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"
17 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_host);
19 /**************************************/
20 /*** Resource Creation & Destruction **/
21 /**************************************/
23 static void ptask_netlink_parse_init(sg_platf_link_cbarg_t link)
25 netlink_parse_init(link);
26 current_property_set = NULL;
29 void surf_host_model_init_ptask_L07(void)
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");
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);
39 surf_host_model = new simgrid::surf::HostL07Model();
40 xbt_dynar_push(all_existing_models, &surf_host_model);
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);
52 routing_model_create(surf_network_model->createLink("__loopback__",
55 SURF_RESOURCE_ON, NULL,
56 SURF_LINK_FATPIPE, NULL));
59 HostL07Model::~HostL07Model() {
60 delete surf_cpu_model_pm;
61 delete surf_network_model;
64 CpuL07Model::CpuL07Model(HostL07Model *hmodel,lmm_system_t sys)
70 CpuL07Model::~CpuL07Model() {
71 surf_cpu_model_pm = NULL;
72 p_maxminSystem = NULL; // Avoid multi-free
74 NetworkL07Model::NetworkL07Model(HostL07Model *hmodel, lmm_system_t sys)
80 NetworkL07Model::~NetworkL07Model()
82 surf_network_model = NULL;
83 p_maxminSystem = NULL; // Avoid multi-free
87 double HostL07Model::shareResources(double /*now*/)
91 ActionList *running_actions = getRunningActionSet();
92 double min = this->shareResourcesMaxMin(running_actions,
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) {
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);
112 XBT_DEBUG("min value : %f", min);
117 void HostL07Model::updateActionsState(double /*now*/, double delta) {
120 ActionList *actionSet = getRunningActionSet();
122 for(ActionList::iterator it = actionSet->begin(), itNext = it
123 ; it != actionSet->end()
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);
131 action->m_latency = 0.0;
133 if ((action->m_latency == 0.0) && (action->isSuspended() == 0)) {
134 action->updateBound();
135 lmm_update_variable_weight(p_maxminSystem, action->getVariable(), 1.0);
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);
142 if (action->getMaxDuration() != NO_MAX_DURATION)
143 action->updateMaxDuration(delta);
145 XBT_DEBUG("Action (%p) : remains (%g).", action, action->getRemains());
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.
153 if ((action->getRemains() <= 0) &&
154 (lmm_get_variable_weight(action->getVariable()) > 0)) {
156 action->setState(SURF_ACTION_DONE);
157 } else if ((action->getMaxDuration() != NO_MAX_DURATION) &&
158 (action->getMaxDuration() <= 0)) {
160 action->setState(SURF_ACTION_DONE);
162 /* Need to check that none of the model has failed */
163 lmm_constraint_t cnst = NULL;
166 while ((cnst = lmm_get_cnst_from_var(p_maxminSystem, action->getVariable(), i++))) {
167 void *constraint_id = lmm_constraint_id(cnst);
169 if (static_cast<Host*>(constraint_id)->getState() == SURF_RESOURCE_OFF) {
170 XBT_DEBUG("Action (%p) Failed!!", action);
172 action->setState(SURF_ACTION_FAILED);
181 Action *HostL07Model::executeParallelTask(int host_nb,
183 double *flops_amount,
184 double *bytes_amount,
187 L07Action *action = new L07Action(this, 1, 0);
190 int nb_used_host = 0; /* Only the hosts with something to compute (>0 flops) are counted) */
191 double latency = 0.0;
193 xbt_dict_t ptask_parallel_task_link_set = xbt_dict_new_homogeneous(NULL);
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]));
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;
204 if (bytes_amount[i * host_nb + j] > 0) {
210 routing_platf->getRouteAndLatency((*action->p_edgeList)[i], (*action->p_edgeList)[j],
212 latency = MAX(latency, lat);
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);
222 nb_link = xbt_dict_length(ptask_parallel_task_link_set);
223 xbt_dict_free(&ptask_parallel_task_link_set);
225 for (int i = 0; i < host_nb; i++)
226 if (flops_amount[i] > 0)
229 XBT_DEBUG("Creating a parallel task (%p) with %d cpus and %d links.",
230 action, host_nb, nb_link);
231 action->m_suspended = 0; /* valgrind seems to want it despite the calloc... */
232 action->p_computationAmount = flops_amount;
233 action->p_communicationAmount = bytes_amount;
234 action->m_latency = latency;
235 action->m_rate = rate;
237 action->p_variable = lmm_variable_new(p_maxminSystem, action, 1.0,
238 (rate > 0 ? rate : -1.0),
241 if (action->m_latency > 0)
242 lmm_update_variable_weight(p_maxminSystem, action->getVariable(), 0.0);
244 for (int i = 0; i < host_nb; i++)
245 lmm_expand(p_maxminSystem,
246 sg_host_surfcpu(host_list[i])->getConstraint(),
247 action->getVariable(), flops_amount[i]);
249 for (int i = 0; i < host_nb; i++) {
250 for (int j = 0; j < host_nb; j++) {
253 xbt_dynar_t route=NULL;
254 if (bytes_amount[i * host_nb + j] == 0.0)
257 routing_platf->getRouteAndLatency((*action->p_edgeList)[i], (*action->p_edgeList)[j],
260 xbt_dynar_foreach(route, cpt, _link) {
261 LinkL07 *link = static_cast<LinkL07*>(_link);
262 lmm_expand_add(p_maxminSystem, link->getConstraint(),
263 action->getVariable(),
264 bytes_amount[i * host_nb + j]);
269 if (nb_link + nb_used_host == 0) {
270 action->setCost(1.0);
271 action->setRemains(0.0);
277 Action *NetworkL07Model::communicate(RoutingEdge *src, RoutingEdge *dst,
278 double size, double rate)
280 sg_host_t*host_list = xbt_new0(sg_host_t, 2);
281 double *flops_amount = xbt_new0(double, 2);
282 double *bytes_amount = xbt_new0(double, 4);
285 host_list[0] = sg_host_by_name(src->getName());
286 host_list[1] = sg_host_by_name(dst->getName());
287 bytes_amount[1] = size;
289 res = p_hostModel->executeParallelTask(2, host_list,
296 Cpu *CpuL07Model::createCpu(simgrid::Host *host, xbt_dynar_t powerPeakList,
297 int pstate, double power_scale,
298 tmgr_trace_t power_trace, int core,
299 e_surf_resource_state_t state_initial,
300 tmgr_trace_t state_trace)
302 CpuL07 *cpu = new CpuL07(this, host, powerPeakList, pstate, power_scale, power_trace,
303 core, state_initial, state_trace);
307 Link* NetworkL07Model::createLink(const char *name,
309 tmgr_trace_t bw_trace,
311 tmgr_trace_t lat_trace,
312 e_surf_resource_state_t state_initial,
313 tmgr_trace_t state_trace,
314 e_surf_link_sharing_policy_t policy,
315 xbt_dict_t properties)
317 xbt_assert(!Link::byName(name),
318 "Link '%s' declared several times in the platform file.", name);
320 Link* link = new LinkL07(this, name, properties,
321 bw_initial, bw_trace,
322 lat_initial, lat_trace,
323 state_initial, state_trace,
325 Link::onCreation(link);
329 void HostL07Model::addTraces()
331 xbt_dict_cursor_t cursor = NULL;
332 char *trace_name, *elm;
334 if (!trace_connect_list_host_avail)
337 /* Connect traces relative to cpu */
338 xbt_dict_foreach(trace_connect_list_host_avail, cursor, trace_name, elm) {
339 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
340 CpuL07 *host = static_cast<CpuL07*>(sg_host_surfcpu(sg_host_by_name(elm)));
342 xbt_assert(host, "Host %s undefined", elm);
343 xbt_assert(trace, "Trace %s undefined", trace_name);
345 host->p_stateEvent = tmgr_history_add_trace(history, trace, 0.0, 0, host);
348 xbt_dict_foreach(trace_connect_list_power, cursor, trace_name, elm) {
349 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
350 CpuL07 *host = static_cast<CpuL07*>(sg_host_surfcpu(sg_host_by_name(elm)));
352 xbt_assert(host, "Host %s undefined", elm);
353 xbt_assert(trace, "Trace %s undefined", trace_name);
355 host->p_speedEvent = tmgr_history_add_trace(history, trace, 0.0, 0, host);
358 /* Connect traces relative to network */
359 xbt_dict_foreach(trace_connect_list_link_avail, cursor, trace_name, elm) {
360 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
361 LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
363 xbt_assert(link, "Link %s undefined", elm);
364 xbt_assert(trace, "Trace %s undefined", trace_name);
366 link->p_stateEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
369 xbt_dict_foreach(trace_connect_list_bandwidth, cursor, trace_name, elm) {
370 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
371 LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
373 xbt_assert(link, "Link %s undefined", elm);
374 xbt_assert(trace, "Trace %s undefined", trace_name);
376 link->p_bwEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
379 xbt_dict_foreach(trace_connect_list_latency, cursor, trace_name, elm) {
380 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
381 LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
383 xbt_assert(link, "Link %s undefined", elm);
384 xbt_assert(trace, "Trace %s undefined", trace_name);
386 link->p_latEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
394 CpuL07::CpuL07(CpuL07Model *model, simgrid::Host *host,
395 xbt_dynar_t speedPeakList, int pstate,
396 double speedScale, tmgr_trace_t speedTrace,
397 int core, e_surf_resource_state_t state_initial, tmgr_trace_t state_trace)
398 : Cpu(model, host, speedPeakList, pstate,
399 core, xbt_dynar_get_as(speedPeakList,pstate,double), speedScale, state_initial)
401 xbt_assert(m_speedScale > 0, "Power has to be >0");
402 p_constraint = lmm_constraint_new(model->getMaxminSystem(), this, xbt_dynar_get_as(speedPeakList,pstate,double) * speedScale);
405 p_speedEvent = tmgr_history_add_trace(history, speedTrace, 0.0, 0, this);
410 p_stateEvent = tmgr_history_add_trace(history, state_trace, 0.0, 0, this);
417 LinkL07::LinkL07(NetworkL07Model *model, const char* name, xbt_dict_t props,
419 tmgr_trace_t bw_trace,
421 tmgr_trace_t lat_trace,
422 e_surf_resource_state_t state_initial,
423 tmgr_trace_t state_trace,
424 e_surf_link_sharing_policy_t policy)
425 : Link(model, name, props, lmm_constraint_new(model->getMaxminSystem(), this, bw_initial), history, state_trace)
427 m_bwCurrent = bw_initial;
429 p_bwEvent = tmgr_history_add_trace(history, bw_trace, 0.0, 0, this);
431 setState(state_initial);
432 m_latCurrent = lat_initial;
435 p_latEvent = tmgr_history_add_trace(history, lat_trace, 0.0, 0, this);
437 if (policy == SURF_LINK_FATPIPE)
438 lmm_constraint_shared(getConstraint());
441 Action *CpuL07::execute(double size)
443 sg_host_t*host_list = xbt_new0(sg_host_t, 1);
444 double *flops_amount = xbt_new0(double, 1);
445 double *bytes_amount = xbt_new0(double, 1);
447 host_list[0] = getHost();
448 flops_amount[0] = size;
450 return static_cast<CpuL07Model*>(getModel())
452 ->executeParallelTask( 1, host_list, flops_amount, bytes_amount, -1);
455 Action *CpuL07::sleep(double duration)
457 L07Action *action = NULL;
459 XBT_IN("(%s,%g)", getName(), duration);
461 action = static_cast<L07Action*>(execute(1.0));
462 action->m_maxDuration = duration;
463 action->m_suspended = 2;
464 lmm_update_variable_weight(getModel()->getMaxminSystem(), action->getVariable(), 0.0);
470 bool CpuL07::isUsed(){
471 return lmm_constraint_used(getModel()->getMaxminSystem(), getConstraint());
474 bool LinkL07::isUsed(){
475 return lmm_constraint_used(getModel()->getMaxminSystem(), getConstraint());
478 void CpuL07::updateState(tmgr_trace_event_t event_type, double value, double /*date*/){
479 XBT_DEBUG("Updating cpu %s (%p) with value %g", getName(), this, value);
480 if (event_type == p_speedEvent) {
481 m_speedScale = value;
482 lmm_update_constraint_bound(getModel()->getMaxminSystem(), getConstraint(), m_speedPeak * m_speedScale);
483 if (tmgr_trace_event_free(event_type))
485 } else if (event_type == p_stateEvent) {
487 setState(SURF_RESOURCE_ON);
489 setState(SURF_RESOURCE_OFF);
490 if (tmgr_trace_event_free(event_type))
493 XBT_CRITICAL("Unknown event ! \n");
499 void LinkL07::updateState(tmgr_trace_event_t event_type, double value, double date) {
500 XBT_DEBUG("Updating link %s (%p) with value=%f for date=%g", getName(), this, value, date);
501 if (event_type == p_bwEvent) {
502 updateBandwidth(value, date);
503 if (tmgr_trace_event_free(event_type))
505 } else if (event_type == p_latEvent) {
506 updateLatency(value, date);
507 if (tmgr_trace_event_free(event_type))
509 } else if (event_type == p_stateEvent) {
511 setState(SURF_RESOURCE_ON);
513 setState(SURF_RESOURCE_OFF);
514 if (tmgr_trace_event_free(event_type))
517 XBT_CRITICAL("Unknown event ! \n");
523 double LinkL07::getBandwidth()
528 void LinkL07::updateBandwidth(double value, double date)
531 lmm_update_constraint_bound(getModel()->getMaxminSystem(), getConstraint(), m_bwCurrent);
534 void LinkL07::updateLatency(double value, double date)
536 lmm_variable_t var = NULL;
538 lmm_element_t elem = NULL;
540 m_latCurrent = value;
541 while ((var = lmm_get_var_from_cnst(getModel()->getMaxminSystem(), getConstraint(), &elem))) {
542 action = static_cast<L07Action*>(lmm_variable_id(var));
543 action->updateBound();
551 L07Action::~L07Action(){
552 free(p_communicationAmount);
553 free(p_computationAmount);
556 void L07Action::updateBound()
558 double lat_current = 0.0;
559 double lat_bound = -1.0;
562 int hostNb = p_edgeList->size();
564 for (i = 0; i < hostNb; i++) {
565 for (j = 0; j < hostNb; j++) {
566 xbt_dynar_t route=NULL;
568 if (p_communicationAmount[i * hostNb + j] > 0) {
570 routing_platf->getRouteAndLatency((*p_edgeList)[i], (*p_edgeList)[j],
573 lat_current = MAX(lat_current, lat * p_communicationAmount[i * hostNb + j]);
577 lat_bound = sg_tcp_gamma / (2.0 * lat_current);
578 XBT_DEBUG("action (%p) : lat_bound = %g", this, lat_bound);
579 if ((m_latency == 0.0) && (m_suspended == 0)) {
581 lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(), lat_bound);
583 lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(),
584 std::min(m_rate, lat_bound));
588 int L07Action::unref()
592 if (action_hook.is_linked())
593 p_stateSet->erase(p_stateSet->iterator_to(*this));
595 lmm_variable_free(getModel()->getMaxminSystem(), getVariable());
602 void L07Action::cancel()
604 setState(SURF_ACTION_FAILED);
608 void L07Action::suspend()
610 XBT_IN("(%p))", this);
611 if (m_suspended != 2) {
613 lmm_update_variable_weight(getModel()->getMaxminSystem(), getVariable(), 0.0);
618 void L07Action::resume()
620 XBT_IN("(%p)", this);
621 if (m_suspended != 2) {
622 lmm_update_variable_weight(getModel()->getMaxminSystem(), getVariable(), 1.0);
628 bool L07Action::isSuspended()
630 return m_suspended == 1;
633 void L07Action::setMaxDuration(double duration)
634 { /* FIXME: should inherit */
635 XBT_IN("(%p,%g)", this, duration);
636 m_maxDuration = duration;
640 void L07Action::setPriority(double priority)
641 { /* FIXME: should inherit */
642 XBT_IN("(%p,%g)", this, priority);
643 m_priority = priority;
647 double L07Action::getRemains()
649 XBT_IN("(%p)", this);