Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
7e467446de3f139c0a334d322740740dea1c5aec
[simgrid.git] / 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   xbt_dict_t ptask_parallel_task_link_set = xbt_dict_new_homogeneous(NULL);
206
207   this->p_netcardList->reserve(host_nb);
208   for (int i = 0; i<host_nb; i++)
209           this->p_netcardList->push_back(host_list[i]->pimpl_netcard);
210
211   /* Compute the number of affected resources... */
212   if(bytes_amount != NULL) {
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
236   nb_link = xbt_dict_length(ptask_parallel_task_link_set);
237   xbt_dict_free(&ptask_parallel_task_link_set);
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 cpus and %d links.",
244          this, host_nb, nb_link);
245   this->p_computationAmount = flops_amount;
246   this->p_communicationAmount = bytes_amount;
247   this->m_latency = latency;
248   this->m_rate = rate;
249
250   this->p_variable = lmm_variable_new(model->getMaxminSystem(), this, 1.0,
251                                         (rate > 0 ? rate : -1.0),
252                                                                                 host_nb + nb_link);
253
254   if (this->m_latency > 0)
255     lmm_update_variable_weight(model->getMaxminSystem(), this->getVariable(), 0.0);
256
257   for (int i = 0; i < host_nb; i++)
258     lmm_expand(model->getMaxminSystem(),
259                host_list[i]->pimpl_cpu->getConstraint(),
260                this->getVariable(), flops_amount[i]);
261
262   if(bytes_amount != NULL) {
263     for (int i = 0; i < host_nb; i++) {
264       for (int j = 0; j < host_nb; j++) {
265         void *_link;
266
267         xbt_dynar_t route=NULL;
268         if (bytes_amount[i * host_nb + j] == 0.0)
269           continue;
270
271         routing_platf->getRouteAndLatency((*this->p_netcardList)[i], (*this->p_netcardList)[j],
272                                                     &route, NULL);
273
274         xbt_dynar_foreach(route, cpt, _link) {
275           LinkL07 *link = static_cast<LinkL07*>(_link);
276           lmm_expand_add(model->getMaxminSystem(), link->getConstraint(),
277                         this->getVariable(), bytes_amount[i * host_nb + j]);
278         }
279       }
280     }
281   }
282
283   if (nb_link + nb_used_host == 0) {
284     this->setCost(1.0);
285     this->setRemains(0.0);
286   }
287 }
288
289 Action *NetworkL07Model::communicate(NetCard *src, NetCard *dst,
290                                        double size, double rate)
291 {
292   sg_host_t*host_list = xbt_new0(sg_host_t, 2);
293   double *flops_amount = xbt_new0(double, 2);
294   double *bytes_amount = xbt_new0(double, 4);
295   Action *res = NULL;
296
297   host_list[0] = sg_host_by_name(src->getName());
298   host_list[1] = sg_host_by_name(dst->getName());
299   bytes_amount[1] = size;
300
301   res = p_hostModel->executeParallelTask(2, host_list,
302                                     flops_amount,
303                                     bytes_amount, rate);
304
305   return res;
306 }
307
308 Cpu *CpuL07Model::createCpu(simgrid::s4u::Host *host,  xbt_dynar_t powerPeakList,
309                           int pstate, double power_scale,
310                           tmgr_trace_t power_trace, int core,
311                           int initiallyOn,
312                           tmgr_trace_t state_trace)
313 {
314   CpuL07 *cpu = new CpuL07(this, host, powerPeakList, pstate, power_scale, power_trace,
315                          core, initiallyOn, state_trace);
316   return cpu;
317 }
318
319 Link* NetworkL07Model::createLink(const char *name,
320                                  double bw_initial,
321                                  tmgr_trace_t bw_trace,
322                                  double lat_initial,
323                                  tmgr_trace_t lat_trace,
324                                  int initiallyOn,
325                                  tmgr_trace_t state_trace,
326                                  e_surf_link_sharing_policy_t policy,
327                                  xbt_dict_t properties)
328 {
329   xbt_assert(!Link::byName(name),
330                  "Link '%s' declared several times in the platform file.", name);
331
332   Link* link = new LinkL07(this, name, properties,
333                              bw_initial, bw_trace,
334                                          lat_initial, lat_trace,
335                                          initiallyOn, state_trace,
336                                          policy);
337   Link::onCreation(link);
338   return link;
339 }
340
341 void HostL07Model::addTraces()
342 {
343   xbt_dict_cursor_t cursor = NULL;
344   char *trace_name, *elm;
345
346   if (!trace_connect_list_host_avail)
347     return;
348
349   /* Connect traces relative to cpu */
350   xbt_dict_foreach(trace_connect_list_host_avail, cursor, trace_name, elm) {
351     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
352     CpuL07 *host = static_cast<CpuL07*>(sg_host_by_name(elm)->pimpl_cpu);
353
354     xbt_assert(host, "Host %s undefined", elm);
355     xbt_assert(trace, "Trace %s undefined", trace_name);
356
357     host->p_stateEvent = future_evt_set->add_trace(trace, 0.0, host);
358   }
359
360   xbt_dict_foreach(trace_connect_list_power, cursor, trace_name, elm) {
361     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
362     CpuL07 *host = static_cast<CpuL07*>(sg_host_by_name(elm)->pimpl_cpu);
363
364     xbt_assert(host, "Host %s undefined", elm);
365     xbt_assert(trace, "Trace %s undefined", trace_name);
366
367     host->p_speedEvent = future_evt_set->add_trace(trace, 0.0, host);
368   }
369
370   /* Connect traces relative to network */
371   xbt_dict_foreach(trace_connect_list_link_avail, cursor, trace_name, elm) {
372     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
373     LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
374
375     xbt_assert(link, "Link %s undefined", elm);
376     xbt_assert(trace, "Trace %s undefined", trace_name);
377
378     link->p_stateEvent = future_evt_set->add_trace(trace, 0.0, link);
379   }
380
381   xbt_dict_foreach(trace_connect_list_bandwidth, cursor, trace_name, elm) {
382     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
383     LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
384
385     xbt_assert(link, "Link %s undefined", elm);
386     xbt_assert(trace, "Trace %s undefined", trace_name);
387
388     link->p_bwEvent = future_evt_set->add_trace(trace, 0.0, link);
389   }
390
391   xbt_dict_foreach(trace_connect_list_latency, cursor, trace_name, elm) {
392     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
393     LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
394
395     xbt_assert(link, "Link %s undefined", elm);
396     xbt_assert(trace, "Trace %s undefined", trace_name);
397
398     link->p_latEvent = future_evt_set->add_trace(trace, 0.0, link);
399   }
400 }
401
402 /************
403  * Resource *
404  ************/
405
406 CpuL07::CpuL07(CpuL07Model *model, simgrid::s4u::Host *host,
407                      xbt_dynar_t speedPeakList, int pstate,
408                                  double speedScale, tmgr_trace_t speedTrace,
409                          int core, int initiallyOn, tmgr_trace_t state_trace)
410  : Cpu(model, host, speedPeakList, pstate,
411            core, xbt_dynar_get_as(speedPeakList,pstate,double), speedScale, initiallyOn)
412 {
413   p_constraint = lmm_constraint_new(model->getMaxminSystem(), this, xbt_dynar_get_as(speedPeakList,pstate,double) * speedScale);
414
415   if (speedTrace)
416     p_speedEvent = future_evt_set->add_trace(speedTrace, 0.0, this);
417
418   if (state_trace)
419     p_stateEvent = future_evt_set->add_trace(state_trace, 0.0, this);
420 }
421
422 CpuL07::~CpuL07()
423 {
424 }
425
426 LinkL07::LinkL07(NetworkL07Model *model, const char* name, xbt_dict_t props,
427                          double bw_initial,
428                          tmgr_trace_t bw_trace,
429                          double lat_initial,
430                          tmgr_trace_t lat_trace,
431                          int initiallyOn,
432                          tmgr_trace_t state_trace,
433                          e_surf_link_sharing_policy_t policy)
434  : Link(model, name, props, lmm_constraint_new(model->getMaxminSystem(), this, bw_initial), future_evt_set, state_trace)
435 {
436   m_bwCurrent = bw_initial;
437   if (bw_trace)
438     p_bwEvent = future_evt_set->add_trace(bw_trace, 0.0, this);
439
440   if (initiallyOn)
441     turnOn();
442   else
443     turnOff();
444   m_latCurrent = lat_initial;
445
446   if (lat_trace)
447     p_latEvent = future_evt_set->add_trace(lat_trace, 0.0, this);
448
449   if (policy == SURF_LINK_FATPIPE)
450         lmm_constraint_shared(getConstraint());
451 }
452
453 Action *CpuL07::execution_start(double size)
454 {
455   sg_host_t*host_list = xbt_new0(sg_host_t, 1);
456   double *flops_amount = xbt_new0(double, 1);
457   double *bytes_amount = xbt_new0(double, 1);
458
459   host_list[0] = getHost();
460   flops_amount[0] = size;
461
462   return static_cast<CpuL07Model*>(getModel())->p_hostModel
463     ->executeParallelTask( 1, host_list, flops_amount, bytes_amount, -1);
464 }
465
466 Action *CpuL07::sleep(double duration)
467 {
468   L07Action *action = static_cast<L07Action*>(execution_start(1.0));
469   action->m_maxDuration = duration;
470   action->m_suspended = 2;
471   lmm_update_variable_weight(getModel()->getMaxminSystem(), action->getVariable(), 0.0);
472
473   return action;
474 }
475
476 bool CpuL07::isUsed(){
477   return lmm_constraint_used(getModel()->getMaxminSystem(), getConstraint());
478 }
479
480 /** @brief take into account changes of speed (either load or max) */
481 void CpuL07::onSpeedChange() {
482         lmm_variable_t var = NULL;
483         lmm_element_t elem = NULL;
484
485     lmm_update_constraint_bound(getModel()->getMaxminSystem(), getConstraint(), m_speedPeak * m_speedScale);
486     while ((var = lmm_get_var_from_cnst
487             (getModel()->getMaxminSystem(), getConstraint(), &elem))) {
488       Action *action = static_cast<Action*>(lmm_variable_id(var));
489
490       lmm_update_variable_bound(getModel()->getMaxminSystem(),
491                                 action->getVariable(),
492                                 m_speedScale * m_speedPeak);
493     }
494
495         Cpu::onSpeedChange();
496 }
497
498
499 bool LinkL07::isUsed(){
500   return lmm_constraint_used(getModel()->getMaxminSystem(), getConstraint());
501 }
502
503 void CpuL07::updateState(tmgr_trace_iterator_t triggered, double value, double /*date*/){
504   XBT_DEBUG("Updating cpu %s (%p) with value %g", getName(), this, value);
505   if (triggered == p_speedEvent) {
506     m_speedScale = value;
507     onSpeedChange();
508     tmgr_trace_event_unref(&p_speedEvent);
509   } else if (triggered == p_stateEvent) {
510     if (value > 0)
511       turnOn();
512     else
513       turnOff();
514     tmgr_trace_event_unref(&p_stateEvent);
515   } else {
516     xbt_die("Unknown event!\n");
517   }
518 }
519
520 void LinkL07::updateState(tmgr_trace_iterator_t triggered, double value, double date) {
521   XBT_DEBUG("Updating link %s (%p) with value=%f for date=%g", getName(), this, value, date);
522   if (triggered == p_bwEvent) {
523     updateBandwidth(value, date);
524     tmgr_trace_event_unref(&p_bwEvent);
525   } else if (triggered == p_latEvent) {
526     updateLatency(value, date);
527     tmgr_trace_event_unref(&p_latEvent);
528   } else if (triggered == p_stateEvent) {
529     if (value > 0)
530       turnOn();
531     else
532       turnOff();
533     tmgr_trace_event_unref(&p_stateEvent);
534   } else {
535     xbt_die("Unknown event ! \n");
536   }
537 }
538
539 double LinkL07::getBandwidth()
540 {
541   return m_bwCurrent;
542 }
543
544 void LinkL07::updateBandwidth(double value, double date)
545 {
546   m_bwCurrent = value;
547   lmm_update_constraint_bound(getModel()->getMaxminSystem(), getConstraint(), m_bwCurrent);
548 }
549
550 void LinkL07::updateLatency(double value, double date)
551 {
552   lmm_variable_t var = NULL;
553   L07Action *action;
554   lmm_element_t elem = NULL;
555
556   m_latCurrent = value;
557   while ((var = lmm_get_var_from_cnst(getModel()->getMaxminSystem(), getConstraint(), &elem))) {
558     action = static_cast<L07Action*>(lmm_variable_id(var));
559     action->updateBound();
560   }
561 }
562
563 /**********
564  * Action *
565  **********/
566
567 L07Action::~L07Action(){
568   free(p_communicationAmount);
569   free(p_computationAmount);
570 }
571
572 void L07Action::updateBound()
573 {
574   double lat_current = 0.0;
575   double lat_bound = -1.0;
576   int i, j;
577
578   int hostNb = p_netcardList->size();
579
580   if (p_communicationAmount != NULL) {
581     for (i = 0; i < hostNb; i++) {
582       for (j = 0; j < hostNb; j++) {
583         xbt_dynar_t route=NULL;
584
585         if (p_communicationAmount[i * hostNb + j] > 0) {
586           double lat = 0.0;
587           routing_platf->getRouteAndLatency((*p_netcardList)[i], (*p_netcardList)[j],
588                                                                 &route, &lat);
589
590           lat_current = MAX(lat_current, lat * p_communicationAmount[i * hostNb + j]);
591         }
592       }
593     }
594   }
595   lat_bound = sg_tcp_gamma / (2.0 * lat_current);
596   XBT_DEBUG("action (%p) : lat_bound = %g", this, lat_bound);
597   if ((m_latency == 0.0) && (m_suspended == 0)) {
598     if (m_rate < 0)
599       lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(), lat_bound);
600     else
601       lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(),
602         std::min(m_rate, lat_bound));
603   }
604 }
605
606 int L07Action::unref()
607 {
608   m_refcount--;
609   if (!m_refcount) {
610     if (action_hook.is_linked())
611           p_stateSet->erase(p_stateSet->iterator_to(*this));
612     if (getVariable())
613       lmm_variable_free(getModel()->getMaxminSystem(), getVariable());
614     delete this;
615     return 1;
616   }
617   return 0;
618 }
619
620 void L07Action::suspend()
621 {
622   XBT_IN("(%p))", this);
623   if (m_suspended != 2) {
624     m_suspended = 1;
625     lmm_update_variable_weight(getModel()->getMaxminSystem(), getVariable(), 0.0);
626   }
627   XBT_OUT();
628 }
629
630 void L07Action::resume()
631 {
632   XBT_IN("(%p)", this);
633   if (m_suspended != 2) {
634     lmm_update_variable_weight(getModel()->getMaxminSystem(), getVariable(), 1.0);
635     m_suspended = 0;
636   }
637   XBT_OUT();
638 }
639
640 double L07Action::getRemains()
641 {
642   XBT_IN("(%p)", this);
643   XBT_OUT();
644   return m_remains;
645 }
646
647 }
648 }