Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
7a6c1eb09e89e4c18b7df4636250d52e2c7630eb
[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 static int ptask_host_count = 0;
20 static xbt_dict_t ptask_parallel_task_link_set = NULL;
21 lmm_system_t ptask_maxmin_system = NULL;
22
23
24 /**************************************/
25 /*** Resource Creation & Destruction **/
26 /**************************************/
27
28 static void ptask_netlink_parse_init(sg_platf_link_cbarg_t link)
29 {
30   netlink_parse_init(link);
31   current_property_set = NULL;
32 }
33
34 void surf_host_model_init_ptask_L07(void)
35 {
36   XBT_INFO("Switching to the L07 model to handle parallel tasks.");
37   xbt_assert(!surf_cpu_model_pm, "CPU model type already defined");
38   xbt_assert(!surf_network_model, "network model type already defined");
39
40   // Define the callbacks to parse the XML
41   sg_platf_link_add_cb(ptask_netlink_parse_init);
42   sg_platf_postparse_add_cb(host_add_traces);
43
44   surf_host_model = new simgrid::surf::HostL07Model();
45   simgrid::surf::Model *model = surf_host_model;
46   xbt_dynar_push(all_existing_models, &model);
47 }
48
49
50 namespace simgrid {
51 namespace surf {
52
53 HostL07Model::HostL07Model() : HostModel() {
54   if (!ptask_maxmin_system)
55         ptask_maxmin_system = lmm_system_new(1);
56   surf_host_model = NULL;
57   surf_network_model = new NetworkL07Model(this);
58   surf_cpu_model_pm = new CpuL07Model(this);
59
60   routing_model_create(surf_network_model->createLink("__loopback__",
61                                                           498000000, NULL,
62                                                           0.000015, NULL,
63                                                           SURF_RESOURCE_ON, NULL,
64                                                           SURF_LINK_FATPIPE, NULL));
65 }
66
67 HostL07Model::~HostL07Model() {
68   xbt_dict_free(&ptask_parallel_task_link_set);
69
70   delete surf_cpu_model_pm;
71   delete surf_network_model;
72   ptask_host_count = 0;
73
74   if (ptask_maxmin_system) {
75     lmm_system_free(ptask_maxmin_system);
76     ptask_maxmin_system = NULL;
77   }
78 }
79
80 double HostL07Model::shareResources(double /*now*/)
81 {
82   L07Action *action;
83
84   ActionList *running_actions = getRunningActionSet();
85   double min = this->shareResourcesMaxMin(running_actions,
86                                               ptask_maxmin_system,
87                                               bottleneck_solve);
88
89   for(ActionList::iterator it(running_actions->begin()), itend(running_actions->end())
90          ; it != itend ; ++it) {
91         action = static_cast<L07Action*>(&*it);
92     if (action->m_latency > 0) {
93       if (min < 0) {
94         min = action->m_latency;
95         XBT_DEBUG("Updating min (value) with %p (start %f): %f", action,
96                action->getStartTime(), min);
97       } else if (action->m_latency < min) {
98         min = action->m_latency;
99         XBT_DEBUG("Updating min (latency) with %p (start %f): %f", action,
100                action->getStartTime(), min);
101       }
102     }
103   }
104
105   XBT_DEBUG("min value : %f", min);
106
107   return min;
108 }
109
110 void HostL07Model::updateActionsState(double /*now*/, double delta) {
111
112   L07Action *action;
113   ActionList *actionSet = getRunningActionSet();
114
115   for(ActionList::iterator it = actionSet->begin(), itNext = it
116          ; it != actionSet->end()
117          ; it =  itNext) {
118         ++itNext;
119     action = static_cast<L07Action*>(&*it);
120     if (action->m_latency > 0) {
121       if (action->m_latency > delta) {
122         double_update(&(action->m_latency), delta, sg_surf_precision);
123       } else {
124         action->m_latency = 0.0;
125       }
126       if ((action->m_latency == 0.0) && (action->isSuspended() == 0)) {
127         action->updateBound();
128         lmm_update_variable_weight(ptask_maxmin_system, action->getVariable(), 1.0);
129       }
130     }
131     XBT_DEBUG("Action (%p) : remains (%g) updated by %g.",
132            action, action->getRemains(), lmm_variable_getvalue(action->getVariable()) * delta);
133     action->updateRemains(lmm_variable_getvalue(action->getVariable()) * delta);
134
135     if (action->getMaxDuration() != NO_MAX_DURATION)
136       action->updateMaxDuration(delta);
137
138     XBT_DEBUG("Action (%p) : remains (%g).", action, action->getRemains());
139
140     /* In the next if cascade, the action can be finished either because:
141      *  - The amount of remaining work reached 0
142      *  - The max duration was reached
143      * If it's not done, it may have failed.
144      */
145
146     if ((action->getRemains() <= 0) &&
147         (lmm_get_variable_weight(action->getVariable()) > 0)) {
148       action->finish();
149       action->setState(SURF_ACTION_DONE);
150     } else if ((action->getMaxDuration() != NO_MAX_DURATION) &&
151                (action->getMaxDuration() <= 0)) {
152       action->finish();
153       action->setState(SURF_ACTION_DONE);
154     } else {
155       /* Need to check that none of the model has failed */
156       lmm_constraint_t cnst = NULL;
157       int i = 0;
158
159       while ((cnst = lmm_get_cnst_from_var(ptask_maxmin_system, action->getVariable(), i++))) {
160         void *constraint_id = lmm_constraint_id(cnst);
161
162         if (static_cast<Host*>(constraint_id)->getState() == SURF_RESOURCE_OFF) {
163           XBT_DEBUG("Action (%p) Failed!!", action);
164           action->finish();
165           action->setState(SURF_ACTION_FAILED);
166           break;
167         }
168       }
169     }
170   }
171   return;
172 }
173
174 Action *HostL07Model::executeParallelTask(int host_nb,
175                                           sg_host_t*host_list,
176                                                                                   double *flops_amount,
177                                                                                   double *bytes_amount,
178                                                                                   double rate)
179 {
180   L07Action *action = new L07Action(this, 1, 0);
181   unsigned int cpt;
182   int nb_link = 0;
183   int nb_host = 0;
184   double latency = 0.0;
185
186   action->p_edgeList->reserve(host_nb);
187   for (int i = 0; i<host_nb; i++)
188           action->p_edgeList->push_back(sg_host_edge(host_list[i]));
189
190   if (ptask_parallel_task_link_set == NULL)
191     ptask_parallel_task_link_set = xbt_dict_new_homogeneous(NULL);
192
193   xbt_dict_reset(ptask_parallel_task_link_set);
194
195   /* Compute the number of affected resources... */
196   for (int i = 0; i < host_nb; i++) {
197     for (int j = 0; j < host_nb; j++) {
198       xbt_dynar_t route=NULL;
199
200       if (bytes_amount[i * host_nb + j] > 0) {
201         double lat=0.0;
202         unsigned int cpt;
203         void *_link;
204         LinkL07 *link;
205
206         routing_platf->getRouteAndLatency((*action->p_edgeList)[i], (*action->p_edgeList)[j],
207                                                   &route, &lat);
208         latency = MAX(latency, lat);
209
210         xbt_dynar_foreach(route, cpt, _link) {
211            link = static_cast<LinkL07*>(_link);
212            xbt_dict_set(ptask_parallel_task_link_set, link->getName(), link, NULL);
213         }
214       }
215     }
216   }
217
218   nb_link = xbt_dict_length(ptask_parallel_task_link_set);
219   xbt_dict_reset(ptask_parallel_task_link_set);
220
221   for (int i = 0; i < host_nb; i++)
222     if (flops_amount[i] > 0)
223       nb_host++;
224
225   XBT_DEBUG("Creating a parallel task (%p) with %d cpus and %d links.",
226          action, host_nb, nb_link);
227   action->m_suspended = 0; /* valgrind seems to want it despite the calloc... */
228   action->p_computationAmount = flops_amount;
229   action->p_communicationAmount = bytes_amount;
230   action->m_latency = latency;
231   action->m_rate = rate;
232
233   action->p_variable = lmm_variable_new(ptask_maxmin_system, action, 1.0,
234                                         (rate > 0 ? rate : -1.0),
235                                                                                 host_nb + nb_link);
236
237   if (action->m_latency > 0)
238     lmm_update_variable_weight(ptask_maxmin_system, action->getVariable(), 0.0);
239
240   for (int i = 0; i < host_nb; i++)
241     lmm_expand(ptask_maxmin_system,
242                sg_host_surfcpu(host_list[i])->getConstraint(),
243                action->getVariable(), flops_amount[i]);
244
245   for (int i = 0; i < host_nb; i++) {
246     for (int j = 0; j < host_nb; j++) {
247       void *_link;
248
249       xbt_dynar_t route=NULL;
250       if (bytes_amount[i * host_nb + j] == 0.0)
251         continue;
252
253       routing_platf->getRouteAndLatency((*action->p_edgeList)[i], (*action->p_edgeList)[j],
254                                             &route, NULL);
255
256       xbt_dynar_foreach(route, cpt, _link) {
257         LinkL07 *link = static_cast<LinkL07*>(_link);
258         lmm_expand_add(ptask_maxmin_system, link->getConstraint(),
259                        action->getVariable(),
260                        bytes_amount[i * host_nb + j]);
261       }
262     }
263   }
264
265   if (nb_link + nb_host == 0) {
266     action->setCost(1.0);
267     action->setRemains(0.0);
268   }
269
270   return action;
271 }
272
273 Host *HostL07Model::createHost(const char *name,RoutingEdge *netElm, Cpu *cpu, xbt_dict_t props)
274 {
275   return new simgrid::surf::Host(this, name, props,
276                   (xbt_dynar_t)xbt_lib_get_or_null(storage_lib, name, ROUTING_STORAGE_HOST_LEVEL),
277                   netElm, cpu);
278 }
279
280 Action *NetworkL07Model::communicate(RoutingEdge *src, RoutingEdge *dst,
281                                        double size, double rate)
282 {
283   sg_host_t*host_list = xbt_new0(sg_host_t, 2);
284   double *flops_amount = xbt_new0(double, 2);
285   double *bytes_amount = xbt_new0(double, 4);
286   Action *res = NULL;
287
288   host_list[0] = sg_host_by_name(src->getName());
289   host_list[1] = sg_host_by_name(dst->getName());
290   bytes_amount[1] = size;
291
292   res = p_hostModel->executeParallelTask(2, host_list,
293                                     flops_amount,
294                                     bytes_amount, rate);
295
296   return res;
297 }
298
299 Cpu *CpuL07Model::createCpu(const char *name,  xbt_dynar_t powerPeak,
300                           int pstate, double power_scale,
301                           tmgr_trace_t power_trace, int core,
302                           e_surf_resource_state_t state_initial,
303                           tmgr_trace_t state_trace)
304 {
305   double power_initial = xbt_dynar_get_as(powerPeak, pstate, double);
306   CpuL07 *cpu = new CpuL07(this, name, power_initial, power_scale, power_trace,
307                          core, state_initial, state_trace);
308   return cpu;
309 }
310
311 Link* NetworkL07Model::createLink(const char *name,
312                                  double bw_initial,
313                                  tmgr_trace_t bw_trace,
314                                  double lat_initial,
315                                  tmgr_trace_t lat_trace,
316                                  e_surf_resource_state_t state_initial,
317                                  tmgr_trace_t state_trace,
318                                  e_surf_link_sharing_policy_t policy,
319                                  xbt_dict_t properties)
320 {
321   xbt_assert(!Link::byName(name),
322                  "Link '%s' declared several times in the platform file.", name);
323
324   Link* link = new LinkL07(this, name, properties,
325                              bw_initial, bw_trace,
326                                          lat_initial, lat_trace,
327                                          state_initial, state_trace,
328                                          policy);
329   Link::onCreation(link);
330   return link;
331 }
332
333 void HostL07Model::addTraces()
334 {
335   xbt_dict_cursor_t cursor = NULL;
336   char *trace_name, *elm;
337
338   if (!trace_connect_list_host_avail)
339     return;
340
341   /* Connect traces relative to cpu */
342   xbt_dict_foreach(trace_connect_list_host_avail, cursor, trace_name, elm) {
343     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
344     CpuL07 *host = static_cast<CpuL07*>(sg_host_surfcpu(sg_host_by_name(elm)));
345
346     xbt_assert(host, "Host %s undefined", elm);
347     xbt_assert(trace, "Trace %s undefined", trace_name);
348
349     host->p_stateEvent = tmgr_history_add_trace(history, trace, 0.0, 0, host);
350   }
351
352   xbt_dict_foreach(trace_connect_list_power, cursor, trace_name, elm) {
353     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
354     CpuL07 *host = static_cast<CpuL07*>(sg_host_surfcpu(sg_host_by_name(elm)));
355
356     xbt_assert(host, "Host %s undefined", elm);
357     xbt_assert(trace, "Trace %s undefined", trace_name);
358
359     host->p_speedEvent = tmgr_history_add_trace(history, trace, 0.0, 0, host);
360   }
361
362   /* Connect traces relative to network */
363   xbt_dict_foreach(trace_connect_list_link_avail, cursor, trace_name, elm) {
364     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
365     LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
366
367     xbt_assert(link, "Link %s undefined", elm);
368     xbt_assert(trace, "Trace %s undefined", trace_name);
369
370     link->p_stateEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
371   }
372
373   xbt_dict_foreach(trace_connect_list_bandwidth, cursor, trace_name, elm) {
374     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
375     LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
376
377     xbt_assert(link, "Link %s undefined", elm);
378     xbt_assert(trace, "Trace %s undefined", trace_name);
379
380     link->p_bwEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
381   }
382
383   xbt_dict_foreach(trace_connect_list_latency, cursor, trace_name, elm) {
384     tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
385     LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
386
387     xbt_assert(link, "Link %s undefined", elm);
388     xbt_assert(trace, "Trace %s undefined", trace_name);
389
390     link->p_latEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
391   }
392 }
393
394 /************
395  * Resource *
396  ************/
397 CpuL07::CpuL07(CpuL07Model *model, const char* name,
398                      double speedInitial, double speedScale, tmgr_trace_t speedTrace,
399                            int core, e_surf_resource_state_t state_initial, tmgr_trace_t state_trace)
400  : Cpu(model, name, lmm_constraint_new(ptask_maxmin_system, this, speedInitial * speedScale),
401            core, speedInitial, speedScale, state_initial)
402 {
403   xbt_assert(m_speedScale > 0, "Power has to be >0");
404
405   if (speedTrace)
406     p_speedEvent = tmgr_history_add_trace(history, speedTrace, 0.0, 0, this);
407   else
408     p_speedEvent = NULL;
409
410   if (state_trace)
411         p_stateEvent = tmgr_history_add_trace(history, state_trace, 0.0, 0, this);
412 }
413
414 CpuL07::~CpuL07()
415 {
416 }
417
418 LinkL07::LinkL07(NetworkL07Model *model, const char* name, xbt_dict_t props,
419                          double bw_initial,
420                          tmgr_trace_t bw_trace,
421                          double lat_initial,
422                          tmgr_trace_t lat_trace,
423                          e_surf_resource_state_t state_initial,
424                          tmgr_trace_t state_trace,
425                          e_surf_link_sharing_policy_t policy)
426  : Link(model, name, props, lmm_constraint_new(ptask_maxmin_system, this, bw_initial), history, state_trace)
427 {
428   m_bwCurrent = bw_initial;
429   if (bw_trace)
430     p_bwEvent = tmgr_history_add_trace(history, bw_trace, 0.0, 0, this);
431
432   setState(state_initial);
433   m_latCurrent = lat_initial;
434
435   if (lat_trace)
436         p_latEvent = tmgr_history_add_trace(history, lat_trace, 0.0, 0, this);
437
438   if (policy == SURF_LINK_FATPIPE)
439         lmm_constraint_shared(getConstraint());
440 }
441
442 Action *CpuL07::execute(double size)
443 {
444   sg_host_t*host_list = xbt_new0(sg_host_t, 1);
445   double *flops_amount = xbt_new0(double, 1);
446   double *bytes_amount = xbt_new0(double, 1);
447
448   host_list[0] = getHost();
449   flops_amount[0] = size;
450
451   return static_cast<CpuL07Model*>(getModel())
452     ->p_hostModel
453     ->executeParallelTask( 1, host_list, flops_amount, bytes_amount, -1);
454 }
455
456 Action *CpuL07::sleep(double duration)
457 {
458   L07Action *action = NULL;
459
460   XBT_IN("(%s,%g)", getName(), duration);
461
462   action = static_cast<L07Action*>(execute(1.0));
463   action->m_maxDuration = duration;
464   action->m_suspended = 2;
465   lmm_update_variable_weight(ptask_maxmin_system, action->getVariable(), 0.0);
466
467   XBT_OUT();
468   return action;
469 }
470
471 bool CpuL07::isUsed(){
472   return lmm_constraint_used(ptask_maxmin_system, getConstraint());
473 }
474
475 bool LinkL07::isUsed(){
476   return lmm_constraint_used(ptask_maxmin_system, getConstraint());
477 }
478
479 void CpuL07::updateState(tmgr_trace_event_t event_type, double value, double /*date*/){
480   XBT_DEBUG("Updating cpu %s (%p) with value %g", getName(), this, value);
481   if (event_type == p_speedEvent) {
482           m_speedScale = value;
483     lmm_update_constraint_bound(ptask_maxmin_system, getConstraint(), m_speedPeak * m_speedScale);
484     if (tmgr_trace_event_free(event_type))
485       p_speedEvent = NULL;
486   } else if (event_type == p_stateEvent) {
487     if (value > 0)
488       setState(SURF_RESOURCE_ON);
489     else
490       setState(SURF_RESOURCE_OFF);
491     if (tmgr_trace_event_free(event_type))
492       p_stateEvent = NULL;
493   } else {
494     XBT_CRITICAL("Unknown event ! \n");
495     xbt_abort();
496   }
497   return;
498 }
499
500 void LinkL07::updateState(tmgr_trace_event_t event_type, double value, double date) {
501   XBT_DEBUG("Updating link %s (%p) with value=%f for date=%g", getName(), this, value, date);
502   if (event_type == p_bwEvent) {
503     updateBandwidth(value, date);
504     if (tmgr_trace_event_free(event_type))
505       p_bwEvent = NULL;
506   } else if (event_type == p_latEvent) {
507     updateLatency(value, date);
508     if (tmgr_trace_event_free(event_type))
509       p_latEvent = NULL;
510   } else if (event_type == p_stateEvent) {
511     if (value > 0)
512       setState(SURF_RESOURCE_ON);
513     else
514       setState(SURF_RESOURCE_OFF);
515     if (tmgr_trace_event_free(event_type))
516       p_stateEvent = NULL;
517   } else {
518     XBT_CRITICAL("Unknown event ! \n");
519     xbt_abort();
520   }
521   return;
522 }
523
524 double LinkL07::getBandwidth()
525 {
526   return m_bwCurrent;
527 }
528
529 void LinkL07::updateBandwidth(double value, double date)
530 {
531   m_bwCurrent = value;
532   lmm_update_constraint_bound(ptask_maxmin_system, getConstraint(), m_bwCurrent);
533 }
534
535 void LinkL07::updateLatency(double value, double date)
536 {
537   lmm_variable_t var = NULL;
538   L07Action *action;
539   lmm_element_t elem = NULL;
540
541   m_latCurrent = value;
542   while ((var = lmm_get_var_from_cnst(ptask_maxmin_system, getConstraint(), &elem))) {
543     action = static_cast<L07Action*>(lmm_variable_id(var));
544     action->updateBound();
545   }
546 }
547
548 /**********
549  * Action *
550  **********/
551
552 L07Action::~L07Action(){
553   free(p_communicationAmount);
554   free(p_computationAmount);
555 }
556
557 void L07Action::updateBound()
558 {
559   double lat_current = 0.0;
560   double lat_bound = -1.0;
561   int i, j;
562
563   int hostNb = p_edgeList->size();
564
565   for (i = 0; i < hostNb; i++) {
566     for (j = 0; j < hostNb; j++) {
567       xbt_dynar_t route=NULL;
568
569       if (p_communicationAmount[i * hostNb + j] > 0) {
570         double lat = 0.0;
571         routing_platf->getRouteAndLatency((*p_edgeList)[i], (*p_edgeList)[j],
572                                                           &route, &lat);
573
574         lat_current = MAX(lat_current, lat * p_communicationAmount[i * hostNb + j]);
575       }
576     }
577   }
578   lat_bound = sg_tcp_gamma / (2.0 * lat_current);
579   XBT_DEBUG("action (%p) : lat_bound = %g", this, lat_bound);
580   if ((m_latency == 0.0) && (m_suspended == 0)) {
581     if (m_rate < 0)
582       lmm_update_variable_bound(ptask_maxmin_system, getVariable(), lat_bound);
583     else
584       lmm_update_variable_bound(ptask_maxmin_system, getVariable(),
585         std::min(m_rate, lat_bound));
586   }
587 }
588
589 int L07Action::unref()
590 {
591   m_refcount--;
592   if (!m_refcount) {
593     if (action_hook.is_linked())
594           p_stateSet->erase(p_stateSet->iterator_to(*this));
595     if (getVariable())
596       lmm_variable_free(ptask_maxmin_system, getVariable());
597     delete this;
598     return 1;
599   }
600   return 0;
601 }
602
603 void L07Action::cancel()
604 {
605   setState(SURF_ACTION_FAILED);
606   return;
607 }
608
609 void L07Action::suspend()
610 {
611   XBT_IN("(%p))", this);
612   if (m_suspended != 2) {
613     m_suspended = 1;
614     lmm_update_variable_weight(ptask_maxmin_system, getVariable(), 0.0);
615   }
616   XBT_OUT();
617 }
618
619 void L07Action::resume()
620 {
621   XBT_IN("(%p)", this);
622   if (m_suspended != 2) {
623     lmm_update_variable_weight(ptask_maxmin_system, getVariable(), 1.0);
624     m_suspended = 0;
625   }
626   XBT_OUT();
627 }
628
629 bool L07Action::isSuspended()
630 {
631   return m_suspended == 1;
632 }
633
634 void L07Action::setMaxDuration(double duration)
635 {                               /* FIXME: should inherit */
636   XBT_IN("(%p,%g)", this, duration);
637   m_maxDuration = duration;
638   XBT_OUT();
639 }
640
641 void L07Action::setPriority(double priority)
642 {                               /* FIXME: should inherit */
643   XBT_IN("(%p,%g)", this, priority);
644   m_priority = priority;
645   XBT_OUT();
646 }
647
648 double L07Action::getRemains()
649 {
650   XBT_IN("(%p)", this);
651   XBT_OUT();
652   return m_remains;
653 }
654
655 }
656 }