1 /* Copyright (c) 2007-2010, 2013-2017. 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. */
10 #include <unordered_set>
12 #include "ptask_L07.hpp"
14 #include "cpu_interface.hpp"
15 #include "xbt/utility.hpp"
17 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_host);
18 XBT_LOG_EXTERNAL_CATEGORY(xbt_cfg);
20 /**************************************/
21 /*** Resource Creation & Destruction **/
22 /**************************************/
23 void surf_host_model_init_ptask_L07()
25 XBT_CINFO(xbt_cfg,"Switching to the L07 model to handle parallel tasks.");
26 xbt_assert(not surf_cpu_model_pm, "CPU model type already defined");
27 xbt_assert(not surf_network_model, "network model type already defined");
29 surf_host_model = new simgrid::surf::HostL07Model();
30 all_existing_models->push_back(surf_host_model);
37 HostL07Model::HostL07Model() : HostModel() {
38 maxminSystem_ = new simgrid::kernel::lmm::System(true /* lazy */);
39 maxminSystem_->solve_fun = &simgrid::kernel::lmm::bottleneck_solve;
40 surf_network_model = new NetworkL07Model(this,maxminSystem_);
41 surf_cpu_model_pm = new CpuL07Model(this,maxminSystem_);
44 HostL07Model::~HostL07Model()
47 maxminSystem_ = nullptr;
48 delete surf_network_model;
49 delete surf_cpu_model_pm;
52 CpuL07Model::CpuL07Model(HostL07Model* hmodel, lmm_system_t sys) : CpuModel(), hostModel_(hmodel)
57 CpuL07Model::~CpuL07Model()
59 maxminSystem_ = nullptr;
62 NetworkL07Model::NetworkL07Model(HostL07Model* hmodel, lmm_system_t sys) : NetworkModel(), hostModel_(hmodel)
65 loopback_ = NetworkL07Model::createLink("__loopback__", 498000000, 0.000015, SURF_LINK_FATPIPE);
68 NetworkL07Model::~NetworkL07Model()
70 maxminSystem_ = nullptr;
73 double HostL07Model::nextOccuringEvent(double now)
75 double min = HostModel::nextOccuringEventFull(now);
76 for (Action const& action : *getRunningActionSet()) {
77 const L07Action& net_action = static_cast<const L07Action&>(action);
78 if (net_action.latency_ > 0 && (min < 0 || net_action.latency_ < min)) {
79 min = net_action.latency_;
80 XBT_DEBUG("Updating min with %p (start %f): %f", &net_action, net_action.getStartTime(), min);
83 XBT_DEBUG("min value: %f", min);
88 void HostL07Model::updateActionsState(double /*now*/, double delta)
90 for (auto it = std::begin(*getRunningActionSet()); it != std::end(*getRunningActionSet());) {
91 L07Action& action = static_cast<L07Action&>(*it);
92 ++it; // increment iterator here since the following calls to action.finish() may invalidate it
93 if (action.latency_ > 0) {
94 if (action.latency_ > delta) {
95 double_update(&(action.latency_), delta, sg_surf_precision);
97 action.latency_ = 0.0;
99 if ((action.latency_ <= 0.0) && (action.isSuspended() == 0)) {
100 action.updateBound();
101 maxminSystem_->update_variable_weight(action.getVariable(), 1.0);
104 XBT_DEBUG("Action (%p) : remains (%g) updated by %g.", &action, action.getRemains(),
105 action.getVariable()->get_value() * delta);
106 action.updateRemains(action.getVariable()->get_value() * delta);
108 if (action.getMaxDuration() > NO_MAX_DURATION)
109 action.updateMaxDuration(delta);
111 XBT_DEBUG("Action (%p) : remains (%g).", &action, action.getRemains());
113 /* In the next if cascade, the action can be finished either because:
114 * - The amount of remaining work reached 0
115 * - The max duration was reached
116 * If it's not done, it may have failed.
119 if (((action.getRemains() <= 0) && (action.getVariable()->get_weight() > 0)) ||
120 ((action.getMaxDuration() > NO_MAX_DURATION) && (action.getMaxDuration() <= 0))) {
121 action.finish(Action::State::done);
123 /* Need to check that none of the model has failed */
125 lmm_constraint_t cnst = action.getVariable()->get_constraint(i);
126 while (cnst != nullptr) {
128 void* constraint_id = cnst->get_id();
129 if (static_cast<simgrid::surf::Resource*>(constraint_id)->isOff()) {
130 XBT_DEBUG("Action (%p) Failed!!", &action);
131 action.finish(Action::State::failed);
134 cnst = action.getVariable()->get_constraint(i);
140 Action *HostL07Model::executeParallelTask(int host_nb, sg_host_t *host_list,
141 double *flops_amount, double *bytes_amount,double rate) {
142 return new L07Action(this, host_nb, host_list, flops_amount, bytes_amount, rate);
145 L07Action::L07Action(Model *model, int host_nb, sg_host_t *host_list,
146 double *flops_amount, double *bytes_amount, double rate)
147 : CpuAction(model, 1, 0)
148 , computationAmount_(flops_amount)
149 , communicationAmount_(bytes_amount)
153 int nb_used_host = 0; /* Only the hosts with something to compute (>0 flops) are counted) */
154 double latency = 0.0;
156 this->hostList_->reserve(host_nb);
157 for (int i = 0; i < host_nb; i++) {
158 this->hostList_->push_back(host_list[i]);
159 if (flops_amount[i] > 0)
163 /* Compute the number of affected resources... */
164 if(bytes_amount != nullptr) {
165 std::unordered_set<const char*> affected_links;
167 for (int i = 0; i < host_nb; i++) {
168 for (int j = 0; j < host_nb; j++) {
170 if (bytes_amount[i * host_nb + j] > 0) {
173 std::vector<LinkImpl*> route;
174 hostList_->at(i)->routeTo(hostList_->at(j), route, &lat);
175 latency = std::max(latency, lat);
177 for (auto const& link : route)
178 affected_links.insert(link->getCname());
183 nb_link = affected_links.size();
186 XBT_DEBUG("Creating a parallel task (%p) with %d hosts and %d unique links.", this, host_nb, nb_link);
189 setVariable(model->getMaxminSystem()->variable_new(this, 1.0, (rate > 0 ? rate : -1.0), host_nb + nb_link));
192 model->getMaxminSystem()->update_variable_weight(getVariable(), 0.0);
194 for (int i = 0; i < host_nb; i++)
195 model->getMaxminSystem()->expand(host_list[i]->pimpl_cpu->constraint(), getVariable(), flops_amount[i]);
197 if(bytes_amount != nullptr) {
198 for (int i = 0; i < host_nb; i++) {
199 for (int j = 0; j < host_nb; j++) {
200 if (bytes_amount[i * host_nb + j] > 0.0) {
201 std::vector<LinkImpl*> route;
202 hostList_->at(i)->routeTo(hostList_->at(j), route, nullptr);
204 for (auto const& link : route)
205 model->getMaxminSystem()->expand_add(link->constraint(), this->getVariable(),
206 bytes_amount[i * host_nb + j]);
212 if (nb_link + nb_used_host == 0) {
214 this->setRemains(0.0);
219 Action* NetworkL07Model::communicate(s4u::Host* src, s4u::Host* dst, double size, double rate)
221 sg_host_t* host_list = new sg_host_t[2]();
222 double* flops_amount = new double[2]();
223 double* bytes_amount = new double[4]();
227 bytes_amount[1] = size;
229 return hostModel_->executeParallelTask(2, host_list, flops_amount, bytes_amount, rate);
232 Cpu *CpuL07Model::createCpu(simgrid::s4u::Host *host, std::vector<double> *speedPerPstate, int core)
234 return new CpuL07(this, host, speedPerPstate, core);
237 LinkImpl* NetworkL07Model::createLink(const std::string& name, double bandwidth, double latency,
238 e_surf_link_sharing_policy_t policy)
240 return new LinkL07(this, name, bandwidth, latency, policy);
247 CpuL07::CpuL07(CpuL07Model* model, simgrid::s4u::Host* host, std::vector<double>* speedPerPstate, int core)
248 : Cpu(model, host, model->getMaxminSystem()->constraint_new(this, speedPerPstate->front()), speedPerPstate, core)
252 CpuL07::~CpuL07()=default;
254 LinkL07::LinkL07(NetworkL07Model* model, const std::string& name, double bandwidth, double latency,
255 e_surf_link_sharing_policy_t policy)
256 : LinkImpl(model, name, model->getMaxminSystem()->constraint_new(this, bandwidth))
258 bandwidth_.peak = bandwidth;
259 latency_.peak = latency;
261 if (policy == SURF_LINK_FATPIPE)
262 constraint()->unshare();
264 s4u::Link::onCreation(this->piface_);
267 Action *CpuL07::execution_start(double size)
269 sg_host_t* host_list = new sg_host_t[1]();
270 double* flops_amount = new double[1]();
272 host_list[0] = getHost();
273 flops_amount[0] = size;
275 return static_cast<CpuL07Model*>(model())->hostModel_->executeParallelTask(1, host_list, flops_amount, nullptr, -1);
278 Action *CpuL07::sleep(double duration)
280 L07Action *action = static_cast<L07Action*>(execution_start(1.0));
281 action->setMaxDuration(duration);
282 action->suspended_ = 2;
283 model()->getMaxminSystem()->update_variable_weight(action->getVariable(), 0.0);
288 bool CpuL07::isUsed(){
289 return model()->getMaxminSystem()->constraint_used(constraint());
292 /** @brief take into account changes of speed (either load or max) */
293 void CpuL07::onSpeedChange() {
294 lmm_variable_t var = nullptr;
295 const_lmm_element_t elem = nullptr;
297 model()->getMaxminSystem()->update_constraint_bound(constraint(), speed_.peak * speed_.scale);
298 while ((var = constraint()->get_variable(&elem))) {
299 Action* action = static_cast<Action*>(var->get_id());
301 model()->getMaxminSystem()->update_variable_bound(action->getVariable(), speed_.scale * speed_.peak);
304 Cpu::onSpeedChange();
308 bool LinkL07::isUsed(){
309 return model()->getMaxminSystem()->constraint_used(constraint());
312 void CpuL07::apply_event(tmgr_trace_event_t triggered, double value)
314 XBT_DEBUG("Updating cpu %s (%p) with value %g", getCname(), this, value);
315 if (triggered == speed_.event) {
316 speed_.scale = value;
318 tmgr_trace_event_unref(&speed_.event);
320 } else if (triggered == stateEvent_) {
325 tmgr_trace_event_unref(&stateEvent_);
328 xbt_die("Unknown event!\n");
332 void LinkL07::apply_event(tmgr_trace_event_t triggered, double value)
334 XBT_DEBUG("Updating link %s (%p) with value=%f", getCname(), this, value);
335 if (triggered == bandwidth_.event) {
337 tmgr_trace_event_unref(&bandwidth_.event);
339 } else if (triggered == latency_.event) {
341 tmgr_trace_event_unref(&latency_.event);
343 } else if (triggered == stateEvent_) {
348 tmgr_trace_event_unref(&stateEvent_);
351 xbt_die("Unknown event ! \n");
355 void LinkL07::setBandwidth(double value)
357 bandwidth_.peak = value;
358 model()->getMaxminSystem()->update_constraint_bound(constraint(), bandwidth_.peak * bandwidth_.scale);
361 void LinkL07::setLatency(double value)
363 lmm_variable_t var = nullptr;
365 const_lmm_element_t elem = nullptr;
367 latency_.peak = value;
368 while ((var = constraint()->get_variable(&elem))) {
369 action = static_cast<L07Action*>(var->get_id());
370 action->updateBound();
373 LinkL07::~LinkL07() = default;
379 L07Action::~L07Action(){
381 delete[] communicationAmount_;
382 delete[] computationAmount_;
385 void L07Action::updateBound()
387 double lat_current = 0.0;
389 int hostNb = hostList_->size();
391 if (communicationAmount_ != nullptr) {
392 for (int i = 0; i < hostNb; i++) {
393 for (int j = 0; j < hostNb; j++) {
395 if (communicationAmount_[i * hostNb + j] > 0) {
397 std::vector<LinkImpl*> route;
398 hostList_->at(i)->routeTo(hostList_->at(j), route, &lat);
400 lat_current = std::max(lat_current, lat * communicationAmount_[i * hostNb + j]);
405 double lat_bound = sg_tcp_gamma / (2.0 * lat_current);
406 XBT_DEBUG("action (%p) : lat_bound = %g", this, lat_bound);
407 if ((latency_ <= 0.0) && (suspended_ == 0)) {
409 getModel()->getMaxminSystem()->update_variable_bound(getVariable(), lat_bound);
411 getModel()->getMaxminSystem()->update_variable_bound(getVariable(), std::min(rate_, lat_bound));
415 int L07Action::unref()
419 if (action_hook.is_linked())
420 simgrid::xbt::intrusive_erase(*stateSet_, *this);
422 getModel()->getMaxminSystem()->variable_free(getVariable());