1 /* Copyright (c) 2010-2018. 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 "simgrid/plugins/energy.h"
7 #include "simgrid/plugins/load.h"
8 #include "simgrid/s4u/Engine.hpp"
9 #include "src/plugins/vm/VirtualMachineImpl.hpp"
10 #include "src/surf/cpu_interface.hpp"
12 #include <boost/algorithm/string/classification.hpp>
13 #include <boost/algorithm/string/split.hpp>
15 /** @addtogroup plugin_energy
17 This is the energy plugin, enabling to account not only for computation time, but also for the dissipated energy in the
19 To activate this plugin, first call sg_host_energy_plugin_init() before your #MSG_init(), and then use
20 MSG_host_get_consumed_energy() to retrieve the consumption of a given host.
22 When the host is on, this energy consumption naturally depends on both the current CPU load and the host energy profile.
23 According to our measurements, the consumption is somehow linear in the amount of cores at full speed, with an
24 abnormality when all the cores are idle. The full details are in
25 <a href="https://hal.inria.fr/hal-01523608">our scientific paper</a> on that topic.
27 As a result, our energy model takes 4 parameters:
29 - \b Idle: instantaneous consumption (in Watt) when your host is up and running, but without anything to do.
30 - \b OneCore: instantaneous consumption (in Watt) when only one core is active, at 100%.
31 - \b AllCores: instantaneous consumption (in Watt) when all cores of the host are at 100%.
32 - \b Off: instantaneous consumption (in Watt) when the host is turned off.
34 Here is an example of XML declaration:
37 <host id="HostA" power="100.0Mf" cores="4">
38 <prop id="watt_per_state" value="100.0:120.0:200.0" />
39 <prop id="watt_off" value="10" />
43 This example gives the following parameters: \b Off is 10 Watts; \b Idle is 100 Watts; \b OneCore is 120 Watts and \b
44 AllCores is 200 Watts.
45 This is enough to compute the consumption as a function of the amount of loaded cores:
48 <tr><th>\#Cores loaded</th><th>Consumption</th><th>Explanation</th></tr>
49 <tr><td>0</td><td> 100 Watts</td><td>Idle value</td></tr>
50 <tr><td>1</td><td> 120 Watts</td><td>OneCore value</td></tr>
51 <tr><td>2</td><td> 147 Watts</td><td>linear extrapolation between OneCore and AllCores</td></tr>
52 <tr><td>3</td><td> 173 Watts</td><td>linear extrapolation between OneCore and AllCores</td></tr>
53 <tr><td>4</td><td> 200 Watts</td><td>AllCores value</td></tr>
56 ### What if a given core is only at load 50%?
58 This is impossible in SimGrid because we recompute everything each time that the CPU starts or stops doing something.
59 So if a core is at load 50% over a period, it means that it is at load 100% half of the time and at load 0% the rest of
60 the time, and our model holds.
62 ### What if the host has only one core?
64 In this case, the parameters \b OneCore and \b AllCores are obviously the same.
65 Actually, SimGrid expect an energetic profile formatted as 'Idle:Running' for mono-cores hosts.
66 If you insist on passing 3 parameters in this case, then you must have the same value for \b OneCore and \b AllCores.
69 <host id="HostC" power="100.0Mf" cores="1">
70 <prop id="watt_per_state" value="95.0:200.0" /> <!-- we may have used '95:200:200' instead -->
71 <prop id="watt_off" value="10" />
75 ### How does DVFS interact with the host energy model?
77 If your host has several DVFS levels (several pstates), then you should give the energetic profile of each pstate level:
80 <host id="HostC" power="100.0Mf,50.0Mf,20.0Mf" cores="4">
81 <prop id="watt_per_state" value="95.0:120.0:200.0, 93.0:115.0:170.0, 90.0:110.0:150.0" />
82 <prop id="watt_off" value="10" />
86 This encodes the following values
88 <tr><th>pstate</th><th>Performance</th><th>Idle</th><th>OneCore</th><th>AllCores</th></tr>
89 <tr><td>0</td><td>100 Mflop/s</td><td>95 Watts</td><td>120 Watts</td><td>200 Watts</td></tr>
90 <tr><td>1</td><td>50 Mflop/s</td><td>93 Watts</td><td>115 Watts</td><td>170 Watts</td></tr>
91 <tr><td>2</td><td>20 Mflop/s</td><td>90 Watts</td><td>110 Watts</td><td>150 Watts</td></tr>
94 To change the pstate of a given CPU, use the following functions:
95 #MSG_host_get_nb_pstates(), simgrid#s4u#Host#setPstate(), #MSG_host_get_power_peak_at().
97 ### How accurate are these models?
99 This model cannot be more accurate than your instantiation: with the default values, your result will not be accurate at
100 all. You can still get accurate energy prediction, provided that you carefully instantiate the model.
101 The first step is to ensure that your timing prediction match perfectly. But this is only the first step of the path,
102 and you really want to read <a href="https://hal.inria.fr/hal-01523608">this paper</a> to see all what you need to do
103 before you can get accurate energy predictions.
106 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_energy, surf, "Logging specific to the SURF energy plugin");
117 PowerRange(double idle, double min, double max) : idle_(idle), min_(min), max_(max) {}
122 static simgrid::xbt::Extension<simgrid::s4u::Host, HostEnergy> EXTENSION_ID;
124 explicit HostEnergy(simgrid::s4u::Host* ptr);
127 double get_current_watts_value();
128 double get_current_watts_value(double cpu_load);
129 double get_consumed_energy();
130 double get_watt_min_at(int pstate);
131 double get_watt_max_at(int pstate);
135 void init_watts_range_list();
136 simgrid::s4u::Host* host_ = nullptr;
137 /*< List of (min_power,max_power) pairs corresponding to each cpu pstate */
138 std::vector<PowerRange> power_range_watts_list_;
140 /* We need to keep track of what pstate has been used, as we will sometimes be notified only *after* a pstate has been
141 * used (but we need to update the energy consumption with the old pstate!)
144 const int pstate_off_ = -1;
147 double watts_off_ = 0.0; /*< Consumption when the machine is turned off (shutdown) */
148 double total_energy_ = 0.0; /*< Total energy consumed by the host */
149 double last_updated_; /*< Timestamp of the last energy update event*/
152 simgrid::xbt::Extension<simgrid::s4u::Host, HostEnergy> HostEnergy::EXTENSION_ID;
154 /* Computes the consumption so far. Called lazily on need. */
155 void HostEnergy::update()
157 double start_time = this->last_updated_;
158 double finish_time = surf_get_clock();
160 if (start_time < finish_time) {
161 double previous_energy = this->total_energy_;
163 double instantaneous_consumption = this->get_current_watts_value();
165 double energy_this_step = instantaneous_consumption * (finish_time - start_time);
167 // TODO Trace: Trace energy_this_step from start_time to finish_time in host->getName()
169 this->total_energy_ = previous_energy + energy_this_step;
170 this->last_updated_ = finish_time;
172 XBT_DEBUG("[update_energy of %s] period=[%.2f-%.2f]; current power peak=%.0E flop/s; consumption change: %.2f J -> "
174 host_->get_cname(), start_time, finish_time, host_->pimpl_cpu->get_speed(1.0), previous_energy,
178 /* Save data for the upcoming time interval: whether it's on/off and the pstate if it's on */
179 this->pstate_ = host_->is_on() ? host_->get_pstate() : pstate_off_;
182 HostEnergy::HostEnergy(simgrid::s4u::Host* ptr) : host_(ptr), last_updated_(surf_get_clock())
184 init_watts_range_list();
186 const char* off_power_str = host_->get_property("watt_off");
187 if (off_power_str != nullptr) {
189 this->watts_off_ = std::stod(std::string(off_power_str));
190 } catch (std::invalid_argument& ia) {
191 throw std::invalid_argument(std::string("Invalid value for property watt_off of host ") + host_->get_cname() +
192 ": " + off_power_str);
195 /* watts_off is 0 by default */
198 HostEnergy::~HostEnergy() = default;
200 double HostEnergy::get_watt_min_at(int pstate)
202 xbt_assert(not power_range_watts_list_.empty(), "No power range properties specified for host %s",
204 return power_range_watts_list_[pstate].min_;
207 double HostEnergy::get_watt_max_at(int pstate)
209 xbt_assert(not power_range_watts_list_.empty(), "No power range properties specified for host %s",
211 return power_range_watts_list_[pstate].max_;
214 /** @brief Computes the power consumed by the host according to the current situation
216 * - If the host is off, that's the watts_off value
217 * - if it's on, take the current pstate and the current processor load into account */
218 double HostEnergy::get_current_watts_value()
220 if (this->pstate_ == pstate_off_) // The host is off (or was off at the beginning of this time interval)
221 return this->watts_off_;
223 double current_speed = host_->getSpeed();
226 // We may have start == finish if the past consumption was updated since the simcall was started
227 // for example if 2 actors requested to update the same host's consumption in a given scheduling round.
229 // Even in this case, we need to save the pstate for the next call (after this big if),
230 // which may have changed since that recent update.
232 if (current_speed <= 0)
233 // Some users declare a pstate of speed 0 flops (e.g., to model boot time).
234 // We consider that the machine is then fully loaded. That's arbitrary but it avoids a NaN
237 cpu_load = host_->pimpl_cpu->get_constraint()->get_usage() / current_speed;
239 /** Divide by the number of cores here **/
240 cpu_load /= host_->pimpl_cpu->get_core_count();
242 if (cpu_load > 1) // A machine with a load > 1 consumes as much as a fully loaded machine, not more
245 /* The problem with this model is that the load is always 0 or 1, never something less.
246 * Another possibility could be to model the total energy as
248 * X/(X+Y)*W_idle + Y/(X+Y)*W_burn
250 * where X is the amount of idling cores, and Y the amount of computing cores.
252 return get_current_watts_value(cpu_load);
255 /** @brief Computes the power that the host would consume at the provided processor load
257 * Whether the host is ON or OFF is not taken into account.
259 double HostEnergy::get_current_watts_value(double cpu_load)
261 xbt_assert(not power_range_watts_list_.empty(), "No power range properties specified for host %s",
264 /* Return watts_off if pstate == pstate_off (ie, if the host is off) */
265 if (this->pstate_ == pstate_off_) {
269 /* min_power corresponds to the power consumed when only one core is active */
270 /* max_power is the power consumed at 100% cpu load */
271 auto range = power_range_watts_list_.at(this->pstate_);
272 double current_power = 0;
273 double min_power = 0;
274 double max_power = 0;
275 double power_slope = 0;
277 if (cpu_load > 0) { /* Something is going on, the machine is not idle */
278 double min_power = range.min_;
279 double max_power = range.max_;
282 * The min_power states how much we consume when only one single
283 * core is working. This means that when cpu_load == 1/coreCount, then
284 * current_power == min_power.
286 * The maximum must be reached when all cores are working (but 1 core was
287 * already accounted for by min_power)
288 * i.e., we need min_power + (maxCpuLoad-1/coreCount)*power_slope == max_power
289 * (maxCpuLoad is by definition 1)
292 int coreCount = host_->get_core_count();
293 double coreReciprocal = static_cast<double>(1) / static_cast<double>(coreCount);
295 power_slope = (max_power - min_power) / (1 - coreReciprocal);
297 power_slope = 0; // Should be 0, since max_power == min_power (in this case)
299 current_power = min_power + (cpu_load - coreReciprocal) * power_slope;
300 } else { /* Our machine is idle, take the dedicated value! */
301 current_power = range.idle_;
304 XBT_DEBUG("[get_current_watts] min_power=%f, max_power=%f, slope=%f", min_power, max_power, power_slope);
305 XBT_DEBUG("[get_current_watts] Current power (watts) = %f, load = %f", current_power, cpu_load);
307 return current_power;
310 double HostEnergy::get_consumed_energy()
312 if (last_updated_ < surf_get_clock()) // We need to simcall this as it modifies the environment
313 simgrid::simix::simcall(std::bind(&HostEnergy::update, this));
315 return total_energy_;
318 void HostEnergy::init_watts_range_list()
320 const char* all_power_values_str = host_->get_property("watt_per_state");
321 if (all_power_values_str == nullptr)
324 std::vector<std::string> all_power_values;
325 boost::split(all_power_values, all_power_values_str, boost::is_any_of(","));
326 XBT_DEBUG("%s: profile: %s, cores: %d", host_->get_cname(), all_power_values_str, host_->get_core_count());
329 for (auto const& current_power_values_str : all_power_values) {
330 /* retrieve the power values associated with the current pstate */
331 std::vector<std::string> current_power_values;
332 boost::split(current_power_values, current_power_values_str, boost::is_any_of(":"));
333 if (host_->get_core_count() == 1) {
334 xbt_assert(current_power_values.size() == 2 || current_power_values.size() == 3,
335 "Power properties incorrectly defined for host %s."
336 "It should be 'Idle:FullSpeed' power values because you have one core only.",
338 if (current_power_values.size() == 2) {
339 // In this case, 1core == AllCores
340 current_power_values.push_back(current_power_values.at(1));
341 } else { // size == 3
342 xbt_assert((current_power_values.at(1)) == (current_power_values.at(2)),
343 "Power properties incorrectly defined for host %s.\n"
344 "The energy profile of mono-cores should be formatted as 'Idle:FullSpeed' only.\n"
345 "If you go for a 'Idle:OneCore:AllCores' power profile on mono-cores, then OneCore and AllCores "
350 xbt_assert(current_power_values.size() == 3,
351 "Power properties incorrectly defined for host %s."
352 "It should be 'Idle:OneCore:AllCores' power values because you have more than one core.",
356 /* min_power corresponds to the idle power (cpu load = 0) */
357 /* max_power is the power consumed at 100% cpu load */
358 char* msg_idle = bprintf("Invalid idle value for pstate %d on host %s: %%s", i, host_->get_cname());
359 char* msg_min = bprintf("Invalid OneCore value for pstate %d on host %s: %%s", i, host_->get_cname());
360 char* msg_max = bprintf("Invalid AllCores value for pstate %d on host %s: %%s", i, host_->get_cname());
361 PowerRange range(xbt_str_parse_double((current_power_values.at(0)).c_str(), msg_idle),
362 xbt_str_parse_double((current_power_values.at(1)).c_str(), msg_min),
363 xbt_str_parse_double((current_power_values.at(2)).c_str(), msg_max));
364 power_range_watts_list_.push_back(range);
371 } // namespace plugin
372 } // namespace simgrid
374 using simgrid::plugin::HostEnergy;
376 /* **************************** events callback *************************** */
377 static void on_creation(simgrid::s4u::Host& host)
379 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
382 // TODO Trace: set to zero the energy variable associated to host->getName()
384 host.extension_set(new HostEnergy(&host));
387 static void on_action_state_change(simgrid::surf::CpuAction* action)
389 for (simgrid::surf::Cpu* const& cpu : action->cpus()) {
390 simgrid::s4u::Host* host = cpu->get_host();
391 if (host != nullptr) {
393 // If it's a VM, take the corresponding PM
394 simgrid::s4u::VirtualMachine* vm = dynamic_cast<simgrid::s4u::VirtualMachine*>(host);
395 if (vm) // If it's a VM, take the corresponding PM
398 // Get the host_energy extension for the relevant host
399 HostEnergy* host_energy = host->extension<HostEnergy>();
401 if (host_energy->last_updated_ < surf_get_clock())
402 host_energy->update();
407 /* This callback is fired either when the host changes its state (on/off) ("onStateChange") or its speed
408 * (because the user changed the pstate, or because of external trace events) ("onSpeedChange") */
409 static void on_host_change(simgrid::s4u::Host& host)
411 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
414 HostEnergy* host_energy = host.extension<HostEnergy>();
416 host_energy->update();
419 static void on_host_destruction(simgrid::s4u::Host& host)
421 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
424 XBT_INFO("Energy consumption of host %s: %f Joules", host.get_cname(),
425 host.extension<HostEnergy>()->get_consumed_energy());
428 static void on_simulation_end()
430 std::vector<simgrid::s4u::Host*> hosts = simgrid::s4u::Engine::get_instance()->get_all_hosts();
432 double total_energy = 0.0; // Total energy consumption (whole platform)
433 double used_hosts_energy = 0.0; // Energy consumed by hosts that computed something
434 for (size_t i = 0; i < hosts.size(); i++) {
435 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(hosts[i]) == nullptr) { // Ignore virtual machines
437 bool host_was_used = (sg_host_get_computed_flops(hosts[i]) != 0);
438 double energy = hosts[i]->extension<HostEnergy>()->get_consumed_energy();
439 total_energy += energy;
441 used_hosts_energy += energy;
444 XBT_INFO("Total energy consumption: %f Joules (used hosts: %f Joules; unused/idle hosts: %f)", total_energy,
445 used_hosts_energy, total_energy - used_hosts_energy);
448 /* **************************** Public interface *************************** */
450 /** \ingroup plugin_energy
451 * \brief Enable host energy plugin
452 * \details Enable energy plugin to get joules consumption of each cpu. Call this function before #MSG_init().
454 void sg_host_energy_plugin_init()
456 if (HostEnergy::EXTENSION_ID.valid())
459 sg_host_load_plugin_init();
461 HostEnergy::EXTENSION_ID = simgrid::s4u::Host::extension_create<HostEnergy>();
463 simgrid::s4u::Host::on_creation.connect(&on_creation);
464 simgrid::s4u::Host::on_state_change.connect(&on_host_change);
465 simgrid::s4u::Host::on_speed_change.connect(&on_host_change);
466 simgrid::s4u::Host::on_destruction.connect(&on_host_destruction);
467 simgrid::s4u::on_simulation_end.connect(&on_simulation_end);
468 simgrid::surf::CpuAction::on_state_change.connect(&on_action_state_change);
471 /** @ingroup plugin_energy
472 * @brief updates the consumption of all hosts
474 * After this call, sg_host_get_consumed_energy() will not interrupt your process
475 * (until after the next clock update).
477 void sg_host_energy_update_all()
479 simgrid::simix::simcall([]() {
480 std::vector<simgrid::s4u::Host*> list = simgrid::s4u::Engine::get_instance()->get_all_hosts();
481 for (auto const& host : list)
482 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(host) == nullptr) // Ignore virtual machines
483 host->extension<HostEnergy>()->update();
487 /** @ingroup plugin_energy
488 * @brief Returns the total energy consumed by the host so far (in Joules)
490 * Please note that since the consumption is lazily updated, it may require a simcall to update it.
491 * The result is that the actor requesting this value will be interrupted,
492 * the value will be updated in kernel mode before returning the control to the requesting actor.
494 double sg_host_get_consumed_energy(sg_host_t host)
496 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
497 "The Energy plugin is not active. Please call sg_host_energy_plugin_init() during initialization.");
498 return host->extension<HostEnergy>()->get_consumed_energy();
501 /** @ingroup plugin_energy
502 * @brief Get the amount of watt dissipated at the given pstate when the host is idling
504 double sg_host_get_wattmin_at(sg_host_t host, int pstate)
506 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
507 "The Energy plugin is not active. Please call sg_host_energy_plugin_init() during initialization.");
508 return host->extension<HostEnergy>()->get_watt_min_at(pstate);
510 /** @ingroup plugin_energy
511 * @brief Returns the amount of watt dissipated at the given pstate when the host burns CPU at 100%
513 double sg_host_get_wattmax_at(sg_host_t host, int pstate)
515 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
516 "The Energy plugin is not active. Please call sg_host_energy_plugin_init() during initialization.");
517 return host->extension<HostEnergy>()->get_watt_max_at(pstate);
520 /** @ingroup plugin_energy
521 * @brief Returns the current consumption of the host
523 double sg_host_get_current_consumption(sg_host_t host)
525 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
526 "The Energy plugin is not active. Please call sg_host_energy_plugin_init() during initialization.");
527 return host->extension<HostEnergy>()->get_current_watts_value();