1 /* Copyright (c) 2007-2010, 2013-2015. The SimGrid Team.
2 * All rights reserved. */
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. */
7 #include "host_ptask_L07.hpp"
9 #include "cpu_interface.hpp"
10 #include "surf_routing.hpp"
13 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_host);
15 static int ptask_host_count = 0;
16 static xbt_dict_t ptask_parallel_task_link_set = NULL;
17 lmm_system_t ptask_maxmin_system = NULL;
20 /**************************************/
21 /*** Resource Creation & Destruction **/
22 /**************************************/
24 static void ptask_netlink_parse_init(sg_platf_link_cbarg_t link)
26 netlink_parse_init(link);
27 current_property_set = NULL;
30 void surf_host_model_init_ptask_L07(void)
32 XBT_INFO("Switching to the L07 model to handle parallel tasks.");
33 xbt_assert(!surf_cpu_model_pm, "CPU model type already defined");
34 xbt_assert(!surf_network_model, "network model type already defined");
36 // Define the callbacks to parse the XML
37 sg_platf_host_add_cb(cpu_parse_init);
38 sg_platf_host_add_cb(host_parse_init);
39 sg_platf_link_add_cb(ptask_netlink_parse_init);
40 sg_platf_postparse_add_cb(host_add_traces);
42 surf_host_model = new HostL07Model();
43 Model *model = surf_host_model;
44 xbt_dynar_push(all_existing_models, &model);
48 HostL07Model::HostL07Model() : HostModel() {
49 if (!ptask_maxmin_system)
50 ptask_maxmin_system = lmm_system_new(1);
51 surf_host_model = NULL;
52 surf_network_model = new NetworkL07Model(this);
53 surf_cpu_model_pm = new CpuL07Model(this);
55 routing_model_create(surf_network_model->createLink("__loopback__",
58 SURF_RESOURCE_ON, NULL,
59 SURF_LINK_FATPIPE, NULL));
62 HostL07Model::~HostL07Model() {
63 xbt_dict_free(&ptask_parallel_task_link_set);
65 delete surf_cpu_model_pm;
66 delete surf_network_model;
69 if (ptask_maxmin_system) {
70 lmm_system_free(ptask_maxmin_system);
71 ptask_maxmin_system = NULL;
75 double HostL07Model::shareResources(double /*now*/)
79 ActionList *running_actions = getRunningActionSet();
80 double min = this->shareResourcesMaxMin(running_actions,
84 for(ActionList::iterator it(running_actions->begin()), itend(running_actions->end())
85 ; it != itend ; ++it) {
86 action = static_cast<L07Action*>(&*it);
87 if (action->m_latency > 0) {
89 min = action->m_latency;
90 XBT_DEBUG("Updating min (value) with %p (start %f): %f", action,
91 action->getStartTime(), min);
92 } else if (action->m_latency < min) {
93 min = action->m_latency;
94 XBT_DEBUG("Updating min (latency) with %p (start %f): %f", action,
95 action->getStartTime(), min);
100 XBT_DEBUG("min value : %f", min);
105 void HostL07Model::updateActionsState(double /*now*/, double delta) {
108 ActionList *actionSet = getRunningActionSet();
110 for(ActionList::iterator it = actionSet->begin(), itNext = it
111 ; it != actionSet->end()
114 action = static_cast<L07Action*>(&*it);
115 if (action->m_latency > 0) {
116 if (action->m_latency > delta) {
117 double_update(&(action->m_latency), delta, sg_surf_precision);
119 action->m_latency = 0.0;
121 if ((action->m_latency == 0.0) && (action->isSuspended() == 0)) {
122 action->updateBound();
123 lmm_update_variable_weight(ptask_maxmin_system, action->getVariable(), 1.0);
126 XBT_DEBUG("Action (%p) : remains (%g) updated by %g.",
127 action, action->getRemains(), lmm_variable_getvalue(action->getVariable()) * delta);
128 action->updateRemains(lmm_variable_getvalue(action->getVariable()) * delta);
130 if (action->getMaxDuration() != NO_MAX_DURATION)
131 action->updateMaxDuration(delta);
133 XBT_DEBUG("Action (%p) : remains (%g).", action, action->getRemains());
135 /* In the next if cascade, the action can be finished either because:
136 * - The amount of remaining work reached 0
137 * - The max duration was reached
138 * If it's not done, it may have failed.
141 if ((action->getRemains() <= 0) &&
142 (lmm_get_variable_weight(action->getVariable()) > 0)) {
144 action->setState(SURF_ACTION_DONE);
145 } else if ((action->getMaxDuration() != NO_MAX_DURATION) &&
146 (action->getMaxDuration() <= 0)) {
148 action->setState(SURF_ACTION_DONE);
150 /* Need to check that none of the model has failed */
151 lmm_constraint_t cnst = NULL;
154 while ((cnst = lmm_get_cnst_from_var(ptask_maxmin_system, action->getVariable(), i++))) {
155 void *constraint_id = lmm_constraint_id(cnst);
157 if (static_cast<Host*>(constraint_id)->getState() == SURF_RESOURCE_OFF) {
158 XBT_DEBUG("Action (%p) Failed!!", action);
160 action->setState(SURF_ACTION_FAILED);
169 Action *HostL07Model::executeParallelTask(int host_nb,
171 double *flops_amount,
172 double *bytes_amount,
175 L07Action *action = new L07Action(this, 1, 0);
179 double latency = 0.0;
181 action->p_edgeList->reserve(host_nb);
182 for (int i = 0; i<host_nb; i++)
183 action->p_edgeList->push_back(sg_host_edge(host_list[i]));
185 if (ptask_parallel_task_link_set == NULL)
186 ptask_parallel_task_link_set = xbt_dict_new_homogeneous(NULL);
188 xbt_dict_reset(ptask_parallel_task_link_set);
190 /* Compute the number of affected resources... */
191 for (int i = 0; i < host_nb; i++) {
192 for (int j = 0; j < host_nb; j++) {
193 xbt_dynar_t route=NULL;
195 if (bytes_amount[i * host_nb + j] > 0) {
201 routing_platf->getRouteAndLatency((*action->p_edgeList)[i], (*action->p_edgeList)[j],
203 latency = MAX(latency, lat);
205 xbt_dynar_foreach(route, cpt, _link) {
206 link = static_cast<LinkL07*>(_link);
207 xbt_dict_set(ptask_parallel_task_link_set, link->getName(), link, NULL);
213 nb_link = xbt_dict_length(ptask_parallel_task_link_set);
214 xbt_dict_reset(ptask_parallel_task_link_set);
216 for (int i = 0; i < host_nb; i++)
217 if (flops_amount[i] > 0)
220 XBT_DEBUG("Creating a parallel task (%p) with %d cpus and %d links.",
221 action, host_nb, nb_link);
222 action->m_suspended = 0; /* valgrind seems to want it despite the calloc... */
223 action->p_computationAmount = flops_amount;
224 action->p_communicationAmount = bytes_amount;
225 action->m_latency = latency;
226 action->m_rate = rate;
228 action->p_variable = lmm_variable_new(ptask_maxmin_system, action, 1.0,
229 (rate > 0 ? rate : -1.0),
232 if (action->m_latency > 0)
233 lmm_update_variable_weight(ptask_maxmin_system, action->getVariable(), 0.0);
235 for (int i = 0; i < host_nb; i++)
236 lmm_expand(ptask_maxmin_system,
237 sg_host_surfcpu(host_list[i])->getConstraint(),
238 action->getVariable(), flops_amount[i]);
240 for (int i = 0; i < host_nb; i++) {
241 for (int j = 0; j < host_nb; j++) {
244 xbt_dynar_t route=NULL;
245 if (bytes_amount[i * host_nb + j] == 0.0)
248 routing_platf->getRouteAndLatency((*action->p_edgeList)[i], (*action->p_edgeList)[j],
251 xbt_dynar_foreach(route, cpt, _link) {
252 LinkL07 *link = static_cast<LinkL07*>(_link);
253 lmm_expand_add(ptask_maxmin_system, link->getConstraint(),
254 action->getVariable(),
255 bytes_amount[i * host_nb + j]);
260 if (nb_link + nb_host == 0) {
261 action->setCost(1.0);
262 action->setRemains(0.0);
268 Host *HostL07Model::createHost(const char *name)
271 sg_host_t sg_host = sg_host_by_name(name);
273 xbt_assert(!surf_host_resource_priv(sg_host),
274 "Host '%s' declared several times in the platform file.",
277 wk = new HostL07(this, name, NULL,
278 sg_host_edge(sg_host),
279 sg_host_surfcpu(sg_host));
281 xbt_lib_set(host_lib, name, SURF_HOST_LEVEL, wk);
286 Action *NetworkL07Model::communicate(RoutingEdge *src, RoutingEdge *dst,
287 double size, double rate)
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);
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;
298 res = p_hostModel->executeParallelTask(2, host_list,
305 xbt_dynar_t HostL07Model::getRoute(Host *src, Host *dst)
307 xbt_dynar_t route=NULL;
308 routing_platf->getRouteAndLatency(src->p_netElm, dst->p_netElm, &route, NULL);
312 Cpu *CpuL07Model::createCpu(const char *name, xbt_dynar_t powerPeak,
313 int pstate, double power_scale,
314 tmgr_trace_t power_trace, int core,
315 e_surf_resource_state_t state_initial,
316 tmgr_trace_t state_trace,
317 xbt_dict_t cpu_properties)
319 double power_initial = xbt_dynar_get_as(powerPeak, pstate, double);
320 xbt_dynar_free(&powerPeak); // kill memory leak
321 sg_host_t sg_host = sg_host_by_name(name);
323 xbt_assert(!surf_host_resource_priv(sg_host),
324 "Host '%s' declared several times in the platform file.",
327 CpuL07 *cpu = new CpuL07(this, name, cpu_properties,
328 power_initial, power_scale, power_trace,
329 core, state_initial, state_trace);
331 sg_host_surfcpu_set(sg_host, cpu);
336 Link* NetworkL07Model::createLink(const char *name,
338 tmgr_trace_t bw_trace,
340 tmgr_trace_t lat_trace,
341 e_surf_resource_state_t state_initial,
342 tmgr_trace_t state_trace,
343 e_surf_link_sharing_policy_t policy,
344 xbt_dict_t properties)
346 xbt_assert(!Link::byName(name),
347 "Link '%s' declared several times in the platform file.", name);
349 return new LinkL07(this, name, properties,
350 bw_initial, bw_trace,
351 lat_initial, lat_trace,
352 state_initial, state_trace,
356 void HostL07Model::addTraces()
358 xbt_dict_cursor_t cursor = NULL;
359 char *trace_name, *elm;
361 if (!trace_connect_list_host_avail)
364 /* Connect traces relative to cpu */
365 xbt_dict_foreach(trace_connect_list_host_avail, cursor, trace_name, elm) {
366 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
367 CpuL07 *host = static_cast<CpuL07*>(sg_host_surfcpu(sg_host_by_name(elm)));
369 xbt_assert(host, "Host %s undefined", elm);
370 xbt_assert(trace, "Trace %s undefined", trace_name);
372 host->p_stateEvent = tmgr_history_add_trace(history, trace, 0.0, 0, host);
375 xbt_dict_foreach(trace_connect_list_power, cursor, trace_name, elm) {
376 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
377 CpuL07 *host = static_cast<CpuL07*>(sg_host_surfcpu(sg_host_by_name(elm)));
379 xbt_assert(host, "Host %s undefined", elm);
380 xbt_assert(trace, "Trace %s undefined", trace_name);
382 host->p_powerEvent = tmgr_history_add_trace(history, trace, 0.0, 0, host);
385 /* Connect traces relative to network */
386 xbt_dict_foreach(trace_connect_list_link_avail, cursor, trace_name, elm) {
387 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
388 LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
390 xbt_assert(link, "Link %s undefined", elm);
391 xbt_assert(trace, "Trace %s undefined", trace_name);
393 link->p_stateEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
396 xbt_dict_foreach(trace_connect_list_bandwidth, cursor, trace_name, elm) {
397 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
398 LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
400 xbt_assert(link, "Link %s undefined", elm);
401 xbt_assert(trace, "Trace %s undefined", trace_name);
403 link->p_bwEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
406 xbt_dict_foreach(trace_connect_list_latency, cursor, trace_name, elm) {
407 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
408 LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
410 xbt_assert(link, "Link %s undefined", elm);
411 xbt_assert(trace, "Trace %s undefined", trace_name);
413 link->p_latEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
421 HostL07::HostL07(HostModel *model, const char* name, xbt_dict_t props, RoutingEdge *netElm, Cpu *cpu)
422 : Host(model, name, props, NULL, netElm, cpu)
426 double HostL07::getPowerPeakAt(int /*pstate_index*/)
431 int HostL07::getNbPstates()
436 void HostL07::setPstate(int /*pstate_index*/)
441 int HostL07::getPstate()
446 double HostL07::getConsumedEnergy()
451 CpuL07::CpuL07(CpuL07Model *model, const char* name, xbt_dict_t props,
452 double power_initial, double power_scale, tmgr_trace_t power_trace,
453 int core, e_surf_resource_state_t state_initial, tmgr_trace_t state_trace)
454 : Cpu(model, name, props, lmm_constraint_new(ptask_maxmin_system, this, power_initial * power_scale),
455 core, power_initial, power_scale)
457 xbt_assert(m_powerScale > 0, "Power has to be >0");
460 p_powerEvent = tmgr_history_add_trace(history, power_trace, 0.0, 0, this);
464 setState(state_initial);
466 p_stateEvent = tmgr_history_add_trace(history, state_trace, 0.0, 0, this);
469 LinkL07::LinkL07(NetworkL07Model *model, const char* name, xbt_dict_t props,
471 tmgr_trace_t bw_trace,
473 tmgr_trace_t lat_trace,
474 e_surf_resource_state_t state_initial,
475 tmgr_trace_t state_trace,
476 e_surf_link_sharing_policy_t policy)
477 : Link(model, name, props, lmm_constraint_new(ptask_maxmin_system, this, bw_initial), history, state_trace)
479 m_bwCurrent = bw_initial;
481 p_bwEvent = tmgr_history_add_trace(history, bw_trace, 0.0, 0, this);
483 setState(state_initial);
484 m_latCurrent = lat_initial;
487 p_latEvent = tmgr_history_add_trace(history, lat_trace, 0.0, 0, this);
489 if (policy == SURF_LINK_FATPIPE)
490 lmm_constraint_shared(getConstraint());
493 bool CpuL07::isUsed(){
494 return lmm_constraint_used(ptask_maxmin_system, getConstraint());
497 bool LinkL07::isUsed(){
498 return lmm_constraint_used(ptask_maxmin_system, getConstraint());
501 void CpuL07::updateState(tmgr_trace_event_t event_type, double value, double /*date*/){
502 XBT_DEBUG("Updating cpu %s (%p) with value %g", getName(), this, value);
503 if (event_type == p_powerEvent) {
504 m_powerScale = value;
505 lmm_update_constraint_bound(ptask_maxmin_system, getConstraint(), m_powerPeak * m_powerScale);
506 if (tmgr_trace_event_free(event_type))
508 } else if (event_type == p_stateEvent) {
510 setState(SURF_RESOURCE_ON);
512 setState(SURF_RESOURCE_OFF);
513 if (tmgr_trace_event_free(event_type))
516 XBT_CRITICAL("Unknown event ! \n");
522 void LinkL07::updateState(tmgr_trace_event_t event_type, double value, double date) {
523 XBT_DEBUG("Updating link %s (%p) with value=%f for date=%g", getName(), this, value, date);
524 if (event_type == p_bwEvent) {
525 updateBandwidth(value, date);
526 if (tmgr_trace_event_free(event_type))
528 } else if (event_type == p_latEvent) {
529 updateLatency(value, date);
530 if (tmgr_trace_event_free(event_type))
532 } else if (event_type == p_stateEvent) {
534 setState(SURF_RESOURCE_ON);
536 setState(SURF_RESOURCE_OFF);
537 if (tmgr_trace_event_free(event_type))
540 XBT_CRITICAL("Unknown event ! \n");
546 e_surf_resource_state_t HostL07::getState() {
547 return p_cpu->getState();
550 Action *HostL07::execute(double size)
552 sg_host_t*host_list = xbt_new0(sg_host_t, 1);
553 double *flops_amount = xbt_new0(double, 1);
554 double *bytes_amount = xbt_new0(double, 1);
556 host_list[0] = sg_host_by_name(getName());
557 flops_amount[0] = size;
559 return static_cast<HostL07Model*>(getModel())->executeParallelTask(1, host_list,
564 Action *HostL07::sleep(double duration)
566 L07Action *action = NULL;
568 XBT_IN("(%s,%g)", getName(), duration);
570 action = static_cast<L07Action*>(execute(1.0));
571 action->m_maxDuration = duration;
572 action->m_suspended = 2;
573 lmm_update_variable_weight(ptask_maxmin_system, action->getVariable(), 0.0);
579 double LinkL07::getBandwidth()
584 void LinkL07::updateBandwidth(double value, double date)
587 lmm_update_constraint_bound(ptask_maxmin_system, getConstraint(), m_bwCurrent);
590 double LinkL07::getLatency()
595 void LinkL07::updateLatency(double value, double date)
597 lmm_variable_t var = NULL;
599 lmm_element_t elem = NULL;
601 m_latCurrent = value;
602 while ((var = lmm_get_var_from_cnst(ptask_maxmin_system, getConstraint(), &elem))) {
603 action = static_cast<L07Action*>(lmm_variable_id(var));
604 action->updateBound();
609 bool LinkL07::isShared()
611 return lmm_constraint_is_shared(getConstraint());
618 L07Action::~L07Action(){
619 free(p_communicationAmount);
620 free(p_computationAmount);
623 void L07Action::updateBound()
625 double lat_current = 0.0;
626 double lat_bound = -1.0;
629 int hostNb = p_edgeList->size();
631 for (i = 0; i < hostNb; i++) {
632 for (j = 0; j < hostNb; j++) {
633 xbt_dynar_t route=NULL;
635 if (p_communicationAmount[i * hostNb + j] > 0) {
637 routing_platf->getRouteAndLatency((*p_edgeList)[i], (*p_edgeList)[j],
640 lat_current = MAX(lat_current, lat * p_communicationAmount[i * hostNb + j]);
644 lat_bound = sg_tcp_gamma / (2.0 * lat_current);
645 XBT_DEBUG("action (%p) : lat_bound = %g", this, lat_bound);
646 if ((m_latency == 0.0) && (m_suspended == 0)) {
648 lmm_update_variable_bound(ptask_maxmin_system, getVariable(), lat_bound);
650 lmm_update_variable_bound(ptask_maxmin_system, getVariable(), min(m_rate, lat_bound));
654 int L07Action::unref()
658 if (action_hook.is_linked())
659 p_stateSet->erase(p_stateSet->iterator_to(*this));
661 lmm_variable_free(ptask_maxmin_system, getVariable());
668 void L07Action::cancel()
670 setState(SURF_ACTION_FAILED);
674 void L07Action::suspend()
676 XBT_IN("(%p))", this);
677 if (m_suspended != 2) {
679 lmm_update_variable_weight(ptask_maxmin_system, getVariable(), 0.0);
684 void L07Action::resume()
686 XBT_IN("(%p)", this);
687 if (m_suspended != 2) {
688 lmm_update_variable_weight(ptask_maxmin_system, getVariable(), 1.0);
694 bool L07Action::isSuspended()
696 return m_suspended == 1;
699 void L07Action::setMaxDuration(double duration)
700 { /* FIXME: should inherit */
701 XBT_IN("(%p,%g)", this, duration);
702 m_maxDuration = duration;
706 void L07Action::setPriority(double priority)
707 { /* FIXME: should inherit */
708 XBT_IN("(%p,%g)", this, priority);
709 m_priority = priority;
713 double L07Action::getRemains()
715 XBT_IN("(%p)", this);