Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
1f8a4879e9720521511849a7181ee64c8da3f087
[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
17 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_host);
18
19 /**************************************/
20 /*** Resource Creation & Destruction **/
21 /**************************************/
22
23 static void ptask_netlink_parse_init(sg_platf_link_cbarg_t link)
24 {
25   netlink_parse_init(link);
26   current_property_set = NULL;
27 }
28
29 void surf_host_model_init_ptask_L07(void)
30 {
31   XBT_INFO("Switching to the L07 model to handle parallel tasks.");
32   xbt_assert(!surf_cpu_model_pm, "CPU model type already defined");
33   xbt_assert(!surf_network_model, "network model type already defined");
34
35   // Define the callbacks to parse the XML
36   sg_platf_link_add_cb(ptask_netlink_parse_init);
37   sg_platf_postparse_add_cb(host_add_traces);
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                                                           SURF_RESOURCE_ON, 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         p_maxminSystem = NULL; // Avoid multi-free
73 }
74 NetworkL07Model::NetworkL07Model(HostL07Model *hmodel, lmm_system_t sys)
75         : NetworkModel()
76         , p_hostModel(hmodel)
77         {
78           p_maxminSystem = sys;
79         }
80 NetworkL07Model::~NetworkL07Model()
81 {
82         surf_network_model = NULL;
83         p_maxminSystem = NULL; // Avoid multi-free
84 }
85
86
87 double HostL07Model::shareResources(double /*now*/)
88 {
89   L07Action *action;
90
91   ActionList *running_actions = getRunningActionSet();
92   double min = this->shareResourcesMaxMin(running_actions,
93                                               p_maxminSystem,
94                                               bottleneck_solve);
95
96   for(ActionList::iterator it(running_actions->begin()), itend(running_actions->end())
97          ; it != itend ; ++it) {
98         action = static_cast<L07Action*>(&*it);
99     if (action->m_latency > 0) {
100       if (min < 0) {
101         min = action->m_latency;
102         XBT_DEBUG("Updating min (value) with %p (start %f): %f", action,
103                action->getStartTime(), min);
104       } else if (action->m_latency < min) {
105         min = action->m_latency;
106         XBT_DEBUG("Updating min (latency) with %p (start %f): %f", action,
107                action->getStartTime(), min);
108       }
109     }
110   }
111
112   XBT_DEBUG("min value : %f", min);
113
114   return min;
115 }
116
117 void HostL07Model::updateActionsState(double /*now*/, double delta) {
118
119   L07Action *action;
120   ActionList *actionSet = getRunningActionSet();
121
122   for(ActionList::iterator it = actionSet->begin(), itNext = it
123          ; it != actionSet->end()
124          ; it =  itNext) {
125         ++itNext;
126     action = static_cast<L07Action*>(&*it);
127     if (action->m_latency > 0) {
128       if (action->m_latency > delta) {
129         double_update(&(action->m_latency), delta, sg_surf_precision);
130       } else {
131         action->m_latency = 0.0;
132       }
133       if ((action->m_latency == 0.0) && (action->isSuspended() == 0)) {
134         action->updateBound();
135         lmm_update_variable_weight(p_maxminSystem, action->getVariable(), 1.0);
136       }
137     }
138     XBT_DEBUG("Action (%p) : remains (%g) updated by %g.",
139            action, action->getRemains(), lmm_variable_getvalue(action->getVariable()) * delta);
140     action->updateRemains(lmm_variable_getvalue(action->getVariable()) * delta);
141
142     if (action->getMaxDuration() != NO_MAX_DURATION)
143       action->updateMaxDuration(delta);
144
145     XBT_DEBUG("Action (%p) : remains (%g).", action, action->getRemains());
146
147     /* In the next if cascade, the action can be finished either because:
148      *  - The amount of remaining work reached 0
149      *  - The max duration was reached
150      * If it's not done, it may have failed.
151      */
152
153     if ((action->getRemains() <= 0) &&
154         (lmm_get_variable_weight(action->getVariable()) > 0)) {
155       action->finish();
156       action->setState(SURF_ACTION_DONE);
157     } else if ((action->getMaxDuration() != NO_MAX_DURATION) &&
158                (action->getMaxDuration() <= 0)) {
159       action->finish();
160       action->setState(SURF_ACTION_DONE);
161     } else {
162       /* Need to check that none of the model has failed */
163       lmm_constraint_t cnst = NULL;
164       int i = 0;
165
166       while ((cnst = lmm_get_cnst_from_var(p_maxminSystem, action->getVariable(), i++))) {
167         void *constraint_id = lmm_constraint_id(cnst);
168
169         if (static_cast<Host*>(constraint_id)->getState() == SURF_RESOURCE_OFF) {
170           XBT_DEBUG("Action (%p) Failed!!", action);
171           action->finish();
172           action->setState(SURF_ACTION_FAILED);
173           break;
174         }
175       }
176     }
177   }
178   return;
179 }
180
181 Action *HostL07Model::executeParallelTask(int host_nb,
182                                           sg_host_t*host_list,
183                                                                                   double *flops_amount,
184                                                                                   double *bytes_amount,
185                                                                                   double rate)
186 {
187   L07Action *action = new L07Action(this, 1, 0);
188   unsigned int cpt;
189   int nb_link = 0;
190   int nb_used_host = 0; /* Only the hosts with something to compute (>0 flops) are counted) */
191   double latency = 0.0;
192
193   xbt_dict_t ptask_parallel_task_link_set = xbt_dict_new_homogeneous(NULL);
194
195   action->p_edgeList->reserve(host_nb);
196   for (int i = 0; i<host_nb; i++)
197           action->p_edgeList->push_back(sg_host_edge(host_list[i]));
198
199   /* Compute the number of affected resources... */
200   for (int i = 0; i < host_nb; i++) {
201     for (int j = 0; j < host_nb; j++) {
202       xbt_dynar_t route=NULL;
203
204       if (bytes_amount[i * host_nb + j] > 0) {
205         double lat=0.0;
206         unsigned int cpt;
207         void *_link;
208         LinkL07 *link;
209
210         routing_platf->getRouteAndLatency((*action->p_edgeList)[i], (*action->p_edgeList)[j],
211                                                   &route, &lat);
212         latency = MAX(latency, lat);
213
214         xbt_dynar_foreach(route, cpt, _link) {
215            link = static_cast<LinkL07*>(_link);
216            xbt_dict_set(ptask_parallel_task_link_set, link->getName(), link, NULL);
217         }
218       }
219     }
220   }
221
222   nb_link = xbt_dict_length(ptask_parallel_task_link_set);
223   xbt_dict_free(&ptask_parallel_task_link_set);
224
225   for (int i = 0; i < host_nb; i++)
226     if (flops_amount[i] > 0)
227       nb_used_host++;
228
229   XBT_DEBUG("Creating a parallel task (%p) with %d cpus and %d links.",
230          action, host_nb, nb_link);
231   action->p_computationAmount = flops_amount;
232   action->p_communicationAmount = bytes_amount;
233   action->m_latency = latency;
234   action->m_rate = rate;
235
236   action->p_variable = lmm_variable_new(p_maxminSystem, action, 1.0,
237                                         (rate > 0 ? rate : -1.0),
238                                                                                 host_nb + nb_link);
239
240   if (action->m_latency > 0)
241     lmm_update_variable_weight(p_maxminSystem, action->getVariable(), 0.0);
242
243   for (int i = 0; i < host_nb; i++)
244     lmm_expand(p_maxminSystem,
245                sg_host_surfcpu(host_list[i])->getConstraint(),
246                action->getVariable(), flops_amount[i]);
247
248   for (int i = 0; i < host_nb; i++) {
249     for (int j = 0; j < host_nb; j++) {
250       void *_link;
251
252       xbt_dynar_t route=NULL;
253       if (bytes_amount[i * host_nb + j] == 0.0)
254         continue;
255
256       routing_platf->getRouteAndLatency((*action->p_edgeList)[i], (*action->p_edgeList)[j],
257                                             &route, NULL);
258
259       xbt_dynar_foreach(route, cpt, _link) {
260         LinkL07 *link = static_cast<LinkL07*>(_link);
261         lmm_expand_add(p_maxminSystem, link->getConstraint(),
262                        action->getVariable(),
263                        bytes_amount[i * host_nb + j]);
264       }
265     }
266   }
267
268   if (nb_link + nb_used_host == 0) {
269     action->setCost(1.0);
270     action->setRemains(0.0);
271   }
272
273   return action;
274 }
275
276 Action *NetworkL07Model::communicate(RoutingEdge *src, RoutingEdge *dst,
277                                        double size, double rate)
278 {
279   sg_host_t*host_list = xbt_new0(sg_host_t, 2);
280   double *flops_amount = xbt_new0(double, 2);
281   double *bytes_amount = xbt_new0(double, 4);
282   Action *res = NULL;
283
284   host_list[0] = sg_host_by_name(src->getName());
285   host_list[1] = sg_host_by_name(dst->getName());
286   bytes_amount[1] = size;
287
288   res = p_hostModel->executeParallelTask(2, host_list,
289                                     flops_amount,
290                                     bytes_amount, rate);
291
292   return res;
293 }
294
295 Cpu *CpuL07Model::createCpu(simgrid::Host *host,  xbt_dynar_t powerPeakList,
296                           int pstate, double power_scale,
297                           tmgr_trace_t power_trace, int core,
298                           e_surf_resource_state_t state_initial,
299                           tmgr_trace_t state_trace)
300 {
301   CpuL07 *cpu = new CpuL07(this, host, powerPeakList, pstate, power_scale, power_trace,
302                          core, state_initial, state_trace);
303   return cpu;
304 }
305
306 Link* NetworkL07Model::createLink(const char *name,
307                                  double bw_initial,
308                                  tmgr_trace_t bw_trace,
309                                  double lat_initial,
310                                  tmgr_trace_t lat_trace,
311                                  e_surf_resource_state_t state_initial,
312                                  tmgr_trace_t state_trace,
313                                  e_surf_link_sharing_policy_t policy,
314                                  xbt_dict_t properties)
315 {
316   xbt_assert(!Link::byName(name),
317                  "Link '%s' declared several times in the platform file.", name);
318
319   Link* link = new LinkL07(this, name, properties,
320                              bw_initial, bw_trace,
321                                          lat_initial, lat_trace,
322                                          state_initial, state_trace,
323                                          policy);
324   Link::onCreation(link);
325   return link;
326 }
327
328 void HostL07Model::addTraces()
329 {
330   xbt_dict_cursor_t cursor = NULL;
331   char *trace_name, *elm;
332
333   if (!trace_connect_list_host_avail)
334     return;
335
336   /* Connect traces relative to cpu */
337   xbt_dict_foreach(trace_connect_list_host_avail, cursor, trace_name, elm) {
338     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
339     CpuL07 *host = static_cast<CpuL07*>(sg_host_surfcpu(sg_host_by_name(elm)));
340
341     xbt_assert(host, "Host %s undefined", elm);
342     xbt_assert(trace, "Trace %s undefined", trace_name);
343
344     host->p_stateEvent = tmgr_history_add_trace(history, trace, 0.0, 0, host);
345   }
346
347   xbt_dict_foreach(trace_connect_list_power, 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_surfcpu(sg_host_by_name(elm)));
350
351     xbt_assert(host, "Host %s undefined", elm);
352     xbt_assert(trace, "Trace %s undefined", trace_name);
353
354     host->p_speedEvent = tmgr_history_add_trace(history, trace, 0.0, 0, host);
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 = tmgr_history_add_trace(history, trace, 0.0, 0, link);
366   }
367
368   xbt_dict_foreach(trace_connect_list_bandwidth, 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 = tmgr_history_add_trace(history, trace, 0.0, 0, link);
376   }
377
378   xbt_dict_foreach(trace_connect_list_latency, 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 = tmgr_history_add_trace(history, trace, 0.0, 0, link);
386   }
387 }
388
389 /************
390  * Resource *
391  ************/
392
393 CpuL07::CpuL07(CpuL07Model *model, simgrid::Host *host,
394                      xbt_dynar_t speedPeakList, int pstate,
395                                  double speedScale, tmgr_trace_t speedTrace,
396                          int core, e_surf_resource_state_t state_initial, tmgr_trace_t state_trace)
397  : Cpu(model, host, speedPeakList, pstate,
398            core, xbt_dynar_get_as(speedPeakList,pstate,double), speedScale, state_initial)
399 {
400   xbt_assert(m_speedScale > 0, "Power has to be >0");
401   p_constraint = lmm_constraint_new(model->getMaxminSystem(), this, xbt_dynar_get_as(speedPeakList,pstate,double) * speedScale);
402
403   if (speedTrace)
404     p_speedEvent = tmgr_history_add_trace(history, speedTrace, 0.0, 0, this);
405   else
406     p_speedEvent = NULL;
407
408   if (state_trace)
409         p_stateEvent = tmgr_history_add_trace(history, state_trace, 0.0, 0, this);
410 }
411
412 CpuL07::~CpuL07()
413 {
414 }
415
416 LinkL07::LinkL07(NetworkL07Model *model, const char* name, xbt_dict_t props,
417                          double bw_initial,
418                          tmgr_trace_t bw_trace,
419                          double lat_initial,
420                          tmgr_trace_t lat_trace,
421                          e_surf_resource_state_t state_initial,
422                          tmgr_trace_t state_trace,
423                          e_surf_link_sharing_policy_t policy)
424  : Link(model, name, props, lmm_constraint_new(model->getMaxminSystem(), this, bw_initial), history, state_trace)
425 {
426   m_bwCurrent = bw_initial;
427   if (bw_trace)
428     p_bwEvent = tmgr_history_add_trace(history, bw_trace, 0.0, 0, this);
429
430   setState(state_initial);
431   m_latCurrent = lat_initial;
432
433   if (lat_trace)
434         p_latEvent = tmgr_history_add_trace(history, lat_trace, 0.0, 0, this);
435
436   if (policy == SURF_LINK_FATPIPE)
437         lmm_constraint_shared(getConstraint());
438 }
439
440 Action *CpuL07::execute(double size)
441 {
442   sg_host_t*host_list = xbt_new0(sg_host_t, 1);
443   double *flops_amount = xbt_new0(double, 1);
444   double *bytes_amount = xbt_new0(double, 1);
445
446   host_list[0] = getHost();
447   flops_amount[0] = size;
448
449   return static_cast<CpuL07Model*>(getModel())
450     ->p_hostModel
451     ->executeParallelTask( 1, host_list, flops_amount, bytes_amount, -1);
452 }
453
454 Action *CpuL07::sleep(double duration)
455 {
456   L07Action *action = NULL;
457
458   XBT_IN("(%s,%g)", getName(), duration);
459
460   action = static_cast<L07Action*>(execute(1.0));
461   action->m_maxDuration = duration;
462   action->m_suspended = 2;
463   lmm_update_variable_weight(getModel()->getMaxminSystem(), action->getVariable(), 0.0);
464
465   XBT_OUT();
466   return action;
467 }
468
469 bool CpuL07::isUsed(){
470   return lmm_constraint_used(getModel()->getMaxminSystem(), getConstraint());
471 }
472
473 bool LinkL07::isUsed(){
474   return lmm_constraint_used(getModel()->getMaxminSystem(), getConstraint());
475 }
476
477 void CpuL07::updateState(tmgr_trace_event_t event_type, double value, double /*date*/){
478   XBT_DEBUG("Updating cpu %s (%p) with value %g", getName(), this, value);
479   if (event_type == p_speedEvent) {
480           m_speedScale = value;
481     lmm_update_constraint_bound(getModel()->getMaxminSystem(), getConstraint(), m_speedPeak * m_speedScale);
482     if (tmgr_trace_event_free(event_type))
483       p_speedEvent = NULL;
484   } else if (event_type == p_stateEvent) {
485     if (value > 0)
486       setState(SURF_RESOURCE_ON);
487     else
488       setState(SURF_RESOURCE_OFF);
489     if (tmgr_trace_event_free(event_type))
490       p_stateEvent = NULL;
491   } else {
492     XBT_CRITICAL("Unknown event ! \n");
493     xbt_abort();
494   }
495   return;
496 }
497
498 void LinkL07::updateState(tmgr_trace_event_t event_type, double value, double date) {
499   XBT_DEBUG("Updating link %s (%p) with value=%f for date=%g", getName(), this, value, date);
500   if (event_type == p_bwEvent) {
501     updateBandwidth(value, date);
502     if (tmgr_trace_event_free(event_type))
503       p_bwEvent = NULL;
504   } else if (event_type == p_latEvent) {
505     updateLatency(value, date);
506     if (tmgr_trace_event_free(event_type))
507       p_latEvent = NULL;
508   } else if (event_type == p_stateEvent) {
509     if (value > 0)
510       setState(SURF_RESOURCE_ON);
511     else
512       setState(SURF_RESOURCE_OFF);
513     if (tmgr_trace_event_free(event_type))
514       p_stateEvent = NULL;
515   } else {
516     XBT_CRITICAL("Unknown event ! \n");
517     xbt_abort();
518   }
519   return;
520 }
521
522 double LinkL07::getBandwidth()
523 {
524   return m_bwCurrent;
525 }
526
527 void LinkL07::updateBandwidth(double value, double date)
528 {
529   m_bwCurrent = value;
530   lmm_update_constraint_bound(getModel()->getMaxminSystem(), getConstraint(), m_bwCurrent);
531 }
532
533 void LinkL07::updateLatency(double value, double date)
534 {
535   lmm_variable_t var = NULL;
536   L07Action *action;
537   lmm_element_t elem = NULL;
538
539   m_latCurrent = value;
540   while ((var = lmm_get_var_from_cnst(getModel()->getMaxminSystem(), getConstraint(), &elem))) {
541     action = static_cast<L07Action*>(lmm_variable_id(var));
542     action->updateBound();
543   }
544 }
545
546 /**********
547  * Action *
548  **********/
549
550 L07Action::~L07Action(){
551   free(p_communicationAmount);
552   free(p_computationAmount);
553 }
554
555 void L07Action::updateBound()
556 {
557   double lat_current = 0.0;
558   double lat_bound = -1.0;
559   int i, j;
560
561   int hostNb = p_edgeList->size();
562
563   for (i = 0; i < hostNb; i++) {
564     for (j = 0; j < hostNb; j++) {
565       xbt_dynar_t route=NULL;
566
567       if (p_communicationAmount[i * hostNb + j] > 0) {
568         double lat = 0.0;
569         routing_platf->getRouteAndLatency((*p_edgeList)[i], (*p_edgeList)[j],
570                                                           &route, &lat);
571
572         lat_current = MAX(lat_current, lat * p_communicationAmount[i * hostNb + j]);
573       }
574     }
575   }
576   lat_bound = sg_tcp_gamma / (2.0 * lat_current);
577   XBT_DEBUG("action (%p) : lat_bound = %g", this, lat_bound);
578   if ((m_latency == 0.0) && (m_suspended == 0)) {
579     if (m_rate < 0)
580       lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(), lat_bound);
581     else
582       lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(),
583         std::min(m_rate, lat_bound));
584   }
585 }
586
587 int L07Action::unref()
588 {
589   m_refcount--;
590   if (!m_refcount) {
591     if (action_hook.is_linked())
592           p_stateSet->erase(p_stateSet->iterator_to(*this));
593     if (getVariable())
594       lmm_variable_free(getModel()->getMaxminSystem(), getVariable());
595     delete this;
596     return 1;
597   }
598   return 0;
599 }
600
601 void L07Action::suspend()
602 {
603   XBT_IN("(%p))", this);
604   if (m_suspended != 2) {
605     m_suspended = 1;
606     lmm_update_variable_weight(getModel()->getMaxminSystem(), getVariable(), 0.0);
607   }
608   XBT_OUT();
609 }
610
611 void L07Action::resume()
612 {
613   XBT_IN("(%p)", this);
614   if (m_suspended != 2) {
615     lmm_update_variable_weight(getModel()->getMaxminSystem(), getVariable(), 1.0);
616     m_suspended = 0;
617   }
618   XBT_OUT();
619 }
620
621 void L07Action::setMaxDuration(double duration)
622 {                               /* FIXME: should inherit */
623   XBT_IN("(%p,%g)", this, duration);
624   m_maxDuration = duration;
625   XBT_OUT();
626 }
627
628 void L07Action::setPriority(double priority)
629 {                               /* FIXME: should inherit */
630   XBT_IN("(%p,%g)", this, priority);
631   m_priority = priority;
632   XBT_OUT();
633 }
634
635 double L07Action::getRemains()
636 {
637   XBT_IN("(%p)", this);
638   XBT_OUT();
639   return m_remains;
640 }
641
642 }
643 }