (int i = 0; i < n; i++){
 184         kernel::lmm::Constraint* constraint = action->getVariable()->get_constraint(i);
 185         NetworkCm02Link* link       = static_cast<NetworkCm02Link*>(constraint->get_id());
 186         double value = action->getVariable()->get_value() * action->getVariable()->get_constraint_weight(i);
 187         TRACE_surf_link_set_utilization(link->getCname(), action->getCategory(), value, action->getLastUpdate(),
 188                                         now - action->getLastUpdate());
 189       }
 190     }
 191
 192     // if I am wearing a latency hat
 193     if (action->getType() == kernel::resource::Action::Type::LATENCY) {
 194       XBT_DEBUG("Latency paid for action %p. Activating", action);
 195       maxminSystem_->update_variable_weight(action->getVariable(), action->weight_);
 196       action->heapRemove(getActionHeap());
 197       action->refreshLastUpdate();
 198
 199         // if I am wearing a max_duration or normal hat
 200     } else if (action->getType() == kernel::resource::Action::Type::MAX_DURATION ||
 201                action->getType() == kernel::resource::Action::Type::NORMAL) {
 202       // no need to communicate anymore
 203       // assume that flows that reached max_duration have remaining of 0
 204       XBT_DEBUG("Action %p finished", action);
 205       action->setRemains(0);
 206       action->finish(kernel::resource::Action::State::done);
 207       action->heapRemove(getActionHeap());
 208     }
 209   }
 210 }
 211
 212
 213 void NetworkCm02Model::updateActionsStateFull(double now, double delta)
 214 {
 215   for (auto it = std::begin(*getRunningActionSet()); it != std::end(*getRunningActionSet());) {
 216     NetworkCm02Action& action = static_cast<NetworkCm02Action&>(*it);
 217     ++it; // increment iterator here since the following calls to action.finish() may invalidate it
 218     XBT_DEBUG("Something happened to action %p", &action);
 219     double deltap = delta;
 220     if (action.latency_ > 0) {
 221       if (action.latency_ > deltap) {
 222         double_update(&action.latency_, deltap, sg_surf_precision);
 223         deltap = 0.0;
 224       } else {
 225         double_update(&deltap, action.latency_, sg_surf_precision);
 226         action.latency_ = 0.0;
 227       }
 228       if (action.latency_ <= 0.0 && not action.isSuspended())
 229         maxminSystem_->update_variable_weight(action.getVariable(), action.weight_);
 230     }
 231     if (TRACE_is_enabled()) {
 232       int n = action.getVariable()->get_number_of_constraint();
 233       for (int i = 0; i < n; i++) {
 234         kernel::lmm::Constraint* constraint = action.getVariable()->get_constraint(i);
 235         NetworkCm02Link* link = static_cast<NetworkCm02Link*>(constraint->get_id());
 236         TRACE_surf_link_set_utilization(
 237             link->getCname(), action.getCategory(),
 238             (action.getVariable()->get_value() * action.getVariable()->get_constraint_weight(i)),
 239             action.getLastUpdate(), now - action.getLastUpdate());
 240       }
 241     }
 242     if (not action.getVariable()->get_number_of_constraint()) {
 243       /* There is actually no link used, hence an infinite bandwidth. This happens often when using models like
 244        * vivaldi. In such case, just make sure that the action completes immediately.
 245        */
 246       action.updateRemains(action.getRemains());
 247     }
 248     action.updateRemains(action.getVariable()->get_value() * delta);
 249
 250     if (action.getMaxDuration() > NO_MAX_DURATION)
 251       action.updateMaxDuration(delta);
 252
 253     if (((action.getRemains() <= 0) && (action.getVariable()->get_weight() > 0)) ||
 254         ((action.getMaxDuration() > NO_MAX_DURATION) && (action.getMaxDuration() <= 0))) {
 255       action.finish(kernel::resource::Action::State::done);
 256     }
 257   }
 258 }
 259
 260 kernel::resource::Action* NetworkCm02Model::communicate(s4u::Host* src, s4u::Host* dst, double size, double rate)
 261 {
 262   int failed = 0;
 263   double latency = 0.0;
 264   std::vector<LinkImpl*> back_route;
 265   std::vector<LinkImpl*> route;
 266
 267   XBT_IN("(%s,%s,%g,%g)", src->getCname(), dst->getCname(), size, rate);
 268
 269   src->routeTo(dst, route, &latency);
 270   xbt_assert(not route.empty() || latency,
 271              "You're trying to send data from %s to %s but there is no connecting path between these two hosts.",
 272              src->getCname(), dst->getCname());
 273
 274   for (auto const& link : route)
 275     if (link->isOff())
 276       failed = 1;
 277
 278   if (sg_network_crosstraffic == 1) {
 279     dst->routeTo(src, back_route, nullptr);
 280     for (auto const& link : back_route)
 281       if (link->isOff())
 282         failed = 1;
 283   }
 284
 285   NetworkCm02Action *action = new NetworkCm02Action(this, size, failed);
 286   action->weight_ = latency;
 287   action->latency_ = latency;
 288   action->rate_ = rate;
 289   if (getUpdateMechanism() == UM_LAZY) {
 290     action->refreshLastUpdate();
 291   }
 292
 293   double bandwidth_bound = -1.0;
 294   if (sg_weight_S_parameter > 0)
 295     for (auto const& link : route)
 296       action->weight_ += sg_weight_S_parameter / link->bandwidth();
 297
 298   for (auto const& link : route) {
 299     double bb       = bandwidthFactor(size) * link->bandwidth();
 300     bandwidth_bound = (bandwidth_bound < 0.0) ? bb : std::min(bandwidth_bound, bb);
 301   }
 302
 303   action->latCurrent_ = action->latency_;
 304   action->latency_ *= latencyFactor(size);
 305   action->rate_ = bandwidthConstraint(action->rate_, bandwidth_bound, size);
 306
 307   int constraints_per_variable = route.size();
 308   constraints_per_variable += back_route.size();
 309
 310   if (action->latency_ > 0) {
 311     action->setVariable(maxminSystem_->variable_new(action, 0.0, -1.0, constraints_per_variable));
 312     if (getUpdateMechanism() == UM_LAZY) {
 313       // add to the heap the event when the latency is payed
 314       XBT_DEBUG("Added action (%p) one latency event at date %f", action, action->latency_ + action->getLastUpdate());
 315       action->heapInsert(getActionHeap(), action->latency_ + action->getLastUpdate(),
 316                          route.empty() ? kernel::resource::Action::Type::NORMAL
 317                                        : kernel::resource::Action::Type::LATENCY);
 318     }
 319   } else
 320     action->setVariable(maxminSystem_->variable_new(action, 1.0, -1.0, constraints_per_variable));
 321
 322   if (action->rate_ < 0) {
 323     maxminSystem_->update_variable_bound(action->getVariable(),
 324                                          (action->latCurrent_ > 0) ? sg_tcp_gamma / (2.0 * action->latCurrent_) : -1.0);
 325   } else {
 326     maxminSystem_->update_variable_bound(action->getVariable(),
 327                                          (action->latCurrent_ > 0)
 328                                              ? std::min(action->rate_, sg_tcp_gamma / (2.0 * action->latCurrent_))
 329                                              : action->rate_);
 330   }
 331
 332   for (auto const& link : route)
 333     maxminSystem_->expand(link->constraint(), action->getVariable(), 1.0);
 334
 335   if (not back_route.empty()) { //  sg_network_crosstraffic was activated
 336     XBT_DEBUG("Crosstraffic active adding backward flow using 5%%");
 337     for (auto const& link : back_route)
 338       maxminSystem_->expand(link->constraint(), action->getVariable(), .05);
 339
 340     // Change concurrency_share here, if you want that cross-traffic is included in the SURF concurrency
 341     // (You would also have to change simgrid::kernel::lmm::Element::get_concurrency())
 342     // action->getVariable()->set_concurrency_share(2)
 343   }
 344   XBT_OUT();
 345
 346   simgrid::s4u::Link::onCommunicate(action, src, dst);
 347   return action;
 348 }
 349
 350 /************
 351  * Resource *
 352  ************/
 353 NetworkCm02Link::NetworkCm02Link(NetworkCm02Model* model, const std::string& name, double bandwidth, double latency,
 354                                  e_surf_link_sharing_policy_t policy, lmm_system_t system)
 355     : LinkImpl(model, name, system->constraint_new(this, sg_bandwidth_factor * bandwidth))
 356 {
 357   bandwidth_.scale = 1.0;
 358   bandwidth_.peak  = bandwidth;
 359
 360   latency_.scale = 1.0;
 361   latency_.peak  = latency;
 362
 363   if (policy == SURF_LINK_FATPIPE)
 364     constraint()->unshare();
 365
 366   simgrid::s4u::Link::onCreation(this->piface_);
 367 }
 368
 369 void NetworkCm02Link::apply_event(tmgr_trace_event_t triggered, double value)
 370 {
 371   /* Find out which of my iterators was triggered, and react accordingly */
 372   if (triggered == bandwidth_.event) {
 373     setBandwidth(value);
 374     tmgr_trace_event_unref(&bandwidth_.event);
 375
 376   } else if (triggered == latency_.event) {
 377     setLatency(value);
 378     tmgr_trace_event_unref(&latency_.event);
 379
 380   } else if (triggered == stateEvent_) {
 381     if (value > 0)
 382       turnOn();
 383     else {
 384       kernel::lmm::Variable* var = nullptr;
 385       const_lmm_element_t elem = nullptr;
 386       double now               = surf_get_clock();
 387
 388       turnOff();
 389       while ((var = constraint()->get_variable(&elem))) {
 390         kernel::resource::Action* action = static_cast<kernel::resource::Action*>(var->get_id());
 391
 392         if (action->getState() == kernel::resource::Action::State::running ||
 393             action->getState() == kernel::resource::Action::State::ready) {
 394           action->setFinishTime(now);
 395           action->setState(kernel::resource::Action::State::failed);
 396         }
 397       }
 398     }
 399     tmgr_trace_event_unref(&stateEvent_);
 400   } else {
 401     xbt_die("Unknown event!\n");
 402   }
 403
 404   XBT_DEBUG("There was a resource state event, need to update actions related to the constraint (%p)", constraint());
 405 }
 406
 407 void NetworkCm02Link::setBandwidth(double value)
 408 {
 409   bandwidth_.peak = value;
 410
 411   model()->getMaxminSystem()->update_constraint_bound(constraint(),
 412                                                       sg_bandwidth_factor * (bandwidth_.peak * bandwidth_.scale));
 413   TRACE_surf_link_set_bandwidth(surf_get_clock(), getCname(), sg_bandwidth_factor * bandwidth_.peak * bandwidth_.scale);
 414
 415   if (sg_weight_S_parameter > 0) {
 416     double delta = sg_weight_S_parameter / value - sg_weight_S_parameter / (bandwidth_.peak * bandwidth_.scale);
 417
 418     kernel::lmm::Variable* var;
 419     const_lmm_element_t elem     = nullptr;
 420     const_lmm_element_t nextelem = nullptr;
 421     int numelem                  = 0;
 422     while ((var = constraint()->get_variable_safe(&elem, &nextelem, &numelem))) {
 423       NetworkCm02Action* action = static_cast<NetworkCm02Action*>(var->get_id());
 424       action->weight_ += delta;
 425       if (not action->isSuspended())
 426         model()->getMaxminSystem()->update_variable_weight(action->getVariable(), action->weight_);
 427     }
 428   }
 429 }
 430
 431 void NetworkCm02Link::setLatency(double value)
 432 {
 433   double delta                 = value - latency_.peak;
 434   kernel::lmm::Variable* var   = nullptr;
 435   const_lmm_element_t elem     = nullptr;
 436   const_lmm_element_t nextelem = nullptr;
 437   int numelem                  = 0;
 438
 439   latency_.peak = value;
 440
 441   while ((var = constraint()->get_variable_safe(&elem, &nextelem, &numelem))) {
 442     NetworkCm02Action* action = static_cast<NetworkCm02Action*>(var->get_id());
 443     action->latCurrent_ += delta;
 444     action->weight_ += delta;
 445     if (action->rate_ < 0)
 446       model()->getMaxminSystem()->update_variable_bound(action->getVariable(),
 447                                                         sg_tcp_gamma / (2.0 * action->latCurrent_));
 448     else {
 449       model()->getMaxminSystem()->update_variable_bound(
 450           action->getVariable(), std::min(action->rate_, sg_tcp_gamma / (2.0 * action->latCurrent_)));
 451
 452       if (action->rate_ < sg_tcp_gamma / (2.0 * action->latCurrent_)) {
 453         XBT_INFO("Flow is limited BYBANDWIDTH");
 454       } else {
 455         XBT_INFO("Flow is limited BYLATENCY, latency of flow is %f", action->latCurrent_);
 456       }
 457     }
 458     if (not action->isSuspended())
 459       model()->getMaxminSystem()->update_variable_weight(action->getVariable(), action->weight_);
 460   }
 461 }
 462
 463 /**********
 464  * Action *
 465  **********/
 466
 467 void NetworkCm02Action::updateRemainingLazy(double now)
 468 {
 469   if (suspended_ != Action::SuspendStates::not_suspended)
 470     return;
 471
 472   double delta        = now - getLastUpdate();
 473   double max_duration = getMaxDuration();
 474
 475   if (getRemainsNoUpdate() > 0) {
 476     XBT_DEBUG("Updating action(%p): remains was %f, last_update was: %f", this, getRemainsNoUpdate(), getLastUpdate());
 477     updateRemains(getLastValue() * delta);
 478
 479     XBT_DEBUG("Updating action(%p): remains is now %f", this, getRemainsNoUpdate());
 480   }
 481
 482   if (max_duration > NO_MAX_DURATION) {
 483     double_update(&max_duration, delta, sg_surf_precision);
 484     setMaxDuration(max_duration);
 485   }
 486
 487   if ((getRemainsNoUpdate() <= 0 && (getVariable()->get_weight() > 0)) ||
 488       ((max_duration > NO_MAX_DURATION) && (max_duration <= 0))) {
 489     finish(Action::State::done);
 490     heapRemove(getModel()->getActionHeap());
 491   }
 492
 493   refreshLastUpdate();
 494   setLastValue(getVariable()->get_value());
 495 }
 496
 497 }
 498 }