1 /* Copyright (c) 2007-2010, 2013-2014. 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"
12 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_host);
14 static int ptask_host_count = 0;
15 static xbt_dict_t ptask_parallel_task_link_set = NULL;
16 lmm_system_t ptask_maxmin_system = NULL;
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_host_add_cb(cpu_parse_init);
37 sg_platf_host_add_cb(host_parse_init);
38 sg_platf_link_add_cb(ptask_netlink_parse_init);
39 sg_platf_postparse_add_cb(host_add_traces);
41 surf_host_model = new HostL07Model();
42 Model *model = surf_host_model;
43 xbt_dynar_push(model_list, &model);
44 xbt_dynar_push(model_list_invoke, &model);
48 HostL07Model::HostL07Model() : HostModel("Host ptask_L07") {
49 if (!ptask_maxmin_system)
50 ptask_maxmin_system = lmm_system_new(1);
51 surf_host_model = NULL;
52 surf_network_model = new NetworkL07Model();
53 surf_cpu_model_pm = new CpuL07Model();
55 routing_model_create(surf_network_model->createLink("__loopback__",
58 SURF_RESOURCE_ON, NULL,
59 SURF_LINK_FATPIPE, NULL));
60 p_cpuModel = surf_cpu_model_pm;
63 HostL07Model::~HostL07Model() {
64 xbt_dict_free(&ptask_parallel_task_link_set);
66 delete surf_cpu_model_pm;
67 delete surf_network_model;
70 if (ptask_maxmin_system) {
71 lmm_system_free(ptask_maxmin_system);
72 ptask_maxmin_system = NULL;
76 double HostL07Model::shareResources(double /*now*/)
80 ActionList *running_actions = getRunningActionSet();
81 double min = this->shareResourcesMaxMin(running_actions,
85 for(ActionList::iterator it(running_actions->begin()), itend(running_actions->end())
86 ; it != itend ; ++it) {
87 action = static_cast<L07Action*>(&*it);
88 if (action->m_latency > 0) {
90 min = action->m_latency;
91 XBT_DEBUG("Updating min (value) with %p (start %f): %f", action,
92 action->getStartTime(), min);
93 } else if (action->m_latency < min) {
94 min = action->m_latency;
95 XBT_DEBUG("Updating min (latency) with %p (start %f): %f", action,
96 action->getStartTime(), min);
101 XBT_DEBUG("min value : %f", min);
106 void HostL07Model::updateActionsState(double /*now*/, double delta) {
109 ActionList *actionSet = getRunningActionSet();
111 for(ActionList::iterator it = actionSet->begin(), itNext = it
112 ; it != actionSet->end()
115 action = static_cast<L07Action*>(&*it);
116 if (action->m_latency > 0) {
117 if (action->m_latency > delta) {
118 double_update(&(action->m_latency), delta, sg_surf_precision);
120 action->m_latency = 0.0;
122 if ((action->m_latency == 0.0) && (action->isSuspended() == 0)) {
123 action->updateBound();
124 lmm_update_variable_weight(ptask_maxmin_system, action->getVariable(), 1.0);
127 XBT_DEBUG("Action (%p) : remains (%g) updated by %g.",
128 action, action->getRemains(), lmm_variable_getvalue(action->getVariable()) * delta);
129 action->updateRemains(lmm_variable_getvalue(action->getVariable()) * delta);
131 if (action->getMaxDuration() != NO_MAX_DURATION)
132 action->updateMaxDuration(delta);
134 XBT_DEBUG("Action (%p) : remains (%g).", action, action->getRemains());
136 /* In the next if cascade, the action can be finished either because:
137 * - The amount of remaining work reached 0
138 * - The max duration was reached
139 * If it's not done, it may have failed.
142 if ((action->getRemains() <= 0) &&
143 (lmm_get_variable_weight(action->getVariable()) > 0)) {
145 action->setState(SURF_ACTION_DONE);
146 } else if ((action->getMaxDuration() != NO_MAX_DURATION) &&
147 (action->getMaxDuration() <= 0)) {
149 action->setState(SURF_ACTION_DONE);
151 /* Need to check that none of the model has failed */
152 lmm_constraint_t cnst = NULL;
155 while ((cnst = lmm_get_cnst_from_var(ptask_maxmin_system, action->getVariable(), i++))) {
156 void *constraint_id = lmm_constraint_id(cnst);
158 if (static_cast<Host*>(constraint_id)->getState() == SURF_RESOURCE_OFF) {
159 XBT_DEBUG("Action (%p) Failed!!", action);
161 action->setState(SURF_ACTION_FAILED);
170 Action *HostL07Model::executeParallelTask(int host_nb,
172 double *flops_amount,
173 double *bytes_amount,
181 double latency = 0.0;
183 if (ptask_parallel_task_link_set == NULL)
184 ptask_parallel_task_link_set = xbt_dict_new_homogeneous(NULL);
186 xbt_dict_reset(ptask_parallel_task_link_set);
188 /* Compute the number of affected resources... */
189 for (i = 0; i < host_nb; i++) {
190 for (j = 0; j < host_nb; j++) {
191 xbt_dynar_t route=NULL;
193 if (bytes_amount[i * host_nb + j] > 0) {
199 routing_platf->getRouteAndLatency(static_cast<HostL07*>(host_list[i])->p_netElm,
200 static_cast<HostL07*>(host_list[j])->p_netElm,
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 (i = 0; i < host_nb; i++)
217 if (flops_amount[i] > 0)
220 action = new L07Action(this, 1, 0);
221 XBT_DEBUG("Creating a parallel task (%p) with %d cpus and %d links.",
222 action, host_nb, nb_link);
223 action->m_suspended = 0; /* Should be useless because of the
224 calloc but it seems to help valgrind... */
225 action->m_hostNb = host_nb;
226 action->p_hostList = (Host **) host_list;
227 action->p_computationAmount = flops_amount;
228 action->p_communicationAmount = bytes_amount;
229 action->m_latency = latency;
230 action->m_rate = rate;
232 action->p_variable = lmm_variable_new(ptask_maxmin_system, action, 1.0,
233 (action->m_rate > 0) ? action->m_rate : -1.0,
236 if (action->m_latency > 0)
237 lmm_update_variable_weight(ptask_maxmin_system, action->getVariable(), 0.0);
239 for (i = 0; i < host_nb; i++)
240 lmm_expand(ptask_maxmin_system,
241 static_cast<HostL07*>(host_list[i])->p_cpu->getConstraint(),
242 action->getVariable(), flops_amount[i]);
244 for (i = 0; i < host_nb; i++) {
245 for (j = 0; j < host_nb; j++) {
249 xbt_dynar_t route=NULL;
250 if (bytes_amount[i * host_nb + j] == 0.0)
253 routing_platf->getRouteAndLatency(static_cast<HostL07*>(host_list[i])->p_netElm,
254 static_cast<HostL07*>(host_list[j])->p_netElm,
257 xbt_dynar_foreach(route, cpt, _link) {
258 link = static_cast<LinkL07*>(_link);
259 lmm_expand_add(ptask_maxmin_system, link->getConstraint(),
260 action->getVariable(),
261 bytes_amount[i * host_nb + j]);
266 if (nb_link + nb_host == 0) {
267 action->setCost(1.0);
268 action->setRemains(0.0);
274 Host *HostL07Model::createHost(const char *name)
277 sg_host_t sg_host = sg_host_by_name(name);
279 xbt_assert(!surf_host_resource_priv(sg_host),
280 "Host '%s' declared several times in the platform file.",
283 wk = new HostL07(this, name, NULL,
284 sg_host_edge(sg_host),
285 sg_host_surfcpu(sg_host));
287 xbt_lib_set(host_lib, name, SURF_HOST_LEVEL, wk);
292 Action *HostL07Model::communicate(Host *src, Host *dst,
293 double size, double rate)
295 void **host_list = xbt_new0(void *, 2);
296 double *flops_amount = xbt_new0(double, 2);
297 double *bytes_amount = xbt_new0(double, 4);
302 bytes_amount[1] = size;
304 res = executeParallelTask(2, host_list,
311 xbt_dynar_t HostL07Model::getRoute(Host *src, Host *dst)
313 xbt_dynar_t route=NULL;
314 routing_platf->getRouteAndLatency(src->p_netElm, dst->p_netElm, &route, NULL);
318 Cpu *CpuL07Model::createCpu(const char *name, xbt_dynar_t powerPeak,
319 int pstate, double power_scale,
320 tmgr_trace_t power_trace, int core,
321 e_surf_resource_state_t state_initial,
322 tmgr_trace_t state_trace,
323 xbt_dict_t cpu_properties)
325 double power_initial = xbt_dynar_get_as(powerPeak, pstate, double);
326 xbt_dynar_free(&powerPeak); // kill memory leak
327 sg_host_t sg_host = sg_host_by_name(name);
329 xbt_assert(!surf_host_resource_priv(sg_host),
330 "Host '%s' declared several times in the platform file.",
333 CpuL07 *cpu = new CpuL07(this, name, cpu_properties,
334 power_initial, power_scale, power_trace,
335 core, state_initial, state_trace);
337 sg_host_surfcpu_set(sg_host, cpu);
342 Link* NetworkL07Model::createLink(const char *name,
344 tmgr_trace_t bw_trace,
346 tmgr_trace_t lat_trace,
347 e_surf_resource_state_t state_initial,
348 tmgr_trace_t state_trace,
349 e_surf_link_sharing_policy_t policy,
350 xbt_dict_t properties)
352 xbt_assert(!Link::byName(name),
353 "Link '%s' declared several times in the platform file.", name);
355 return new LinkL07(this, name, properties,
356 bw_initial, bw_trace,
357 lat_initial, lat_trace,
358 state_initial, state_trace,
362 void HostL07Model::addTraces()
364 xbt_dict_cursor_t cursor = NULL;
365 char *trace_name, *elm;
367 if (!trace_connect_list_host_avail)
370 /* Connect traces relative to cpu */
371 xbt_dict_foreach(trace_connect_list_host_avail, cursor, trace_name, elm) {
372 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
373 CpuL07 *host = static_cast<CpuL07*>(sg_host_surfcpu(sg_host_by_name(elm)));
375 xbt_assert(host, "Host %s undefined", elm);
376 xbt_assert(trace, "Trace %s undefined", trace_name);
378 host->p_stateEvent = tmgr_history_add_trace(history, trace, 0.0, 0, host);
381 xbt_dict_foreach(trace_connect_list_power, cursor, trace_name, elm) {
382 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
383 CpuL07 *host = static_cast<CpuL07*>(sg_host_surfcpu(sg_host_by_name(elm)));
385 xbt_assert(host, "Host %s undefined", elm);
386 xbt_assert(trace, "Trace %s undefined", trace_name);
388 host->p_powerEvent = tmgr_history_add_trace(history, trace, 0.0, 0, host);
391 /* Connect traces relative to network */
392 xbt_dict_foreach(trace_connect_list_link_avail, cursor, trace_name, elm) {
393 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
394 LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
396 xbt_assert(link, "Link %s undefined", elm);
397 xbt_assert(trace, "Trace %s undefined", trace_name);
399 link->p_stateEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
402 xbt_dict_foreach(trace_connect_list_bandwidth, cursor, trace_name, elm) {
403 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
404 LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
406 xbt_assert(link, "Link %s undefined", elm);
407 xbt_assert(trace, "Trace %s undefined", trace_name);
409 link->p_bwEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
412 xbt_dict_foreach(trace_connect_list_latency, cursor, trace_name, elm) {
413 tmgr_trace_t trace = (tmgr_trace_t) xbt_dict_get_or_null(traces_set_list, trace_name);
414 LinkL07 *link = static_cast<LinkL07*>(Link::byName(elm));
416 xbt_assert(link, "Link %s undefined", elm);
417 xbt_assert(trace, "Trace %s undefined", trace_name);
419 link->p_latEvent = tmgr_history_add_trace(history, trace, 0.0, 0, link);
427 HostL07::HostL07(HostModel *model, const char* name, xbt_dict_t props, RoutingEdge *netElm, Cpu *cpu)
428 : Host(model, name, props, NULL, netElm, cpu)
432 double HostL07::getPowerPeakAt(int /*pstate_index*/)
437 int HostL07::getNbPstates()
442 void HostL07::setPstate(int /*pstate_index*/)
447 int HostL07::getPstate()
452 double HostL07::getConsumedEnergy()
457 CpuL07::CpuL07(CpuL07Model *model, const char* name, xbt_dict_t props,
458 double power_initial, double power_scale, tmgr_trace_t power_trace,
459 int core, e_surf_resource_state_t state_initial, tmgr_trace_t state_trace)
460 : Cpu(model, name, props, lmm_constraint_new(ptask_maxmin_system, this, power_initial * power_scale),
461 core, power_initial, power_scale)
463 xbt_assert(m_powerScale > 0, "Power has to be >0");
466 p_powerEvent = tmgr_history_add_trace(history, power_trace, 0.0, 0, this);
470 setState(state_initial);
472 p_stateEvent = tmgr_history_add_trace(history, state_trace, 0.0, 0, this);
475 LinkL07::LinkL07(NetworkL07Model *model, const char* name, xbt_dict_t props,
477 tmgr_trace_t bw_trace,
479 tmgr_trace_t lat_trace,
480 e_surf_resource_state_t state_initial,
481 tmgr_trace_t state_trace,
482 e_surf_link_sharing_policy_t policy)
483 : Link(model, name, props, lmm_constraint_new(ptask_maxmin_system, this, bw_initial), history, state_trace)
485 m_bwCurrent = bw_initial;
487 p_bwEvent = tmgr_history_add_trace(history, bw_trace, 0.0, 0, this);
489 setState(state_initial);
490 m_latCurrent = lat_initial;
493 p_latEvent = tmgr_history_add_trace(history, lat_trace, 0.0, 0, this);
495 if (policy == SURF_LINK_FATPIPE)
496 lmm_constraint_shared(getConstraint());
499 bool CpuL07::isUsed(){
500 return lmm_constraint_used(ptask_maxmin_system, getConstraint());
503 bool LinkL07::isUsed(){
504 return lmm_constraint_used(ptask_maxmin_system, getConstraint());
507 void CpuL07::updateState(tmgr_trace_event_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_powerEvent) {
510 m_powerScale = value;
511 lmm_update_constraint_bound(ptask_maxmin_system, getConstraint(), m_powerPeak * m_powerScale);
512 if (tmgr_trace_event_free(event_type))
514 } else if (event_type == p_stateEvent) {
516 setState(SURF_RESOURCE_ON);
518 setState(SURF_RESOURCE_OFF);
519 if (tmgr_trace_event_free(event_type))
522 XBT_CRITICAL("Unknown event ! \n");
528 void LinkL07::updateState(tmgr_trace_event_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) {
540 setState(SURF_RESOURCE_ON);
542 setState(SURF_RESOURCE_OFF);
543 if (tmgr_trace_event_free(event_type))
546 XBT_CRITICAL("Unknown event ! \n");
552 e_surf_resource_state_t HostL07::getState() {
553 return p_cpu->getState();
556 Action *HostL07::execute(double size)
558 void **host_list = xbt_new0(void *, 1);
559 double *flops_amount = xbt_new0(double, 1);
560 double *bytes_amount = xbt_new0(double, 1);
563 bytes_amount[0] = 0.0;
564 flops_amount[0] = size;
566 return static_cast<HostL07Model*>(getModel())->executeParallelTask(1, host_list,
571 Action *HostL07::sleep(double duration)
573 L07Action *action = NULL;
575 XBT_IN("(%s,%g)", getName(), duration);
577 action = static_cast<L07Action*>(execute(1.0));
578 action->m_maxDuration = duration;
579 action->m_suspended = 2;
580 lmm_update_variable_weight(ptask_maxmin_system, action->getVariable(), 0.0);
586 double LinkL07::getBandwidth()
591 void LinkL07::updateBandwidth(double value, double date)
594 lmm_update_constraint_bound(ptask_maxmin_system, getConstraint(), m_bwCurrent);
597 double LinkL07::getLatency()
602 void LinkL07::updateLatency(double value, double date)
604 lmm_variable_t var = NULL;
606 lmm_element_t elem = NULL;
608 m_latCurrent = value;
609 while ((var = lmm_get_var_from_cnst(ptask_maxmin_system, getConstraint(), &elem))) {
610 action = static_cast<L07Action*>(lmm_variable_id(var));
611 action->updateBound();
616 bool LinkL07::isShared()
618 return lmm_constraint_is_shared(getConstraint());
625 L07Action::~L07Action(){
627 free(p_communicationAmount);
628 free(p_computationAmount);
631 void L07Action::updateBound()
633 double lat_current = 0.0;
634 double lat_bound = -1.0;
637 for (i = 0; i < m_hostNb; i++) {
638 for (j = 0; j < m_hostNb; j++) {
639 xbt_dynar_t route=NULL;
641 if (p_communicationAmount[i * m_hostNb + j] > 0) {
643 routing_platf->getRouteAndLatency(static_cast<HostL07*>(((void**)p_hostList)[i])->p_netElm,
644 static_cast<HostL07*>(((void**)p_hostList)[j])->p_netElm,
647 lat_current = MAX(lat_current, lat * p_communicationAmount[i * m_hostNb + j]);
651 lat_bound = sg_tcp_gamma / (2.0 * lat_current);
652 XBT_DEBUG("action (%p) : lat_bound = %g", this, lat_bound);
653 if ((m_latency == 0.0) && (m_suspended == 0)) {
655 lmm_update_variable_bound(ptask_maxmin_system, getVariable(), lat_bound);
657 lmm_update_variable_bound(ptask_maxmin_system, getVariable(), min(m_rate, lat_bound));
661 int L07Action::unref()
665 if (actionHook::is_linked())
666 p_stateSet->erase(p_stateSet->iterator_to(*this));
668 lmm_variable_free(ptask_maxmin_system, getVariable());
675 void L07Action::cancel()
677 setState(SURF_ACTION_FAILED);
681 void L07Action::suspend()
683 XBT_IN("(%p))", this);
684 if (m_suspended != 2) {
686 lmm_update_variable_weight(ptask_maxmin_system, getVariable(), 0.0);
691 void L07Action::resume()
693 XBT_IN("(%p)", this);
694 if (m_suspended != 2) {
695 lmm_update_variable_weight(ptask_maxmin_system, getVariable(), 1.0);
701 bool L07Action::isSuspended()
703 return m_suspended == 1;
706 void L07Action::setMaxDuration(double duration)
707 { /* FIXME: should inherit */
708 XBT_IN("(%p,%g)", this, duration);
709 m_maxDuration = duration;
713 void L07Action::setPriority(double priority)
714 { /* FIXME: should inherit */
715 XBT_IN("(%p,%g)", this, priority);
716 m_priority = priority;
720 double L07Action::getRemains()
722 XBT_IN("(%p)", this);