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         xbt_dynar_t route=NULL;
 216
 217         if (bytes_amount[i * host_nb + j] > 0) {
 218           double lat=0.0;
 219           unsigned int cpt;
 220           void *_link;
 221           LinkL07 *link;
 222
 223           routing_platf->getRouteAndLatency((*this->p_netcardList)[i], (*this->p_netcardList)[j],
 224                                                         &route, &lat);
 225           latency = MAX(latency, lat);
 226
 227           xbt_dynar_foreach(route, cpt, _link) {
 228             link = static_cast<LinkL07*>(_link);
 229             xbt_dict_set(ptask_parallel_task_link_set, link->getName(), link, NULL);
 230           }
 231         }
 232       }
 233     }
 234
 235     nb_link = xbt_dict_length(ptask_parallel_task_link_set);
 236     xbt_dict_free(&ptask_parallel_task_link_set);
 237   }
 238
 239   for (int i = 0; i < host_nb; i++)
 240     if (flops_amount[i] > 0)
 241       nb_used_host++;
 242
 243   XBT_DEBUG("Creating a parallel task (%p) with %d hosts and %d unique links.", this, host_nb, nb_link);
 244   this->p_computationAmount = flops_amount;
 245   this->p_communicationAmount = bytes_amount;
 246   this->m_latency = latency;
 247   this->m_rate = rate;
 248
 249   this->p_variable = lmm_variable_new(model->getMaxminSystem(), this, 1.0,
 250       (rate > 0 ? rate : -1.0),
 251       host_nb + nb_link);
 252
 253   if (this->m_latency > 0)
 254     lmm_update_variable_weight(model->getMaxminSystem(), this->getVariable(), 0.0);
 255
 256   for (int i = 0; i < host_nb; i++)
 257     lmm_expand(model->getMaxminSystem(), host_list[i]->pimpl_cpu->getConstraint(),
 258         this->getVariable(), flops_amount[i]);
 259
 260   if(bytes_amount != NULL) {
 261     for (int i = 0; i < host_nb; i++) {
 262       for (int j = 0; j < host_nb; j++) {
 263         void *_link;
 264
 265         xbt_dynar_t route=NULL;
 266         if (bytes_amount[i * host_nb + j] == 0.0)
 267           continue;
 268
 269         routing_platf->getRouteAndLatency((*this->p_netcardList)[i], (*this->p_netcardList)[j],
 270                                                     &route, NULL);
 271
 272         xbt_dynar_foreach(route, cpt, _link) {
 273           LinkL07 *link = static_cast<LinkL07*>(_link);
 274           lmm_expand_add(model->getMaxminSystem(), link->getConstraint(),
 275                         this->getVariable(), bytes_amount[i * host_nb + j]);
 276         }
 277       }
 278     }
 279   }
 280
 281   if (nb_link + nb_used_host == 0) {
 282     this->setCost(1.0);
 283     this->setRemains(0.0);
 284   }
 285   xbt_free(host_list);
 286 }
 287
 288 Action *NetworkL07Model::communicate(NetCard *src, NetCard *dst,
 289                                        double size, double rate)
 290 {
 291   sg_host_t*host_list = xbt_new0(sg_host_t, 2);
 292   double *flops_amount = xbt_new0(double, 2);
 293   double *bytes_amount = xbt_new0(double, 4);
 294   Action *res = NULL;
 295
 296   host_list[0] = sg_host_by_name(src->getName());
 297   host_list[1] = sg_host_by_name(dst->getName());
 298   bytes_amount[1] = size;
 299
 300   res = p_hostModel->executeParallelTask(2, host_list, flops_amount, bytes_amount, rate);
 301
 302   return res;
 303 }
 304
 305 Cpu *CpuL07Model::createCpu(simgrid::s4u::Host *host,  xbt_dynar_t powerPeakList,
 306                           int pstate, double power_scale,
 307                           tmgr_trace_t power_trace, int core,
 308                           int initiallyOn,
 309                           tmgr_trace_t state_trace)
 310 {
 311   CpuL07 *cpu = new CpuL07(this, host, powerPeakList, pstate, power_scale, power_trace,
 312                          core, initiallyOn, state_trace);
 313   return cpu;
 314 }
 315
 316 Link* NetworkL07Model::createLink(const char *name,
 317                                  double bw_initial,
 318                                  tmgr_trace_t bw_trace,
 319                                  double lat_initial,
 320                                  tmgr_trace_t lat_trace,
 321                                  int initiallyOn,
 322                                  tmgr_trace_t state_trace,
 323                                  e_surf_link_sharing_policy_t policy,
 324                                  xbt_dict_t properties)
 325 {
 326   xbt_assert(!Link::byName(name),
 327                  "Link '%s' declared several times in the platform file.", name);
 328
 329   Link* link = new LinkL07(this, name, properties,
 330                              bw_initial, bw_trace,
 331                                          lat_initial, lat_trace,
 332                                          initiallyOn, state_trace,
 333                                          policy);
 334   Link::onCreation(link);
 335   return link;
 336 }
 337
 338 void HostL07Model::addTraces()
 339 {
 340   xbt_dict_cursor_t cursor = NULL;
 341   char *trace_name, *elm;
 342
 343   if (!trace_connect_list_host_speed)
 344     return;
 345
 346   /* Connect traces relative to cpu */
 347   xbt_dict_foreach(trace_connect_list_host_speed, cursor, trace_name, elm) {
 348     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
 349     Cpu *cpu = sg_host_by_name(elm)->pimpl_cpu;
 350
 351     xbt_assert(cpu, "Host %s undefined", elm);
 352     xbt_assert(trace, "Trace %s undefined", trace_name);
 353
 354     cpu->set_speed_trace(trace);
 355   }
 356
 357   /* Connect traces relative to network */
 358   xbt_dict_foreach(trace_connect_list_link_avail, cursor, trace_name, elm) {
 359     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
 360     LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
 361
 362     xbt_assert(link, "Link %s undefined", elm);
 363     xbt_assert(trace, "Trace %s undefined", trace_name);
 364
 365     link->p_stateEvent = future_evt_set->add_trace(trace, 0.0, link);
 366   }
 367
 368   xbt_dict_foreach(trace_connect_list_link_bw, cursor, trace_name, elm) {
 369     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
 370     LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
 371
 372     xbt_assert(link, "Link %s undefined", elm);
 373     xbt_assert(trace, "Trace %s undefined", trace_name);
 374
 375     link->p_bwEvent = future_evt_set->add_trace(trace, 0.0, link);
 376   }
 377
 378   xbt_dict_foreach(trace_connect_list_link_lat, cursor, trace_name, elm) {
 379     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
 380     LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
 381
 382     xbt_assert(link, "Link %s undefined", elm);
 383     xbt_assert(trace, "Trace %s undefined", trace_name);
 384
 385     link->p_latEvent = future_evt_set->add_trace(trace, 0.0, link);
 386   }
 387 }
 388
 389 /************
 390  * Resource *
 391  ************/
 392
 393 CpuL07::CpuL07(CpuL07Model *model, simgrid::s4u::Host *host,
 394                      xbt_dynar_t speedPeakList, int pstate,
 395                                  double speedScale, tmgr_trace_t speedTrace,
 396                          int core, int initiallyOn, tmgr_trace_t state_trace)
 397  : Cpu(model, host, speedPeakList, pstate,
 398            core, xbt_dynar_get_as(speedPeakList,pstate,double), speedScale, initiallyOn)
 399 {
 400   p_constraint = lmm_constraint_new(model->getMaxminSystem(), this, xbt_dynar_get_as(speedPeakList,pstate,double) * speedScale);
 401
 402   if (speedTrace)
 403     p_speedEvent = future_evt_set->add_trace(speedTrace, 0.0, this);
 404
 405   if (state_trace)
 406     p_stateEvent = future_evt_set->add_trace(state_trace, 0.0, this);
 407 }
 408
 409 CpuL07::~CpuL07()
 410 {
 411 }
 412
 413 LinkL07::LinkL07(NetworkL07Model *model, const char* name, xbt_dict_t props,
 414                          double bw_initial,
 415                          tmgr_trace_t bw_trace,
 416                          double lat_initial,
 417                          tmgr_trace_t lat_trace,
 418                          int initiallyOn,
 419                          tmgr_trace_t state_trace,
 420                          e_surf_link_sharing_policy_t policy)
 421  : Link(model, name, props, lmm_constraint_new(model->getMaxminSystem(), this, bw_initial), future_evt_set, state_trace)
 422 {
 423   m_bwCurrent = bw_initial;
 424   if (bw_trace)
 425     p_bwEvent = future_evt_set->add_trace(bw_trace, 0.0, this);
 426
 427   if (initiallyOn)
 428     turnOn();
 429   else
 430     turnOff();
 431   m_latCurrent = lat_initial;
 432
 433   if (lat_trace)
 434     p_latEvent = future_evt_set->add_trace(lat_trace, 0.0, this);
 435
 436   if (policy == SURF_LINK_FATPIPE)
 437         lmm_constraint_shared(getConstraint());
 438 }
 439
 440 Action *CpuL07::execution_start(double size)
 441 {
 442   sg_host_t*host_list = xbt_new0(sg_host_t, 1);
 443   double *flops_amount = xbt_new0(double, 1);
 444
 445   host_list[0] = getHost();
 446   flops_amount[0] = size;
 447
 448   return static_cast<CpuL07Model*>(getModel())->p_hostModel
 449     ->executeParallelTask( 1, host_list, flops_amount, NULL, -1);
 450 }
 451
 452 Action *CpuL07::sleep(double duration)
 453 {
 454   L07Action *action = static_cast<L07Action*>(execution_start(1.0));
 455   action->m_maxDuration = duration;
 456   action->m_suspended = 2;
 457   lmm_update_variable_weight(getModel()->getMaxminSystem(), action->getVariable(), 0.0);
 458
 459   return action;
 460 }
 461
 462 bool CpuL07::isUsed(){
 463   return lmm_constraint_used(getModel()->getMaxminSystem(), getConstraint());
 464 }
 465
 466 /** @brief take into account changes of speed (either load or max) */
 467 void CpuL07::onSpeedChange() {
 468         lmm_variable_t var = NULL;
 469         lmm_element_t elem = NULL;
 470
 471     lmm_update_constraint_bound(getModel()->getMaxminSystem(), getConstraint(), m_speedPeak * m_speedScale);
 472     while ((var = lmm_get_var_from_cnst
 473             (getModel()->getMaxminSystem(), getConstraint(), &elem))) {
 474       Action *action = static_cast<Action*>(lmm_variable_id(var));
 475
 476       lmm_update_variable_bound(getModel()->getMaxminSystem(),
 477                                 action->getVariable(),
 478                                 m_speedScale * m_speedPeak);
 479     }
 480
 481         Cpu::onSpeedChange();
 482 }
 483
 484
 485 bool LinkL07::isUsed(){
 486   return lmm_constraint_used(getModel()->getMaxminSystem(), getConstraint());
 487 }
 488
 489 void CpuL07::updateState(tmgr_trace_iterator_t triggered, double value, double /*date*/){
 490   XBT_DEBUG("Updating cpu %s (%p) with value %g", getName(), this, value);
 491   if (triggered == p_speedEvent) {
 492     m_speedScale = value;
 493     onSpeedChange();
 494     tmgr_trace_event_unref(&p_speedEvent);
 495   } else if (triggered == p_stateEvent) {
 496     if (value > 0)
 497       turnOn();
 498     else
 499       turnOff();
 500     tmgr_trace_event_unref(&p_stateEvent);
 501   } else {
 502     xbt_die("Unknown event!\n");
 503   }
 504 }
 505
 506 void LinkL07::updateState(tmgr_trace_iterator_t triggered, double value, double date) {
 507   XBT_DEBUG("Updating link %s (%p) with value=%f for date=%g", getName(), this, value, date);
 508   if (triggered == p_bwEvent) {
 509     updateBandwidth(value, date);
 510     tmgr_trace_event_unref(&p_bwEvent);
 511   } else if (triggered == p_latEvent) {
 512     updateLatency(value, date);
 513     tmgr_trace_event_unref(&p_latEvent);
 514   } else if (triggered == p_stateEvent) {
 515     if (value > 0)
 516       turnOn();
 517     else
 518       turnOff();
 519     tmgr_trace_event_unref(&p_stateEvent);
 520   } else {
 521     xbt_die("Unknown event ! \n");
 522   }
 523 }
 524
 525 double LinkL07::getBandwidth()
 526 {
 527   return m_bwCurrent;
 528 }
 529
 530 void LinkL07::updateBandwidth(double value, double date)
 531 {
 532   m_bwCurrent = value;
 533   lmm_update_constraint_bound(getModel()->getMaxminSystem(), getConstraint(), m_bwCurrent);
 534 }
 535
 536 void LinkL07::updateLatency(double value, double date)
 537 {
 538   lmm_variable_t var = NULL;
 539   L07Action *action;
 540   lmm_element_t elem = NULL;
 541
 542   m_latCurrent = value;
 543   while ((var = lmm_get_var_from_cnst(getModel()->getMaxminSystem(), getConstraint(), &elem))) {
 544     action = static_cast<L07Action*>(lmm_variable_id(var));
 545     action->updateBound();
 546   }
 547 }
 548
 549 /**********
 550  * Action *
 551  **********/
 552
 553 L07Action::~L07Action(){
 554   delete p_netcardList;
 555   free(p_communicationAmount);
 556   free(p_computationAmount);
 557 }
 558
 559 void L07Action::updateBound()
 560 {
 561   double lat_current = 0.0;
 562   double lat_bound = -1.0;
 563   int i, j;
 564
 565   int hostNb = p_netcardList->size();
 566
 567   if (p_communicationAmount != NULL) {
 568     for (i = 0; i < hostNb; i++) {
 569       for (j = 0; j < hostNb; j++) {
 570         xbt_dynar_t route=NULL;
 571
 572         if (p_communicationAmount[i * hostNb + j] > 0) {
 573           double lat = 0.0;
 574           routing_platf->getRouteAndLatency((*p_netcardList)[i], (*p_netcardList)[j],
 575                                                                 &route, &lat);
 576
 577           lat_current = MAX(lat_current, lat * p_communicationAmount[i * hostNb + j]);
 578         }
 579       }
 580     }
 581   }
 582   lat_bound = sg_tcp_gamma / (2.0 * lat_current);
 583   XBT_DEBUG("action (%p) : lat_bound = %g", this, lat_bound);
 584   if ((m_latency == 0.0) && (m_suspended == 0)) {
 585     if (m_rate < 0)
 586       lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(), lat_bound);
 587     else
 588       lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(),
 589         std::min(m_rate, lat_bound));
 590   }
 591 }
 592
 593 int L07Action::unref()
 594 {
 595   m_refcount--;
 596   if (!m_refcount) {
 597     if (action_hook.is_linked())
 598       p_stateSet->erase(p_stateSet->iterator_to(*this));
 599     if (getVariable())
 600       lmm_variable_free(getModel()->getMaxminSystem(), getVariable());
 601     delete this;
 602     return 1;
 603   }
 604   return 0;
 605 }
 606
 607 }
 608 }