1 /* Copyright (c) 2007-2018. The SimGrid Team. All rights reserved. */
3 /* This program is free software; you can redistribute it and/or modify it
4 * under the terms of the license (GNU LGPL) which comes with this package. */
6 #include "ptask_L07.hpp"
7 #include "surf/surf.hpp"
8 #include "xbt/config.hpp"
10 #include <unordered_set>
12 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_host);
13 XBT_LOG_EXTERNAL_CATEGORY(xbt_cfg);
15 /**************************************/
16 /*** Resource Creation & Destruction **/
17 /**************************************/
18 void surf_host_model_init_ptask_L07()
20 XBT_CINFO(xbt_cfg,"Switching to the L07 model to handle parallel tasks.");
21 xbt_assert(not surf_cpu_model_pm, "CPU model type already defined");
22 xbt_assert(not surf_network_model, "network model type already defined");
24 surf_host_model = new simgrid::surf::HostL07Model();
25 all_existing_models->push_back(surf_host_model);
32 HostL07Model::HostL07Model() : HostModel() {
33 auto* maxmin_system = new simgrid::kernel::lmm::FairBottleneck(true /* selective update */);
34 set_maxmin_system(maxmin_system);
35 surf_network_model = new NetworkL07Model(this, maxmin_system);
36 surf_cpu_model_pm = new CpuL07Model(this, maxmin_system);
39 HostL07Model::~HostL07Model()
41 delete surf_network_model;
42 delete surf_cpu_model_pm;
45 CpuL07Model::CpuL07Model(HostL07Model* hmodel, kernel::lmm::System* sys)
46 : CpuModel(Model::UpdateAlgo::Full), hostModel_(hmodel)
48 set_maxmin_system(sys);
51 CpuL07Model::~CpuL07Model()
53 set_maxmin_system(nullptr);
56 NetworkL07Model::NetworkL07Model(HostL07Model* hmodel, kernel::lmm::System* sys)
57 : NetworkModel(Model::UpdateAlgo::Full), hostModel_(hmodel)
59 set_maxmin_system(sys);
60 loopback_ = NetworkL07Model::createLink("__loopback__", 498000000, 0.000015, s4u::Link::SharingPolicy::FATPIPE);
63 NetworkL07Model::~NetworkL07Model()
65 set_maxmin_system(nullptr);
68 double HostL07Model::next_occuring_event(double now)
70 double min = HostModel::next_occuring_event_full(now);
71 for (kernel::resource::Action const& action : *get_running_action_set()) {
72 const L07Action& net_action = static_cast<const L07Action&>(action);
73 if (net_action.latency_ > 0 && (min < 0 || net_action.latency_ < min)) {
74 min = net_action.latency_;
75 XBT_DEBUG("Updating min with %p (start %f): %f", &net_action, net_action.get_start_time(), min);
78 XBT_DEBUG("min value: %f", min);
83 void HostL07Model::update_actions_state(double /*now*/, double delta)
85 for (auto it = std::begin(*get_running_action_set()); it != std::end(*get_running_action_set());) {
86 L07Action& action = static_cast<L07Action&>(*it);
87 ++it; // increment iterator here since the following calls to action.finish() may invalidate it
88 if (action.latency_ > 0) {
89 if (action.latency_ > delta) {
90 double_update(&(action.latency_), delta, sg_surf_precision);
92 action.latency_ = 0.0;
94 if ((action.latency_ <= 0.0) && (action.is_suspended() == 0)) {
96 get_maxmin_system()->update_variable_weight(action.get_variable(), 1.0);
99 XBT_DEBUG("Action (%p) : remains (%g) updated by %g.", &action, action.get_remains(),
100 action.get_variable()->get_value() * delta);
101 action.update_remains(action.get_variable()->get_value() * delta);
103 if (action.get_max_duration() > NO_MAX_DURATION)
104 action.update_max_duration(delta);
106 XBT_DEBUG("Action (%p) : remains (%g).", &action, action.get_remains());
108 /* In the next if cascade, the action can be finished either because:
109 * - The amount of remaining work reached 0
110 * - The max duration was reached
111 * If it's not done, it may have failed.
114 if (((action.get_remains() <= 0) && (action.get_variable()->get_weight() > 0)) ||
115 ((action.get_max_duration() > NO_MAX_DURATION) && (action.get_max_duration() <= 0))) {
116 action.finish(kernel::resource::Action::State::done);
118 /* Need to check that none of the model has failed */
120 kernel::lmm::Constraint* cnst = action.get_variable()->get_constraint(i);
121 while (cnst != nullptr) {
123 void* constraint_id = cnst->get_id();
124 if (static_cast<simgrid::kernel::resource::Resource*>(constraint_id)->is_off()) {
125 XBT_DEBUG("Action (%p) Failed!!", &action);
126 action.finish(kernel::resource::Action::State::failed);
129 cnst = action.get_variable()->get_constraint(i);
135 kernel::resource::Action* HostL07Model::execute_parallel(int host_nb, sg_host_t* host_list, double* flops_amount,
136 double* bytes_amount, double rate)
138 return new L07Action(this, host_nb, host_list, flops_amount, bytes_amount, rate);
141 L07Action::L07Action(kernel::resource::Model* model, int host_nb, sg_host_t* host_list, double* flops_amount,
142 double* bytes_amount, double rate)
143 : CpuAction(model, 1, 0), computationAmount_(flops_amount), communicationAmount_(bytes_amount), rate_(rate)
146 int nb_used_host = 0; /* Only the hosts with something to compute (>0 flops) are counted) */
147 double latency = 0.0;
149 this->hostList_->reserve(host_nb);
150 for (int i = 0; i < host_nb; i++) {
151 this->hostList_->push_back(host_list[i]);
152 if (flops_amount[i] > 0)
156 /* Compute the number of affected resources... */
157 if(bytes_amount != nullptr) {
158 std::unordered_set<const char*> affected_links;
160 for (int i = 0; i < host_nb; i++) {
161 for (int j = 0; j < host_nb; j++) {
163 if (bytes_amount[i * host_nb + j] > 0) {
166 std::vector<kernel::resource::LinkImpl*> route;
167 hostList_->at(i)->routeTo(hostList_->at(j), route, &lat);
168 latency = std::max(latency, lat);
170 for (auto const& link : route)
171 affected_links.insert(link->get_cname());
176 nb_link = affected_links.size();
179 XBT_DEBUG("Creating a parallel task (%p) with %d hosts and %d unique links.", this, host_nb, nb_link);
182 set_variable(model->get_maxmin_system()->variable_new(this, 1.0, (rate > 0 ? rate : -1.0), host_nb + nb_link));
185 model->get_maxmin_system()->update_variable_weight(get_variable(), 0.0);
187 for (int i = 0; i < host_nb; i++)
188 model->get_maxmin_system()->expand(host_list[i]->pimpl_cpu->get_constraint(), get_variable(), flops_amount[i]);
190 if(bytes_amount != nullptr) {
191 for (int i = 0; i < host_nb; i++) {
192 for (int j = 0; j < host_nb; j++) {
193 if (bytes_amount[i * host_nb + j] > 0.0) {
194 std::vector<kernel::resource::LinkImpl*> route;
195 hostList_->at(i)->routeTo(hostList_->at(j), route, nullptr);
197 for (auto const& link : route)
198 model->get_maxmin_system()->expand_add(link->get_constraint(), this->get_variable(),
199 bytes_amount[i * host_nb + j]);
205 if (nb_link + nb_used_host == 0) {
207 this->set_remains(0.0);
212 kernel::resource::Action* NetworkL07Model::communicate(s4u::Host* src, s4u::Host* dst, double size, double rate)
214 sg_host_t* host_list = new sg_host_t[2]();
215 double* flops_amount = new double[2]();
216 double* bytes_amount = new double[4]();
220 bytes_amount[1] = size;
222 return hostModel_->execute_parallel(2, host_list, flops_amount, bytes_amount, rate);
225 Cpu *CpuL07Model::createCpu(simgrid::s4u::Host *host, std::vector<double> *speedPerPstate, int core)
227 return new CpuL07(this, host, speedPerPstate, core);
230 kernel::resource::LinkImpl* NetworkL07Model::createLink(const std::string& name, double bandwidth, double latency,
231 s4u::Link::SharingPolicy policy)
233 return new LinkL07(this, name, bandwidth, latency, policy);
240 CpuL07::CpuL07(CpuL07Model* model, simgrid::s4u::Host* host, std::vector<double>* speedPerPstate, int core)
241 : Cpu(model, host, model->get_maxmin_system()->constraint_new(this, speedPerPstate->front()), speedPerPstate, core)
245 CpuL07::~CpuL07()=default;
247 LinkL07::LinkL07(NetworkL07Model* model, const std::string& name, double bandwidth, double latency,
248 s4u::Link::SharingPolicy policy)
249 : LinkImpl(model, name, model->get_maxmin_system()->constraint_new(this, bandwidth))
251 bandwidth_.peak = bandwidth;
252 latency_.peak = latency;
254 if (policy == s4u::Link::SharingPolicy::FATPIPE)
255 get_constraint()->unshare();
257 s4u::Link::onCreation(this->piface_);
260 kernel::resource::Action* CpuL07::execution_start(double size)
262 sg_host_t* host_list = new sg_host_t[1]();
263 double* flops_amount = new double[1]();
265 host_list[0] = getHost();
266 flops_amount[0] = size;
268 return static_cast<CpuL07Model*>(get_model())->hostModel_->execute_parallel(1, host_list, flops_amount, nullptr, -1);
271 kernel::resource::Action* CpuL07::sleep(double duration)
273 L07Action *action = static_cast<L07Action*>(execution_start(1.0));
274 action->set_max_duration(duration);
275 action->suspended_ = kernel::resource::Action::SuspendStates::sleeping;
276 get_model()->get_maxmin_system()->update_variable_weight(action->get_variable(), 0.0);
281 bool CpuL07::is_used()
283 return get_model()->get_maxmin_system()->constraint_used(get_constraint());
286 /** @brief take into account changes of speed (either load or max) */
287 void CpuL07::onSpeedChange() {
288 kernel::lmm::Variable* var = nullptr;
289 const kernel::lmm::Element* elem = nullptr;
291 get_model()->get_maxmin_system()->update_constraint_bound(get_constraint(), speed_.peak * speed_.scale);
292 while ((var = get_constraint()->get_variable(&elem))) {
293 kernel::resource::Action* action = static_cast<kernel::resource::Action*>(var->get_id());
295 get_model()->get_maxmin_system()->update_variable_bound(action->get_variable(), speed_.scale * speed_.peak);
298 Cpu::onSpeedChange();
301 bool LinkL07::is_used()
303 return get_model()->get_maxmin_system()->constraint_used(get_constraint());
306 void CpuL07::apply_event(tmgr_trace_event_t triggered, double value)
308 XBT_DEBUG("Updating cpu %s (%p) with value %g", get_cname(), this, value);
309 if (triggered == speed_.event) {
310 speed_.scale = value;
312 tmgr_trace_event_unref(&speed_.event);
314 } else if (triggered == stateEvent_) {
319 tmgr_trace_event_unref(&stateEvent_);
322 xbt_die("Unknown event!\n");
326 void LinkL07::apply_event(tmgr_trace_event_t triggered, double value)
328 XBT_DEBUG("Updating link %s (%p) with value=%f", get_cname(), this, value);
329 if (triggered == bandwidth_.event) {
331 tmgr_trace_event_unref(&bandwidth_.event);
333 } else if (triggered == latency_.event) {
335 tmgr_trace_event_unref(&latency_.event);
337 } else if (triggered == stateEvent_) {
342 tmgr_trace_event_unref(&stateEvent_);
345 xbt_die("Unknown event ! \n");
349 void LinkL07::setBandwidth(double value)
351 bandwidth_.peak = value;
352 get_model()->get_maxmin_system()->update_constraint_bound(get_constraint(), bandwidth_.peak * bandwidth_.scale);
355 void LinkL07::setLatency(double value)
357 kernel::lmm::Variable* var = nullptr;
359 const kernel::lmm::Element* elem = nullptr;
361 latency_.peak = value;
362 while ((var = get_constraint()->get_variable(&elem))) {
363 action = static_cast<L07Action*>(var->get_id());
364 action->updateBound();
367 LinkL07::~LinkL07() = default;
373 L07Action::~L07Action(){
375 delete[] communicationAmount_;
376 delete[] computationAmount_;
379 void L07Action::updateBound()
381 double lat_current = 0.0;
383 int hostNb = hostList_->size();
385 if (communicationAmount_ != nullptr) {
386 for (int i = 0; i < hostNb; i++) {
387 for (int j = 0; j < hostNb; j++) {
389 if (communicationAmount_[i * hostNb + j] > 0) {
391 std::vector<kernel::resource::LinkImpl*> route;
392 hostList_->at(i)->routeTo(hostList_->at(j), route, &lat);
394 lat_current = std::max(lat_current, lat * communicationAmount_[i * hostNb + j]);
399 double lat_bound = kernel::resource::NetworkModel::cfg_tcp_gamma / (2.0 * lat_current);
400 XBT_DEBUG("action (%p) : lat_bound = %g", this, lat_bound);
401 if ((latency_ <= 0.0) && (suspended_ == Action::SuspendStates::not_suspended)) {
403 get_model()->get_maxmin_system()->update_variable_bound(get_variable(), lat_bound);
405 get_model()->get_maxmin_system()->update_variable_bound(get_variable(), std::min(rate_, lat_bound));