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