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. */
11 #include "host_ptask_L07.hpp"
13 #include "cpu_interface.hpp"
14 #include "surf_routing.hpp"
16 #include "src/surf/platform.hpp"
18 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_host);
19 XBT_LOG_EXTERNAL_CATEGORY(xbt_cfg);
21 /**************************************/
22 /*** Resource Creation & Destruction **/
23 /**************************************/
25 static void ptask_netlink_parse_init(sg_platf_link_cbarg_t link)
27 netlink_parse_init(link);
28 current_property_set = NULL;
31 void surf_host_model_init_ptask_L07(void)
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");
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();
43 surf_host_model = new simgrid::surf::HostL07Model();
44 xbt_dynar_push(all_existing_models, &surf_host_model);
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);
56 routing_model_create(surf_network_model->createLink("__loopback__",
60 SURF_LINK_FATPIPE, NULL));
63 HostL07Model::~HostL07Model() {
64 delete surf_cpu_model_pm;
65 delete surf_network_model;
68 CpuL07Model::CpuL07Model(HostL07Model *hmodel,lmm_system_t sys)
74 CpuL07Model::~CpuL07Model() {
75 surf_cpu_model_pm = NULL;
76 p_maxminSystem = NULL; // Avoid multi-free
78 NetworkL07Model::NetworkL07Model(HostL07Model *hmodel, lmm_system_t sys)
84 NetworkL07Model::~NetworkL07Model()
86 surf_network_model = NULL;
87 p_maxminSystem = NULL; // Avoid multi-free
91 double HostL07Model::shareResources(double /*now*/)
95 ActionList *running_actions = getRunningActionSet();
96 double min = this->shareResourcesMaxMin(running_actions,
100 for(ActionList::iterator it(running_actions->begin()), itend(running_actions->end())
101 ; it != itend ; ++it) {
102 action = static_cast<L07Action*>(&*it);
103 if (action->m_latency > 0) {
105 min = action->m_latency;
106 XBT_DEBUG("Updating min (value) with %p (start %f): %f", action,
107 action->getStartTime(), min);
108 } else if (action->m_latency < min) {
109 min = action->m_latency;
110 XBT_DEBUG("Updating min (latency) with %p (start %f): %f", action,
111 action->getStartTime(), min);
116 XBT_DEBUG("min value : %f", min);
121 void HostL07Model::updateActionsState(double /*now*/, double delta) {
124 ActionList *actionSet = getRunningActionSet();
126 for(ActionList::iterator it = actionSet->begin(), itNext = it
127 ; it != actionSet->end()
130 action = static_cast<L07Action*>(&*it);
131 if (action->m_latency > 0) {
132 if (action->m_latency > delta) {
133 double_update(&(action->m_latency), delta, sg_surf_precision);
135 action->m_latency = 0.0;
137 if ((action->m_latency == 0.0) && (action->isSuspended() == 0)) {
138 action->updateBound();
139 lmm_update_variable_weight(p_maxminSystem, action->getVariable(), 1.0);
142 XBT_DEBUG("Action (%p) : remains (%g) updated by %g.",
143 action, action->getRemains(), lmm_variable_getvalue(action->getVariable()) * delta);
144 action->updateRemains(lmm_variable_getvalue(action->getVariable()) * delta);
146 if (action->getMaxDuration() != NO_MAX_DURATION)
147 action->updateMaxDuration(delta);
149 XBT_DEBUG("Action (%p) : remains (%g).", action, action->getRemains());
151 /* In the next if cascade, the action can be finished either because:
152 * - The amount of remaining work reached 0
153 * - The max duration was reached
154 * If it's not done, it may have failed.
157 if ((action->getRemains() <= 0) &&
158 (lmm_get_variable_weight(action->getVariable()) > 0)) {
160 action->setState(SURF_ACTION_DONE);
161 } else if ((action->getMaxDuration() != NO_MAX_DURATION) &&
162 (action->getMaxDuration() <= 0)) {
164 action->setState(SURF_ACTION_DONE);
166 /* Need to check that none of the model has failed */
167 lmm_constraint_t cnst = NULL;
170 while ((cnst = lmm_get_cnst_from_var(p_maxminSystem, action->getVariable(), i++))) {
171 void *constraint_id = lmm_constraint_id(cnst);
173 if (static_cast<Host*>(constraint_id)->isOff()) {
174 XBT_DEBUG("Action (%p) Failed!!", action);
176 action->setState(SURF_ACTION_FAILED);
185 Action *HostL07Model::executeParallelTask(int host_nb, sg_host_t *host_list,
186 double *flops_amount, double *bytes_amount,
188 return new L07Action(this, host_nb, host_list, flops_amount, bytes_amount, rate);
192 L07Action::L07Action(Model *model, int host_nb,
194 double *flops_amount,
195 double *bytes_amount,
197 : CpuAction(model, 1, 0)
201 int nb_used_host = 0; /* Only the hosts with something to compute (>0 flops) are counted) */
202 double latency = 0.0;
204 xbt_dict_t ptask_parallel_task_link_set = xbt_dict_new_homogeneous(NULL);
206 this->p_netcardList->reserve(host_nb);
207 for (int i = 0; i<host_nb; i++)
208 this->p_netcardList->push_back(host_list[i]->pimpl_netcard);
210 /* Compute the number of affected resources... */
211 for (int i = 0; i < host_nb; i++) {
212 for (int j = 0; j < host_nb; j++) {
213 xbt_dynar_t route=NULL;
215 if (bytes_amount[i * host_nb + j] > 0) {
221 routing_platf->getRouteAndLatency((*this->p_netcardList)[i], (*this->p_netcardList)[j],
223 latency = MAX(latency, lat);
225 xbt_dynar_foreach(route, cpt, _link) {
226 link = static_cast<LinkL07*>(_link);
227 xbt_dict_set(ptask_parallel_task_link_set, link->getName(), link, NULL);
233 nb_link = xbt_dict_length(ptask_parallel_task_link_set);
234 xbt_dict_free(&ptask_parallel_task_link_set);
236 for (int i = 0; i < host_nb; i++)
237 if (flops_amount[i] > 0)
240 XBT_DEBUG("Creating a parallel task (%p) with %d cpus and %d links.",
241 this, host_nb, nb_link);
242 this->p_computationAmount = flops_amount;
243 this->p_communicationAmount = bytes_amount;
244 this->m_latency = latency;
247 this->p_variable = lmm_variable_new(model->getMaxminSystem(), this, 1.0,
248 (rate > 0 ? rate : -1.0),
251 if (this->m_latency > 0)
252 lmm_update_variable_weight(model->getMaxminSystem(), this->getVariable(), 0.0);
254 for (int i = 0; i < host_nb; i++)
255 lmm_expand(model->getMaxminSystem(),
256 host_list[i]->pimpl_cpu->getConstraint(),
257 this->getVariable(), flops_amount[i]);
259 for (int i = 0; i < host_nb; i++) {
260 for (int j = 0; j < host_nb; j++) {
263 xbt_dynar_t route=NULL;
264 if (bytes_amount[i * host_nb + j] == 0.0)
267 routing_platf->getRouteAndLatency((*this->p_netcardList)[i], (*this->p_netcardList)[j],
270 xbt_dynar_foreach(route, cpt, _link) {
271 LinkL07 *link = static_cast<LinkL07*>(_link);
272 lmm_expand_add(model->getMaxminSystem(), link->getConstraint(),
274 bytes_amount[i * host_nb + j]);
279 if (nb_link + nb_used_host == 0) {
281 this->setRemains(0.0);
285 Action *NetworkL07Model::communicate(NetCard *src, NetCard *dst,
286 double size, double rate)
288 sg_host_t*host_list = xbt_new0(sg_host_t, 2);
289 double *flops_amount = xbt_new0(double, 2);
290 double *bytes_amount = xbt_new0(double, 4);
293 host_list[0] = sg_host_by_name(src->getName());
294 host_list[1] = sg_host_by_name(dst->getName());
295 bytes_amount[1] = size;
297 res = p_hostModel->executeParallelTask(2, host_list,
304 Cpu *CpuL07Model::createCpu(simgrid::s4u::Host *host, xbt_dynar_t powerPeakList,
305 int pstate, double power_scale,
306 tmgr_trace_t power_trace, int core,
308 tmgr_trace_t state_trace)
310 CpuL07 *cpu = new CpuL07(this, host, powerPeakList, pstate, power_scale, power_trace,
311 core, initiallyOn, state_trace);
315 Link* NetworkL07Model::createLink(const char *name,
317 tmgr_trace_t bw_trace,
319 tmgr_trace_t lat_trace,
321 tmgr_trace_t state_trace,
322 e_surf_link_sharing_policy_t policy,
323 xbt_dict_t properties)
325 xbt_assert(!Link::byName(name),
326 "Link '%s' declared several times in the platform file.", name);
328 Link* link = new LinkL07(this, name, properties,
329 bw_initial, bw_trace,
330 lat_initial, lat_trace,
331 initiallyOn, state_trace,
333 Link::onCreation(link);
337 void HostL07Model::addTraces()
339 xbt_dict_cursor_t cursor = NULL;
340 char *trace_name, *elm;
342 if (!trace_connect_list_host_avail)
345 /* Connect traces relative to cpu */
346 xbt_dict_foreach(trace_connect_list_host_avail, cursor, trace_name, elm) {
347 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
348 CpuL07 *host = static_cast<CpuL07*>(sg_host_by_name(elm)->pimpl_cpu);
350 xbt_assert(host, "Host %s undefined", elm);
351 xbt_assert(trace, "Trace %s undefined", trace_name);
353 host->p_stateEvent = tmgr_history_add_trace(history, trace, 0.0, 0, host);
356 xbt_dict_foreach(trace_connect_list_power, cursor, trace_name, elm) {
357 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
358 CpuL07 *host = static_cast<CpuL07*>(sg_host_by_name(elm)->pimpl_cpu);
360 xbt_assert(host, "Host %s undefined", elm);
361 xbt_assert(trace, "Trace %s undefined", trace_name);
363 host->p_speedEvent = tmgr_history_add_trace(history, trace, 0.0, 0, host);
366 /* Connect traces relative to network */
367 xbt_dict_foreach(trace_connect_list_link_avail, cursor, trace_name, elm) {
368 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
369 LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
371 xbt_assert(link, "Link %s undefined", elm);
372 xbt_assert(trace, "Trace %s undefined", trace_name);
374 link->p_stateEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
377 xbt_dict_foreach(trace_connect_list_bandwidth, cursor, trace_name, elm) {
378 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
379 LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
381 xbt_assert(link, "Link %s undefined", elm);
382 xbt_assert(trace, "Trace %s undefined", trace_name);
384 link->p_bwEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
387 xbt_dict_foreach(trace_connect_list_latency, cursor, trace_name, elm) {
388 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
389 LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
391 xbt_assert(link, "Link %s undefined", elm);
392 xbt_assert(trace, "Trace %s undefined", trace_name);
394 link->p_latEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
402 CpuL07::CpuL07(CpuL07Model *model, simgrid::s4u::Host *host,
403 xbt_dynar_t speedPeakList, int pstate,
404 double speedScale, tmgr_trace_t speedTrace,
405 int core, int initiallyOn, tmgr_trace_t state_trace)
406 : Cpu(model, host, speedPeakList, pstate,
407 core, xbt_dynar_get_as(speedPeakList,pstate,double), speedScale, initiallyOn)
409 p_constraint = lmm_constraint_new(model->getMaxminSystem(), this, xbt_dynar_get_as(speedPeakList,pstate,double) * speedScale);
412 p_speedEvent = tmgr_history_add_trace(history, speedTrace, 0.0, 0, this);
417 p_stateEvent = tmgr_history_add_trace(history, state_trace, 0.0, 0, this);
424 LinkL07::LinkL07(NetworkL07Model *model, const char* name, xbt_dict_t props,
426 tmgr_trace_t bw_trace,
428 tmgr_trace_t lat_trace,
430 tmgr_trace_t state_trace,
431 e_surf_link_sharing_policy_t policy)
432 : Link(model, name, props, lmm_constraint_new(model->getMaxminSystem(), this, bw_initial), history, state_trace)
434 m_bwCurrent = bw_initial;
436 p_bwEvent = tmgr_history_add_trace(history, bw_trace, 0.0, 0, this);
442 m_latCurrent = lat_initial;
445 p_latEvent = tmgr_history_add_trace(history, lat_trace, 0.0, 0, this);
447 if (policy == SURF_LINK_FATPIPE)
448 lmm_constraint_shared(getConstraint());
451 Action *CpuL07::execute(double size)
453 sg_host_t*host_list = xbt_new0(sg_host_t, 1);
454 double *flops_amount = xbt_new0(double, 1);
455 double *bytes_amount = xbt_new0(double, 1);
457 host_list[0] = getHost();
458 flops_amount[0] = size;
460 return static_cast<CpuL07Model*>(getModel())
462 ->executeParallelTask( 1, host_list, flops_amount, bytes_amount, -1);
465 Action *CpuL07::sleep(double duration)
467 L07Action *action = NULL;
469 XBT_IN("(%s,%g)", getName(), duration);
471 action = static_cast<L07Action*>(execute(1.0));
472 action->m_maxDuration = duration;
473 action->m_suspended = 2;
474 lmm_update_variable_weight(getModel()->getMaxminSystem(), action->getVariable(), 0.0);
480 bool CpuL07::isUsed(){
481 return lmm_constraint_used(getModel()->getMaxminSystem(), getConstraint());
484 /** @brief take into account changes of speed (either load or max) */
485 void CpuL07::onSpeedChange() {
486 lmm_variable_t var = NULL;
487 lmm_element_t elem = NULL;
489 lmm_update_constraint_bound(getModel()->getMaxminSystem(), getConstraint(), m_speedPeak * m_speedScale);
490 while ((var = lmm_get_var_from_cnst
491 (getModel()->getMaxminSystem(), getConstraint(), &elem))) {
492 Action *action = static_cast<Action*>(lmm_variable_id(var));
494 lmm_update_variable_bound(getModel()->getMaxminSystem(),
495 action->getVariable(),
496 m_speedScale * m_speedPeak);
499 Cpu::onSpeedChange();
503 bool LinkL07::isUsed(){
504 return lmm_constraint_used(getModel()->getMaxminSystem(), getConstraint());
507 void CpuL07::updateState(tmgr_trace_iterator_t event_type, double value, double /*date*/){
508 XBT_DEBUG("Updating cpu %s (%p) with value %g", getName(), this, value);
509 if (event_type == p_speedEvent) {
510 m_speedScale = value;
512 if (tmgr_trace_event_free(event_type))
514 } else if (event_type == p_stateEvent) {
519 if (tmgr_trace_event_free(event_type))
522 XBT_CRITICAL("Unknown event ! \n");
528 void LinkL07::updateState(tmgr_trace_iterator_t event_type, double value, double date) {
529 XBT_DEBUG("Updating link %s (%p) with value=%f for date=%g", getName(), this, value, date);
530 if (event_type == p_bwEvent) {
531 updateBandwidth(value, date);
532 if (tmgr_trace_event_free(event_type))
534 } else if (event_type == p_latEvent) {
535 updateLatency(value, date);
536 if (tmgr_trace_event_free(event_type))
538 } else if (event_type == p_stateEvent) {
543 if (tmgr_trace_event_free(event_type))
546 XBT_CRITICAL("Unknown event ! \n");
552 double LinkL07::getBandwidth()
557 void LinkL07::updateBandwidth(double value, double date)
560 lmm_update_constraint_bound(getModel()->getMaxminSystem(), getConstraint(), m_bwCurrent);
563 void LinkL07::updateLatency(double value, double date)
565 lmm_variable_t var = NULL;
567 lmm_element_t elem = NULL;
569 m_latCurrent = value;
570 while ((var = lmm_get_var_from_cnst(getModel()->getMaxminSystem(), getConstraint(), &elem))) {
571 action = static_cast<L07Action*>(lmm_variable_id(var));
572 action->updateBound();
580 L07Action::~L07Action(){
581 free(p_communicationAmount);
582 free(p_computationAmount);
585 void L07Action::updateBound()
587 double lat_current = 0.0;
588 double lat_bound = -1.0;
591 int hostNb = p_netcardList->size();
593 for (i = 0; i < hostNb; i++) {
594 for (j = 0; j < hostNb; j++) {
595 xbt_dynar_t route=NULL;
597 if (p_communicationAmount[i * hostNb + j] > 0) {
599 routing_platf->getRouteAndLatency((*p_netcardList)[i], (*p_netcardList)[j],
602 lat_current = MAX(lat_current, lat * p_communicationAmount[i * hostNb + j]);
606 lat_bound = sg_tcp_gamma / (2.0 * lat_current);
607 XBT_DEBUG("action (%p) : lat_bound = %g", this, lat_bound);
608 if ((m_latency == 0.0) && (m_suspended == 0)) {
610 lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(), lat_bound);
612 lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(),
613 std::min(m_rate, lat_bound));
617 int L07Action::unref()
621 if (action_hook.is_linked())
622 p_stateSet->erase(p_stateSet->iterator_to(*this));
624 lmm_variable_free(getModel()->getMaxminSystem(), getVariable());
631 void L07Action::suspend()
633 XBT_IN("(%p))", this);
634 if (m_suspended != 2) {
636 lmm_update_variable_weight(getModel()->getMaxminSystem(), getVariable(), 0.0);
641 void L07Action::resume()
643 XBT_IN("(%p)", this);
644 if (m_suspended != 2) {
645 lmm_update_variable_weight(getModel()->getMaxminSystem(), getVariable(), 1.0);
651 void L07Action::setMaxDuration(double duration)
652 { /* FIXME: should inherit */
653 XBT_IN("(%p,%g)", this, duration);
654 m_maxDuration = duration;
658 void L07Action::setPriority(double priority)
659 { /* FIXME: should inherit */
660 XBT_IN("(%p,%g)", this, priority);
661 m_priority = priority;
665 double L07Action::getRemains()
667 XBT_IN("(%p)", this);