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