src/surf/host_ptask_L07.cpp

   1 /* Copyright (c) 2007-2010, 2013-2015. The SimGrid Team.
   2  * All rights reserved.                                                     */
   3
   4 /* This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the license (GNU LGPL) which comes with this package. */
   6
   7 #include <cstdlib>
   8
   9 #include <algorithm>
  10
  11 #include "host_ptask_L07.hpp"
  12
  13 #include "cpu_interface.hpp"
  14 #include "surf_routing.hpp"
  15 #include "xbt/lib.h"
  16 #include "src/surf/platform.hpp"
  17
  18 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_host);
  19 XBT_LOG_EXTERNAL_CATEGORY(xbt_cfg);
  20
  21 /**************************************/
  22 /*** Resource Creation & Destruction **/
  23 /**************************************/
  24
  25 static void ptask_netlink_parse_init(sg_platf_link_cbarg_t link)
  26 {
  27   netlink_parse_init(link);
  28   current_property_set = NULL;
  29 }
  30
  31 void surf_host_model_init_ptask_L07(void)
  32 {
  33   XBT_CINFO(xbt_cfg,"Switching to the L07 model to handle parallel tasks.");
  34   xbt_assert(!surf_cpu_model_pm, "CPU model type already defined");
  35   xbt_assert(!surf_network_model, "network model type already defined");
  36
  37   // Define the callbacks to parse the XML
  38   simgrid::surf::on_link.connect(ptask_netlink_parse_init);
  39   simgrid::surf::on_postparse.connect([](){
  40       surf_host_model->addTraces();
  41   });
  42
  43   surf_host_model = new simgrid::surf::HostL07Model();
  44   xbt_dynar_push(all_existing_models, &surf_host_model);
  45 }
  46
  47
  48 namespace simgrid {
  49 namespace surf {
  50
  51 HostL07Model::HostL07Model() : HostModel() {
  52   p_maxminSystem = lmm_system_new(1);
  53   surf_network_model = new NetworkL07Model(this,p_maxminSystem);
  54   surf_cpu_model_pm = new CpuL07Model(this,p_maxminSystem);
  55
  56   routing_model_create(surf_network_model->createLink("__loopback__",
  57                                                     498000000, NULL,
  58                                                     0.000015, NULL,
  59                                                     1/*ON*/, NULL,
  60                                                     SURF_LINK_FATPIPE, NULL));
  61 }
  62
  63 HostL07Model::~HostL07Model() {
  64   delete surf_cpu_model_pm;
  65   delete surf_network_model;
  66 }
  67
  68 CpuL07Model::CpuL07Model(HostL07Model *hmodel,lmm_system_t sys)
  69   : CpuModel()
  70   , p_hostModel(hmodel)
  71   {
  72     p_maxminSystem = sys;
  73   }
  74 CpuL07Model::~CpuL07Model() {
  75   surf_cpu_model_pm = NULL;
  76   lmm_system_free(p_maxminSystem);
  77   p_maxminSystem = NULL;
  78 }
  79 NetworkL07Model::NetworkL07Model(HostL07Model *hmodel, lmm_system_t sys)
  80   : NetworkModel()
  81   , p_hostModel(hmodel)
  82   {
  83     p_maxminSystem = sys;
  84   }
  85 NetworkL07Model::~NetworkL07Model()
  86 {
  87   surf_network_model = NULL;
  88   p_maxminSystem = NULL; // Avoid multi-free
  89 }
  90
  91
  92 double HostL07Model::shareResources(double /*now*/)
  93 {
  94   L07Action *action;
  95
  96   ActionList *running_actions = getRunningActionSet();
  97   double min = this->shareResourcesMaxMin(running_actions,
  98                                               p_maxminSystem,
  99                                               bottleneck_solve);
 100
 101   for(ActionList::iterator it(running_actions->begin()), itend(running_actions->end())
 102    ; it != itend ; ++it) {
 103   action = static_cast<L07Action*>(&*it);
 104     if (action->m_latency > 0) {
 105       if (min < 0) {
 106         min = action->m_latency;
 107         XBT_DEBUG("Updating min (value) with %p (start %f): %f", action,
 108                action->getStartTime(), min);
 109       } else if (action->m_latency < min) {
 110         min = action->m_latency;
 111         XBT_DEBUG("Updating min (latency) with %p (start %f): %f", action,
 112                action->getStartTime(), min);
 113       }
 114     }
 115   }
 116
 117   XBT_DEBUG("min value : %f", min);
 118
 119   return min;
 120 }
 121
 122 void HostL07Model::updateActionsState(double /*now*/, double delta) {
 123
 124   L07Action *action;
 125   ActionList *actionSet = getRunningActionSet();
 126
 127   for(ActionList::iterator it = actionSet->begin(), itNext = it
 128    ; it != actionSet->end()
 129    ; it =  itNext) {
 130   ++itNext;
 131     action = static_cast<L07Action*>(&*it);
 132     if (action->m_latency > 0) {
 133       if (action->m_latency > delta) {
 134         double_update(&(action->m_latency), delta, sg_surf_precision);
 135       } else {
 136         action->m_latency = 0.0;
 137       }
 138       if ((action->m_latency == 0.0) && (action->isSuspended() == 0)) {
 139         action->updateBound();
 140         lmm_update_variable_weight(p_maxminSystem, action->getVariable(), 1.0);
 141       }
 142     }
 143     XBT_DEBUG("Action (%p) : remains (%g) updated by %g.",
 144            action, action->getRemains(), lmm_variable_getvalue(action->getVariable()) * delta);
 145     action->updateRemains(lmm_variable_getvalue(action->getVariable()) * delta);
 146
 147     if (action->getMaxDuration() != NO_MAX_DURATION)
 148       action->updateMaxDuration(delta);
 149
 150     XBT_DEBUG("Action (%p) : remains (%g).", action, action->getRemains());
 151
 152     /* In the next if cascade, the action can be finished either because:
 153      *  - The amount of remaining work reached 0
 154      *  - The max duration was reached
 155      * If it's not done, it may have failed.
 156      */
 157
 158     if ((action->getRemains() <= 0) &&
 159         (lmm_get_variable_weight(action->getVariable()) > 0)) {
 160       action->finish();
 161       action->setState(SURF_ACTION_DONE);
 162     } else if ((action->getMaxDuration() != NO_MAX_DURATION) &&
 163                (action->getMaxDuration() <= 0)) {
 164       action->finish();
 165       action->setState(SURF_ACTION_DONE);
 166     } else {
 167       /* Need to check that none of the model has failed */
 168       lmm_constraint_t cnst = NULL;
 169       int i = 0;
 170
 171       while ((cnst = lmm_get_cnst_from_var(p_maxminSystem, action->getVariable(), i++))) {
 172         void *constraint_id = lmm_constraint_id(cnst);
 173
 174         if (static_cast<Host*>(constraint_id)->isOff()) {
 175           XBT_DEBUG("Action (%p) Failed!!", action);
 176           action->finish();
 177           action->setState(SURF_ACTION_FAILED);
 178           break;
 179         }
 180       }
 181     }
 182   }
 183   return;
 184 }
 185
 186 Action *HostL07Model::executeParallelTask(int host_nb, sg_host_t *host_list,
 187       double *flops_amount, double *bytes_amount,
 188       double rate) {
 189   return new L07Action(this, host_nb, host_list, flops_amount, bytes_amount, rate);
 190 }
 191
 192
 193 L07Action::L07Action(Model *model, int host_nb,
 194     sg_host_t*host_list,
 195     double *flops_amount,
 196     double *bytes_amount,
 197     double rate)
 198   : CpuAction(model, 1, 0)
 199 {
 200   unsigned int cpt;
 201   int nb_link = 0;
 202   int nb_used_host = 0; /* Only the hosts with something to compute (>0 flops) are counted) */
 203   double latency = 0.0;
 204
 205   this->p_netcardList->reserve(host_nb);
 206   for (int i = 0; i<host_nb; i++)
 207     this->p_netcardList->push_back(host_list[i]->pimpl_netcard);
 208
 209   /* Compute the number of affected resources... */
 210   if(bytes_amount != NULL) {
 211     xbt_dict_t ptask_parallel_task_link_set = xbt_dict_new_homogeneous(NULL);
 212
 213     for (int i = 0; i < host_nb; i++) {
 214       for (int j = 0; j < host_nb; j++) {
 215
 216         if (bytes_amount[i * host_nb + j] > 0) {
 217           double lat=0.0;
 218           xbt_dynar_t route=NULL;
 219
 220           routing_platf->getRouteAndLatency((*p_netcardList)[i], (*p_netcardList)[j], &route, &lat);
 221           latency = MAX(latency, lat);
 222
 223           void *_link;
 224           xbt_dynar_foreach(route, cpt, _link) {
 225             LinkL07 *link = static_cast<LinkL07*>(_link);
 226             xbt_dict_set(ptask_parallel_task_link_set, link->getName(), link, NULL);
 227           }
 228         }
 229       }
 230     }
 231
 232     nb_link = xbt_dict_length(ptask_parallel_task_link_set);
 233     xbt_dict_free(&ptask_parallel_task_link_set);
 234   }
 235
 236   for (int i = 0; i < host_nb; i++)
 237     if (flops_amount[i] > 0)
 238       nb_used_host++;
 239
 240   XBT_DEBUG("Creating a parallel task (%p) with %d hosts and %d unique links.", this, host_nb, nb_link);
 241   this->p_computationAmount = flops_amount;
 242   this->p_communicationAmount = bytes_amount;
 243   this->m_latency = latency;
 244   this->m_rate = rate;
 245
 246   this->p_variable = lmm_variable_new(model->getMaxminSystem(), this, 1.0,
 247       (rate > 0 ? rate : -1.0),
 248       host_nb + nb_link);
 249
 250   if (this->m_latency > 0)
 251     lmm_update_variable_weight(model->getMaxminSystem(), this->getVariable(), 0.0);
 252
 253   for (int i = 0; i < host_nb; i++)
 254     lmm_expand(model->getMaxminSystem(), host_list[i]->pimpl_cpu->getConstraint(),
 255         this->getVariable(), flops_amount[i]);
 256
 257   if(bytes_amount != NULL) {
 258     for (int i = 0; i < host_nb; i++) {
 259       for (int j = 0; j < host_nb; j++) {
 260
 261         xbt_dynar_t route=NULL;
 262         if (bytes_amount[i * host_nb + j] == 0.0)
 263           continue;
 264
 265         routing_platf->getRouteAndLatency((*p_netcardList)[i], (*p_netcardList)[j],
 266                                                     &route, NULL);
 267
 268         void *_link;
 269         xbt_dynar_foreach(route, cpt, _link) {
 270           LinkL07 *link = static_cast<LinkL07*>(_link);
 271           lmm_expand_add(model->getMaxminSystem(), link->getConstraint(),
 272                         this->getVariable(), bytes_amount[i * host_nb + j]);
 273         }
 274       }
 275     }
 276   }
 277
 278   if (nb_link + nb_used_host == 0) {
 279     this->setCost(1.0);
 280     this->setRemains(0.0);
 281   }
 282   xbt_free(host_list);
 283 }
 284
 285 Action *NetworkL07Model::communicate(NetCard *src, NetCard *dst,
 286                                        double size, double rate)
 287 {
 288   sg_host_t*host_list = xbt_new0(sg_host_t, 2);
 289   double *flops_amount = xbt_new0(double, 2);
 290   double *bytes_amount = xbt_new0(double, 4);
 291   Action *res = NULL;
 292
 293   host_list[0] = sg_host_by_name(src->getName());
 294   host_list[1] = sg_host_by_name(dst->getName());
 295   bytes_amount[1] = size;
 296
 297   res = p_hostModel->executeParallelTask(2, host_list, flops_amount, bytes_amount, rate);
 298
 299   return res;
 300 }
 301
 302 Cpu *CpuL07Model::createCpu(simgrid::s4u::Host *host,  xbt_dynar_t powerPeakList,
 303                           int pstate, double power_scale,
 304                           tmgr_trace_t power_trace, int core,
 305                           int initiallyOn,
 306                           tmgr_trace_t state_trace)
 307 {
 308   CpuL07 *cpu = new CpuL07(this, host, powerPeakList, pstate, power_scale, power_trace,
 309                          core, initiallyOn, state_trace);
 310   return cpu;
 311 }
 312
 313 Link* NetworkL07Model::createLink(const char *name,
 314                                  double bw_initial,
 315                                  tmgr_trace_t bw_trace,
 316                                  double lat_initial,
 317                                  tmgr_trace_t lat_trace,
 318                                  int initiallyOn,
 319                                  tmgr_trace_t state_trace,
 320                                  e_surf_link_sharing_policy_t policy,
 321                                  xbt_dict_t properties)
 322 {
 323   xbt_assert(!Link::byName(name),
 324            "Link '%s' declared several times in the platform file.", name);
 325
 326   Link* link = new LinkL07(this, name, properties,
 327                  bw_initial, bw_trace,
 328            lat_initial, lat_trace,
 329            initiallyOn, state_trace,
 330            policy);
 331   Link::onCreation(link);
 332   return link;
 333 }
 334
 335 void HostL07Model::addTraces()
 336 {
 337   xbt_dict_cursor_t cursor = NULL;
 338   char *trace_name, *elm;
 339
 340   if (!trace_connect_list_host_speed)
 341     return;
 342
 343   /* Connect traces relative to cpu */
 344   xbt_dict_foreach(trace_connect_list_host_speed, cursor, trace_name, elm) {
 345     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
 346     Cpu *cpu = sg_host_by_name(elm)->pimpl_cpu;
 347
 348     xbt_assert(cpu, "Host %s undefined", elm);
 349     xbt_assert(trace, "Trace %s undefined", trace_name);
 350
 351     cpu->set_speed_trace(trace);
 352   }
 353
 354   /* Connect traces relative to network */
 355   xbt_dict_foreach(trace_connect_list_link_avail, cursor, trace_name, elm) {
 356     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
 357     LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
 358
 359     xbt_assert(link, "Link %s undefined", elm);
 360     xbt_assert(trace, "Trace %s undefined", trace_name);
 361
 362     link->p_stateEvent = future_evt_set->add_trace(trace, 0.0, link);
 363   }
 364
 365   xbt_dict_foreach(trace_connect_list_link_bw, cursor, trace_name, elm) {
 366     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
 367     LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
 368
 369     xbt_assert(link, "Link %s undefined", elm);
 370     xbt_assert(trace, "Trace %s undefined", trace_name);
 371
 372     link->p_bwEvent = future_evt_set->add_trace(trace, 0.0, link);
 373   }
 374
 375   xbt_dict_foreach(trace_connect_list_link_lat, cursor, trace_name, elm) {
 376     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
 377     LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
 378
 379     xbt_assert(link, "Link %s undefined", elm);
 380     xbt_assert(trace, "Trace %s undefined", trace_name);
 381
 382     link->p_latEvent = future_evt_set->add_trace(trace, 0.0, link);
 383   }
 384 }
 385
 386 /************
 387  * Resource *
 388  ************/
 389
 390 CpuL07::CpuL07(CpuL07Model *model, simgrid::s4u::Host *host,
 391                xbt_dynar_t speedPeakList, int pstate,
 392          double speedScale, tmgr_trace_t speedTrace,
 393              int core, int initiallyOn, tmgr_trace_t state_trace)
 394  : Cpu(model, host, speedPeakList, pstate,
 395      core, xbt_dynar_get_as(speedPeakList,pstate,double), speedScale, initiallyOn)
 396 {
 397   p_constraint = lmm_constraint_new(model->getMaxminSystem(), this, xbt_dynar_get_as(speedPeakList,pstate,double) * speedScale);
 398
 399   if (speedTrace)
 400     p_speedEvent = future_evt_set->add_trace(speedTrace, 0.0, this);
 401
 402   if (state_trace)
 403     p_stateEvent = future_evt_set->add_trace(state_trace, 0.0, this);
 404 }
 405
 406 CpuL07::~CpuL07()
 407 {
 408 }
 409
 410 LinkL07::LinkL07(NetworkL07Model *model, const char* name, xbt_dict_t props,
 411              double bw_initial,
 412              tmgr_trace_t bw_trace,
 413              double lat_initial,
 414              tmgr_trace_t lat_trace,
 415              int initiallyOn,
 416              tmgr_trace_t state_trace,
 417              e_surf_link_sharing_policy_t policy)
 418  : Link(model, name, props, lmm_constraint_new(model->getMaxminSystem(), this, bw_initial), future_evt_set, state_trace)
 419 {
 420   m_bwCurrent = bw_initial;
 421   if (bw_trace)
 422     p_bwEvent = future_evt_set->add_trace(bw_trace, 0.0, this);
 423
 424   if (initiallyOn)
 425     turnOn();
 426   else
 427     turnOff();
 428   m_latCurrent = lat_initial;
 429
 430   if (lat_trace)
 431     p_latEvent = future_evt_set->add_trace(lat_trace, 0.0, this);
 432
 433   if (policy == SURF_LINK_FATPIPE)
 434   lmm_constraint_shared(getConstraint());
 435 }
 436
 437 Action *CpuL07::execution_start(double size)
 438 {
 439   sg_host_t*host_list = xbt_new0(sg_host_t, 1);
 440   double *flops_amount = xbt_new0(double, 1);
 441
 442   host_list[0] = getHost();
 443   flops_amount[0] = size;
 444
 445   return static_cast<CpuL07Model*>(getModel())->p_hostModel
 446     ->executeParallelTask( 1, host_list, flops_amount, NULL, -1);
 447 }
 448
 449 Action *CpuL07::sleep(double duration)
 450 {
 451   L07Action *action = static_cast<L07Action*>(execution_start(1.0));
 452   action->m_maxDuration = duration;
 453   action->m_suspended = 2;
 454   lmm_update_variable_weight(getModel()->getMaxminSystem(), action->getVariable(), 0.0);
 455
 456   return action;
 457 }
 458
 459 bool CpuL07::isUsed(){
 460   return lmm_constraint_used(getModel()->getMaxminSystem(), getConstraint());
 461 }
 462
 463 /** @brief take into account changes of speed (either load or max) */
 464 void CpuL07::onSpeedChange() {
 465   lmm_variable_t var = NULL;
 466   lmm_element_t elem = NULL;
 467
 468     lmm_update_constraint_bound(getModel()->getMaxminSystem(), getConstraint(), m_speedPeak * m_speedScale);
 469     while ((var = lmm_get_var_from_cnst
 470             (getModel()->getMaxminSystem(), getConstraint(), &elem))) {
 471       Action *action = static_cast<Action*>(lmm_variable_id(var));
 472
 473       lmm_update_variable_bound(getModel()->getMaxminSystem(),
 474                                 action->getVariable(),
 475                                 m_speedScale * m_speedPeak);
 476     }
 477
 478   Cpu::onSpeedChange();
 479 }
 480
 481
 482 bool LinkL07::isUsed(){
 483   return lmm_constraint_used(getModel()->getMaxminSystem(), getConstraint());
 484 }
 485
 486 void CpuL07::updateState(tmgr_trace_iterator_t triggered, double value, double /*date*/){
 487   XBT_DEBUG("Updating cpu %s (%p) with value %g", getName(), this, value);
 488   if (triggered == p_speedEvent) {
 489     m_speedScale = value;
 490     onSpeedChange();
 491     tmgr_trace_event_unref(&p_speedEvent);
 492   } else if (triggered == p_stateEvent) {
 493     if (value > 0)
 494       turnOn();
 495     else
 496       turnOff();
 497     tmgr_trace_event_unref(&p_stateEvent);
 498   } else {
 499     xbt_die("Unknown event!\n");
 500   }
 501 }
 502
 503 void LinkL07::updateState(tmgr_trace_iterator_t triggered, double value, double date) {
 504   XBT_DEBUG("Updating link %s (%p) with value=%f for date=%g", getName(), this, value, date);
 505   if (triggered == p_bwEvent) {
 506     updateBandwidth(value, date);
 507     tmgr_trace_event_unref(&p_bwEvent);
 508   } else if (triggered == p_latEvent) {
 509     updateLatency(value, date);
 510     tmgr_trace_event_unref(&p_latEvent);
 511   } else if (triggered == p_stateEvent) {
 512     if (value > 0)
 513       turnOn();
 514     else
 515       turnOff();
 516     tmgr_trace_event_unref(&p_stateEvent);
 517   } else {
 518     xbt_die("Unknown event ! \n");
 519   }
 520 }
 521
 522 double LinkL07::getBandwidth()
 523 {
 524   return m_bwCurrent;
 525 }
 526
 527 void LinkL07::updateBandwidth(double value, double date)
 528 {
 529   m_bwCurrent = value;
 530   lmm_update_constraint_bound(getModel()->getMaxminSystem(), getConstraint(), m_bwCurrent);
 531 }
 532
 533 void LinkL07::updateLatency(double value, double date)
 534 {
 535   lmm_variable_t var = NULL;
 536   L07Action *action;
 537   lmm_element_t elem = NULL;
 538
 539   m_latCurrent = value;
 540   while ((var = lmm_get_var_from_cnst(getModel()->getMaxminSystem(), getConstraint(), &elem))) {
 541     action = static_cast<L07Action*>(lmm_variable_id(var));
 542     action->updateBound();
 543   }
 544 }
 545
 546 /**********
 547  * Action *
 548  **********/
 549
 550 L07Action::~L07Action(){
 551   delete p_netcardList;
 552   free(p_communicationAmount);
 553   free(p_computationAmount);
 554 }
 555
 556 void L07Action::updateBound()
 557 {
 558   double lat_current = 0.0;
 559   double lat_bound = -1.0;
 560   int i, j;
 561
 562   int hostNb = p_netcardList->size();
 563
 564   if (p_communicationAmount != NULL) {
 565     for (i = 0; i < hostNb; i++) {
 566       for (j = 0; j < hostNb; j++) {
 567         xbt_dynar_t route=NULL;
 568
 569         if (p_communicationAmount[i * hostNb + j] > 0) {
 570           double lat = 0.0;
 571           routing_platf->getRouteAndLatency((*p_netcardList)[i], (*p_netcardList)[j],
 572                                                                 &route, &lat);
 573
 574           lat_current = MAX(lat_current, lat * p_communicationAmount[i * hostNb + j]);
 575         }
 576       }
 577     }
 578   }
 579   lat_bound = sg_tcp_gamma / (2.0 * lat_current);
 580   XBT_DEBUG("action (%p) : lat_bound = %g", this, lat_bound);
 581   if ((m_latency == 0.0) && (m_suspended == 0)) {
 582     if (m_rate < 0)
 583       lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(), lat_bound);
 584     else
 585       lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(),
 586         std::min(m_rate, lat_bound));
 587   }
 588 }
 589
 590 int L07Action::unref()
 591 {
 592   m_refcount--;
 593   if (!m_refcount) {
 594     if (action_hook.is_linked())
 595       p_stateSet->erase(p_stateSet->iterator_to(*this));
 596     if (getVariable())
 597       lmm_variable_free(getModel()->getMaxminSystem(), getVariable());
 598     delete this;
 599     return 1;
 600   }
 601   return 0;
 602 }
 603
 604 }
 605 }