Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
9ba3c567a3b04e4b0f5a3b66875238db56ed8b09
[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   for (int i = 0; i < host_nb; i++) {
213     for (int j = 0; j < host_nb; j++) {
214       xbt_dynar_t route=NULL;
215
216       if (bytes_amount[i * host_nb + j] > 0) {
217         double lat=0.0;
218         unsigned int cpt;
219         void *_link;
220         LinkL07 *link;
221
222         routing_platf->getRouteAndLatency((*this->p_netcardList)[i], (*this->p_netcardList)[j],
223                                                   &route, &lat);
224         latency = MAX(latency, lat);
225
226         xbt_dynar_foreach(route, cpt, _link) {
227            link = static_cast<LinkL07*>(_link);
228            xbt_dict_set(ptask_parallel_task_link_set, link->getName(), link, NULL);
229         }
230       }
231     }
232   }
233
234   nb_link = xbt_dict_length(ptask_parallel_task_link_set);
235   xbt_dict_free(&ptask_parallel_task_link_set);
236
237   for (int i = 0; i < host_nb; i++)
238     if (flops_amount[i] > 0)
239       nb_used_host++;
240
241   XBT_DEBUG("Creating a parallel task (%p) with %d cpus and %d links.",
242          this, host_nb, nb_link);
243   this->p_computationAmount = flops_amount;
244   this->p_communicationAmount = bytes_amount;
245   this->m_latency = latency;
246   this->m_rate = rate;
247
248   this->p_variable = lmm_variable_new(model->getMaxminSystem(), this, 1.0,
249                                         (rate > 0 ? rate : -1.0),
250                                                                                 host_nb + nb_link);
251
252   if (this->m_latency > 0)
253     lmm_update_variable_weight(model->getMaxminSystem(), this->getVariable(), 0.0);
254
255   for (int i = 0; i < host_nb; i++)
256     lmm_expand(model->getMaxminSystem(),
257                host_list[i]->pimpl_cpu->getConstraint(),
258                this->getVariable(), flops_amount[i]);
259
260   for (int i = 0; i < host_nb; i++) {
261     for (int j = 0; j < host_nb; j++) {
262       void *_link;
263
264       xbt_dynar_t route=NULL;
265       if (bytes_amount[i * host_nb + j] == 0.0)
266         continue;
267
268       routing_platf->getRouteAndLatency((*this->p_netcardList)[i], (*this->p_netcardList)[j],
269                                             &route, NULL);
270
271       xbt_dynar_foreach(route, cpt, _link) {
272         LinkL07 *link = static_cast<LinkL07*>(_link);
273         lmm_expand_add(model->getMaxminSystem(), link->getConstraint(),
274                        this->getVariable(),
275                        bytes_amount[i * host_nb + j]);
276       }
277     }
278   }
279
280   if (nb_link + nb_used_host == 0) {
281     this->setCost(1.0);
282     this->setRemains(0.0);
283   }
284 }
285
286 Action *NetworkL07Model::communicate(NetCard *src, NetCard *dst,
287                                        double size, double rate)
288 {
289   sg_host_t*host_list = xbt_new0(sg_host_t, 2);
290   double *flops_amount = xbt_new0(double, 2);
291   double *bytes_amount = xbt_new0(double, 4);
292   Action *res = NULL;
293
294   host_list[0] = sg_host_by_name(src->getName());
295   host_list[1] = sg_host_by_name(dst->getName());
296   bytes_amount[1] = size;
297
298   res = p_hostModel->executeParallelTask(2, host_list,
299                                     flops_amount,
300                                     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_avail)
344     return;
345
346   /* Connect traces relative to cpu */
347   xbt_dict_foreach(trace_connect_list_host_avail, cursor, trace_name, elm) {
348     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
349     CpuL07 *host = static_cast<CpuL07*>(sg_host_by_name(elm)->pimpl_cpu);
350
351     xbt_assert(host, "Host %s undefined", elm);
352     xbt_assert(trace, "Trace %s undefined", trace_name);
353
354     host->p_stateEvent = future_evt_set->add_trace(trace, 0.0, host);
355   }
356
357   xbt_dict_foreach(trace_connect_list_power, cursor, trace_name, elm) {
358     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
359     CpuL07 *host = static_cast<CpuL07*>(sg_host_by_name(elm)->pimpl_cpu);
360
361     xbt_assert(host, "Host %s undefined", elm);
362     xbt_assert(trace, "Trace %s undefined", trace_name);
363
364     host->p_speedEvent = future_evt_set->add_trace(trace, 0.0, host);
365   }
366
367   /* Connect traces relative to network */
368   xbt_dict_foreach(trace_connect_list_link_avail, 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_stateEvent = future_evt_set->add_trace(trace, 0.0, link);
376   }
377
378   xbt_dict_foreach(trace_connect_list_bandwidth, 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_bwEvent = future_evt_set->add_trace(trace, 0.0, link);
386   }
387
388   xbt_dict_foreach(trace_connect_list_latency, cursor, trace_name, elm) {
389     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
390     LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
391
392     xbt_assert(link, "Link %s undefined", elm);
393     xbt_assert(trace, "Trace %s undefined", trace_name);
394
395     link->p_latEvent = future_evt_set->add_trace(trace, 0.0, link);
396   }
397 }
398
399 /************
400  * Resource *
401  ************/
402
403 CpuL07::CpuL07(CpuL07Model *model, simgrid::s4u::Host *host,
404                      xbt_dynar_t speedPeakList, int pstate,
405                                  double speedScale, tmgr_trace_t speedTrace,
406                          int core, int initiallyOn, tmgr_trace_t state_trace)
407  : Cpu(model, host, speedPeakList, pstate,
408            core, xbt_dynar_get_as(speedPeakList,pstate,double), speedScale, initiallyOn)
409 {
410   p_constraint = lmm_constraint_new(model->getMaxminSystem(), this, xbt_dynar_get_as(speedPeakList,pstate,double) * speedScale);
411
412   if (speedTrace)
413     p_speedEvent = future_evt_set->add_trace(speedTrace, 0.0, this);
414
415   if (state_trace)
416     p_stateEvent = future_evt_set->add_trace(state_trace, 0.0, this);
417 }
418
419 CpuL07::~CpuL07()
420 {
421 }
422
423 LinkL07::LinkL07(NetworkL07Model *model, const char* name, xbt_dict_t props,
424                          double bw_initial,
425                          tmgr_trace_t bw_trace,
426                          double lat_initial,
427                          tmgr_trace_t lat_trace,
428                          int initiallyOn,
429                          tmgr_trace_t state_trace,
430                          e_surf_link_sharing_policy_t policy)
431  : Link(model, name, props, lmm_constraint_new(model->getMaxminSystem(), this, bw_initial), future_evt_set, state_trace)
432 {
433   m_bwCurrent = bw_initial;
434   if (bw_trace)
435     p_bwEvent = future_evt_set->add_trace(bw_trace, 0.0, this);
436
437   if (initiallyOn)
438     turnOn();
439   else
440     turnOff();
441   m_latCurrent = lat_initial;
442
443   if (lat_trace)
444     p_latEvent = future_evt_set->add_trace(lat_trace, 0.0, this);
445
446   if (policy == SURF_LINK_FATPIPE)
447         lmm_constraint_shared(getConstraint());
448 }
449
450 Action *CpuL07::execute(double size)
451 {
452   sg_host_t*host_list = xbt_new0(sg_host_t, 1);
453   double *flops_amount = xbt_new0(double, 1);
454   double *bytes_amount = xbt_new0(double, 1);
455
456   host_list[0] = getHost();
457   flops_amount[0] = size;
458
459   return static_cast<CpuL07Model*>(getModel())
460     ->p_hostModel
461     ->executeParallelTask( 1, host_list, flops_amount, bytes_amount, -1);
462 }
463
464 Action *CpuL07::sleep(double duration)
465 {
466   L07Action *action = NULL;
467
468   XBT_IN("(%s,%g)", getName(), duration);
469
470   action = static_cast<L07Action*>(execute(1.0));
471   action->m_maxDuration = duration;
472   action->m_suspended = 2;
473   lmm_update_variable_weight(getModel()->getMaxminSystem(), action->getVariable(), 0.0);
474
475   XBT_OUT();
476   return action;
477 }
478
479 bool CpuL07::isUsed(){
480   return lmm_constraint_used(getModel()->getMaxminSystem(), getConstraint());
481 }
482
483 /** @brief take into account changes of speed (either load or max) */
484 void CpuL07::onSpeedChange() {
485         lmm_variable_t var = NULL;
486         lmm_element_t elem = NULL;
487
488     lmm_update_constraint_bound(getModel()->getMaxminSystem(), getConstraint(), m_speedPeak * m_speedScale);
489     while ((var = lmm_get_var_from_cnst
490             (getModel()->getMaxminSystem(), getConstraint(), &elem))) {
491       Action *action = static_cast<Action*>(lmm_variable_id(var));
492
493       lmm_update_variable_bound(getModel()->getMaxminSystem(),
494                                 action->getVariable(),
495                                 m_speedScale * m_speedPeak);
496     }
497
498         Cpu::onSpeedChange();
499 }
500
501
502 bool LinkL07::isUsed(){
503   return lmm_constraint_used(getModel()->getMaxminSystem(), getConstraint());
504 }
505
506 void CpuL07::updateState(tmgr_trace_iterator_t triggered, double value, double /*date*/){
507   XBT_DEBUG("Updating cpu %s (%p) with value %g", getName(), this, value);
508   if (triggered == p_speedEvent) {
509     m_speedScale = value;
510     onSpeedChange();
511     tmgr_trace_event_unref(&p_speedEvent);
512   } else if (triggered == p_stateEvent) {
513     if (value > 0)
514       turnOn();
515     else
516       turnOff();
517     tmgr_trace_event_unref(&p_stateEvent);
518   } else {
519     xbt_die("Unknown event!\n");
520   }
521 }
522
523 void LinkL07::updateState(tmgr_trace_iterator_t triggered, double value, double date) {
524   XBT_DEBUG("Updating link %s (%p) with value=%f for date=%g", getName(), this, value, date);
525   if (triggered == p_bwEvent) {
526     updateBandwidth(value, date);
527     tmgr_trace_event_unref(&p_bwEvent);
528   } else if (triggered == p_latEvent) {
529     updateLatency(value, date);
530     tmgr_trace_event_unref(&p_latEvent);
531   } else if (triggered == p_stateEvent) {
532     if (value > 0)
533       turnOn();
534     else
535       turnOff();
536     tmgr_trace_event_unref(&p_stateEvent);
537   } else {
538     xbt_die("Unknown event ! \n");
539   }
540 }
541
542 double LinkL07::getBandwidth()
543 {
544   return m_bwCurrent;
545 }
546
547 void LinkL07::updateBandwidth(double value, double date)
548 {
549   m_bwCurrent = value;
550   lmm_update_constraint_bound(getModel()->getMaxminSystem(), getConstraint(), m_bwCurrent);
551 }
552
553 void LinkL07::updateLatency(double value, double date)
554 {
555   lmm_variable_t var = NULL;
556   L07Action *action;
557   lmm_element_t elem = NULL;
558
559   m_latCurrent = value;
560   while ((var = lmm_get_var_from_cnst(getModel()->getMaxminSystem(), getConstraint(), &elem))) {
561     action = static_cast<L07Action*>(lmm_variable_id(var));
562     action->updateBound();
563   }
564 }
565
566 /**********
567  * Action *
568  **********/
569
570 L07Action::~L07Action(){
571   free(p_communicationAmount);
572   free(p_computationAmount);
573 }
574
575 void L07Action::updateBound()
576 {
577   double lat_current = 0.0;
578   double lat_bound = -1.0;
579   int i, j;
580
581   int hostNb = p_netcardList->size();
582
583   for (i = 0; i < hostNb; i++) {
584     for (j = 0; j < hostNb; j++) {
585       xbt_dynar_t route=NULL;
586
587       if (p_communicationAmount[i * hostNb + j] > 0) {
588         double lat = 0.0;
589         routing_platf->getRouteAndLatency((*p_netcardList)[i], (*p_netcardList)[j],
590                                                           &route, &lat);
591
592         lat_current = MAX(lat_current, lat * p_communicationAmount[i * hostNb + j]);
593       }
594     }
595   }
596   lat_bound = sg_tcp_gamma / (2.0 * lat_current);
597   XBT_DEBUG("action (%p) : lat_bound = %g", this, lat_bound);
598   if ((m_latency == 0.0) && (m_suspended == 0)) {
599     if (m_rate < 0)
600       lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(), lat_bound);
601     else
602       lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(),
603         std::min(m_rate, lat_bound));
604   }
605 }
606
607 int L07Action::unref()
608 {
609   m_refcount--;
610   if (!m_refcount) {
611     if (action_hook.is_linked())
612           p_stateSet->erase(p_stateSet->iterator_to(*this));
613     if (getVariable())
614       lmm_variable_free(getModel()->getMaxminSystem(), getVariable());
615     delete this;
616     return 1;
617   }
618   return 0;
619 }
620
621 void L07Action::suspend()
622 {
623   XBT_IN("(%p))", this);
624   if (m_suspended != 2) {
625     m_suspended = 1;
626     lmm_update_variable_weight(getModel()->getMaxminSystem(), getVariable(), 0.0);
627   }
628   XBT_OUT();
629 }
630
631 void L07Action::resume()
632 {
633   XBT_IN("(%p)", this);
634   if (m_suspended != 2) {
635     lmm_update_variable_weight(getModel()->getMaxminSystem(), getVariable(), 1.0);
636     m_suspended = 0;
637   }
638   XBT_OUT();
639 }
640
641 void L07Action::setMaxDuration(double duration)
642 {                               /* FIXME: should inherit */
643   XBT_IN("(%p,%g)", this, duration);
644   m_maxDuration = duration;
645   XBT_OUT();
646 }
647
648 void L07Action::setPriority(double priority)
649 {                               /* FIXME: should inherit */
650   XBT_IN("(%p,%g)", this, priority);
651   m_priority = priority;
652   XBT_OUT();
653 }
654
655 double L07Action::getRemains()
656 {
657   XBT_IN("(%p)", this);
658   XBT_OUT();
659   return m_remains;
660 }
661
662 }
663 }