1 /* Copyright (c) 2010-2019. 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/s4u/Engine.hpp"
8 #include "src/kernel/activity/ExecImpl.hpp"
9 #include "src/include/surf/surf.hpp"
10 #include "src/plugins/vm/VirtualMachineImpl.hpp"
11 #include "src/surf/cpu_interface.hpp"
13 #include <boost/algorithm/string/classification.hpp>
14 #include <boost/algorithm/string/split.hpp>
16 SIMGRID_REGISTER_PLUGIN(host_energy, "Cpu energy consumption.", &sg_host_energy_plugin_init)
18 /** @addtogroup plugin_energy
20 This is the energy plugin, enabling to account not only for computation time, but also for the dissipated energy in the
22 To activate this plugin, first call sg_host_energy_plugin_init() before your #MSG_init(), and then use
23 MSG_host_get_consumed_energy() to retrieve the consumption of a given host.
25 When the host is on, this energy consumption naturally depends on both the current CPU load and the host energy profile.
26 According to our measurements, the consumption is somehow linear in the amount of cores at full speed, with an
27 abnormality when all the cores are idle. The full details are in
28 <a href="https://hal.inria.fr/hal-01523608">our scientific paper</a> on that topic.
30 As a result, our energy model takes 4 parameters:
32 - @b Idle: instantaneous consumption (in Watt) when your host is up and running, but without anything to do.
33 - @b OneCore: instantaneous consumption (in Watt) when only one core is active, at 100%.
34 - @b AllCores: instantaneous consumption (in Watt) when all cores of the host are at 100%.
35 - @b Off: instantaneous consumption (in Watt) when the host is turned off.
37 Here is an example of XML declaration:
40 <host id="HostA" power="100.0Mf" cores="4">
41 <prop id="watt_per_state" value="100.0:120.0:200.0" />
42 <prop id="watt_off" value="10" />
46 This example gives the following parameters: @b Off is 10 Watts; @b Idle is 100 Watts; @b OneCore is 120 Watts and @b
47 AllCores is 200 Watts.
48 This is enough to compute the consumption as a function of the amount of loaded cores:
51 <tr><th>@#Cores loaded</th><th>Consumption</th><th>Explanation</th></tr>
52 <tr><td>0</td><td> 100 Watts</td><td>Idle value</td></tr>
53 <tr><td>1</td><td> 120 Watts</td><td>OneCore value</td></tr>
54 <tr><td>2</td><td> 147 Watts</td><td>linear extrapolation between OneCore and AllCores</td></tr>
55 <tr><td>3</td><td> 173 Watts</td><td>linear extrapolation between OneCore and AllCores</td></tr>
56 <tr><td>4</td><td> 200 Watts</td><td>AllCores value</td></tr>
59 ### What if a given core is only at load 50%?
61 This is impossible in SimGrid because we recompute everything each time that the CPU starts or stops doing something.
62 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
63 the time, and our model holds.
65 ### What if the host has only one core?
67 In this case, the parameters @b OneCore and @b AllCores are obviously the same.
68 Actually, SimGrid expect an energetic profile formatted as 'Idle:Running' for mono-cores hosts.
69 If you insist on passing 3 parameters in this case, then you must have the same value for @b OneCore and @b AllCores.
72 <host id="HostC" power="100.0Mf" cores="1">
73 <prop id="watt_per_state" value="95.0:200.0" /> <!-- we may have used '95:200:200' instead -->
74 <prop id="watt_off" value="10" />
78 ### How does DVFS interact with the host energy model?
80 If your host has several DVFS levels (several pstates), then you should give the energetic profile of each pstate level:
83 <host id="HostC" power="100.0Mf,50.0Mf,20.0Mf" cores="4">
84 <prop id="watt_per_state" value="95.0:120.0:200.0, 93.0:115.0:170.0, 90.0:110.0:150.0" />
85 <prop id="watt_off" value="10" />
89 This encodes the following values
91 <tr><th>pstate</th><th>Performance</th><th>Idle</th><th>OneCore</th><th>AllCores</th></tr>
92 <tr><td>0</td><td>100 Mflop/s</td><td>95 Watts</td><td>120 Watts</td><td>200 Watts</td></tr>
93 <tr><td>1</td><td>50 Mflop/s</td><td>93 Watts</td><td>115 Watts</td><td>170 Watts</td></tr>
94 <tr><td>2</td><td>20 Mflop/s</td><td>90 Watts</td><td>110 Watts</td><td>150 Watts</td></tr>
97 To change the pstate of a given CPU, use the following functions:
98 #MSG_host_get_nb_pstates(), simgrid#s4u#Host#setPstate(), #MSG_host_get_power_peak_at().
100 ### How accurate are these models?
102 This model cannot be more accurate than your instantiation: with the default values, your result will not be accurate at
103 all. You can still get accurate energy prediction, provided that you carefully instantiate the model.
104 The first step is to ensure that your timing prediction match perfectly. But this is only the first step of the path,
105 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
106 before you can get accurate energy predictions.
109 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_energy, surf, "Logging specific to the SURF energy plugin");
111 // Forwards declaration needed to make this function a friend (because friends have external linkage by default)
112 static void on_simulation_end();
123 PowerRange(double idle, double min, double max) : idle_(idle), min_(min), max_(max) {}
127 friend void ::on_simulation_end(); // For access to host_was_used_
129 static simgrid::xbt::Extension<simgrid::s4u::Host, HostEnergy> EXTENSION_ID;
131 explicit HostEnergy(simgrid::s4u::Host* ptr);
134 double get_current_watts_value();
135 double get_current_watts_value(double cpu_load);
136 double get_consumed_energy();
137 double get_idle_consumption();
138 double get_watt_min_at(int pstate);
139 double get_watt_max_at(int pstate);
143 void init_watts_range_list();
144 simgrid::s4u::Host* host_ = nullptr;
145 /*< List of (min_power,max_power) pairs corresponding to each cpu pstate */
146 std::vector<PowerRange> power_range_watts_list_;
148 /* We need to keep track of what pstate has been used, as we will sometimes be notified only *after* a pstate has been
149 * used (but we need to update the energy consumption with the old pstate!)
152 const int pstate_off_ = -1;
154 /* Only used to split total energy into unused/used hosts.
155 * If you want to get this info for something else, rather use the host_load plugin
157 bool host_was_used_ = false;
159 double watts_off_ = 0.0; /*< Consumption when the machine is turned off (shutdown) */
160 double total_energy_ = 0.0; /*< Total energy consumed by the host */
161 double last_updated_; /*< Timestamp of the last energy update event*/
164 simgrid::xbt::Extension<simgrid::s4u::Host, HostEnergy> HostEnergy::EXTENSION_ID;
166 /* Computes the consumption so far. Called lazily on need. */
167 void HostEnergy::update()
169 double start_time = this->last_updated_;
170 double finish_time = surf_get_clock();
172 // We may have start == finish if the past consumption was updated since the simcall was started
173 // for example if 2 actors requested to update the same host's consumption in a given scheduling round.
175 // Even in this case, we need to save the pstate for the next call (after this if),
176 // which may have changed since that recent update.
177 if (start_time < finish_time) {
178 double previous_energy = this->total_energy_;
180 double instantaneous_consumption = this->get_current_watts_value();
182 double energy_this_step = instantaneous_consumption * (finish_time - start_time);
184 // TODO Trace: Trace energy_this_step from start_time to finish_time in host->getName()
186 this->total_energy_ = previous_energy + energy_this_step;
187 this->last_updated_ = finish_time;
189 XBT_DEBUG("[update_energy of %s] period=[%.8f-%.8f]; current speed=%.2E flop/s (pstate %i); total consumption before: consumption change: %.8f J -> added now: %.8f J",
190 host_->get_cname(), start_time, finish_time, host_->pimpl_cpu->get_pstate_peak_speed(this->pstate_), this->pstate_, previous_energy,
194 /* Save data for the upcoming time interval: whether it's on/off and the pstate if it's on */
195 this->pstate_ = host_->is_on() ? host_->get_pstate() : pstate_off_;
198 HostEnergy::HostEnergy(simgrid::s4u::Host* ptr) : host_(ptr), last_updated_(surf_get_clock())
200 init_watts_range_list();
202 const char* off_power_str = host_->get_property("watt_off");
203 if (off_power_str != nullptr) {
205 this->watts_off_ = std::stod(std::string(off_power_str));
206 } catch (const std::invalid_argument&) {
207 throw std::invalid_argument(std::string("Invalid value for property watt_off of host ") + host_->get_cname() +
208 ": " + off_power_str);
211 /* watts_off is 0 by default */
214 HostEnergy::~HostEnergy() = default;
216 double HostEnergy::get_idle_consumption()
218 xbt_assert(not power_range_watts_list_.empty(), "No power range properties specified for host %s",
221 return power_range_watts_list_[0].idle_;
224 double HostEnergy::get_watt_min_at(int pstate)
226 xbt_assert(not power_range_watts_list_.empty(), "No power range properties specified for host %s",
228 return power_range_watts_list_[pstate].min_;
231 double HostEnergy::get_watt_max_at(int pstate)
233 xbt_assert(not power_range_watts_list_.empty(), "No power range properties specified for host %s",
235 return power_range_watts_list_[pstate].max_;
238 /** @brief Computes the power consumed by the host according to the current situation
240 * - If the host is off, that's the watts_off value
241 * - if it's on, take the current pstate and the current processor load into account */
242 double HostEnergy::get_current_watts_value()
244 if (this->pstate_ == pstate_off_) // The host is off (or was off at the beginning of this time interval)
245 return this->watts_off_;
247 double current_speed = host_->get_pstate_speed(this->pstate_);
251 if (current_speed <= 0)
252 // Some users declare a pstate of speed 0 flops (e.g., to model boot time).
253 // We consider that the machine is then fully loaded. That's arbitrary but it avoids a NaN
256 cpu_load = host_->pimpl_cpu->get_constraint()->get_usage() / current_speed;
258 /** Divide by the number of cores here **/
259 cpu_load /= host_->pimpl_cpu->get_core_count();
261 if (cpu_load > 1) // A machine with a load > 1 consumes as much as a fully loaded machine, not more
264 host_was_used_ = true;
267 /* The problem with this model is that the load is always 0 or 1, never something less.
268 * Another possibility could be to model the total energy as
270 * X/(X+Y)*W_idle + Y/(X+Y)*W_burn
272 * where X is the amount of idling cores, and Y the amount of computing cores.
274 return get_current_watts_value(cpu_load);
277 /** @brief Computes the power that the host would consume at the provided processor load
279 * Whether the host is ON or OFF is not taken into account.
281 double HostEnergy::get_current_watts_value(double cpu_load)
283 xbt_assert(not power_range_watts_list_.empty(), "No power range properties specified for host %s",
286 /* Return watts_off if pstate == pstate_off (ie, if the host is off) */
287 if (this->pstate_ == pstate_off_) {
291 /* min_power corresponds to the power consumed when only one core is active */
292 /* max_power is the power consumed at 100% cpu load */
293 auto range = power_range_watts_list_.at(this->pstate_);
294 double current_power;
299 if (cpu_load > 0) { /* Something is going on, the machine is not idle */
300 min_power = range.min_;
301 max_power = range.max_;
304 * The min_power states how much we consume when only one single
305 * core is working. This means that when cpu_load == 1/coreCount, then
306 * current_power == min_power.
308 * The maximum must be reached when all cores are working (but 1 core was
309 * already accounted for by min_power)
310 * i.e., we need min_power + (maxCpuLoad-1/coreCount)*power_slope == max_power
311 * (maxCpuLoad is by definition 1)
313 int coreCount = host_->get_core_count();
314 double coreReciprocal = static_cast<double>(1) / static_cast<double>(coreCount);
316 power_slope = (max_power - min_power) / (1 - coreReciprocal);
318 power_slope = 0; // Should be 0, since max_power == min_power (in this case)
320 current_power = min_power + (cpu_load - coreReciprocal) * power_slope;
321 } else { /* Our machine is idle, take the dedicated value! */
325 current_power = range.idle_;
328 XBT_DEBUG("[get_current_watts] pstate=%i, min_power=%f, max_power=%f, slope=%f", this->pstate_, min_power, max_power, power_slope);
329 XBT_DEBUG("[get_current_watts] Current power (watts) = %f, load = %f", current_power, cpu_load);
331 return current_power;
334 double HostEnergy::get_consumed_energy()
336 if (last_updated_ < surf_get_clock()) // We need to simcall this as it modifies the environment
337 simgrid::simix::simcall(std::bind(&HostEnergy::update, this));
339 return total_energy_;
342 void HostEnergy::init_watts_range_list()
344 const char* all_power_values_str = host_->get_property("watt_per_state");
345 if (all_power_values_str == nullptr)
348 std::vector<std::string> all_power_values;
349 boost::split(all_power_values, all_power_values_str, boost::is_any_of(","));
350 XBT_DEBUG("%s: profile: %s, cores: %d", host_->get_cname(), all_power_values_str, host_->get_core_count());
353 for (auto const& current_power_values_str : all_power_values) {
354 /* retrieve the power values associated with the current pstate */
355 std::vector<std::string> current_power_values;
356 boost::split(current_power_values, current_power_values_str, boost::is_any_of(":"));
357 if (host_->get_core_count() == 1) {
358 xbt_assert(current_power_values.size() == 2 || current_power_values.size() == 3,
359 "Power properties incorrectly defined for host %s."
360 "It should be 'Idle:FullSpeed' power values because you have one core only.",
362 if (current_power_values.size() == 2) {
363 // In this case, 1core == AllCores
364 current_power_values.push_back(current_power_values.at(1));
365 } else { // size == 3
366 current_power_values[1] = current_power_values.at(2);
367 current_power_values[2] = current_power_values.at(2);
368 static bool displayed_warning = false;
369 if (not displayed_warning) { // Otherwise we get in the worst case no_pstate*no_hosts warnings
370 XBT_WARN("Host %s is a single-core machine and part of the power profile is '%s'"
371 ", which is in the 'Idle:OneCore:AllCores' format."
372 " Here, only the value for 'AllCores' is used.", host_->get_cname(), current_power_values_str.c_str());
373 displayed_warning = true;
377 xbt_assert(current_power_values.size() == 3,
378 "Power properties incorrectly defined for host %s."
379 "It should be 'Idle:OneCore:AllCores' power values because you have more than one core.",
383 /* min_power corresponds to the idle power (cpu load = 0) */
384 /* max_power is the power consumed at 100% cpu load */
385 char* msg_idle = bprintf("Invalid idle value for pstate %d on host %s: %%s", i, host_->get_cname());
386 char* msg_min = bprintf("Invalid OneCore value for pstate %d on host %s: %%s", i, host_->get_cname());
387 char* msg_max = bprintf("Invalid AllCores value for pstate %d on host %s: %%s", i, host_->get_cname());
388 PowerRange range(xbt_str_parse_double((current_power_values.at(0)).c_str(), msg_idle),
389 xbt_str_parse_double((current_power_values.at(1)).c_str(), msg_min),
390 xbt_str_parse_double((current_power_values.at(2)).c_str(), msg_max));
391 power_range_watts_list_.push_back(range);
398 } // namespace plugin
399 } // namespace simgrid
401 using simgrid::plugin::HostEnergy;
403 /* **************************** events callback *************************** */
404 static void on_creation(simgrid::s4u::Host& host)
406 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
409 // TODO Trace: set to zero the energy variable associated to host->getName()
411 host.extension_set(new HostEnergy(&host));
414 static void on_action_state_change(simgrid::kernel::resource::CpuAction const& action,
415 simgrid::kernel::resource::Action::State /*previous*/)
417 for (simgrid::kernel::resource::Cpu* const& cpu : action.cpus()) {
418 simgrid::s4u::Host* host = cpu->get_host();
419 if (host != nullptr) {
421 // If it's a VM, take the corresponding PM
422 simgrid::s4u::VirtualMachine* vm = dynamic_cast<simgrid::s4u::VirtualMachine*>(host);
423 if (vm) // If it's a VM, take the corresponding PM
426 // Get the host_energy extension for the relevant host
427 HostEnergy* host_energy = host->extension<HostEnergy>();
429 if (host_energy->last_updated_ < surf_get_clock())
430 host_energy->update();
435 /* This callback is fired either when the host changes its state (on/off) ("onStateChange") or its speed
436 * (because the user changed the pstate, or because of external trace events) ("onSpeedChange") */
437 static void on_host_change(simgrid::s4u::Host const& host)
439 if (dynamic_cast<simgrid::s4u::VirtualMachine const*>(&host)) // Ignore virtual machines
442 HostEnergy* host_energy = host.extension<HostEnergy>();
444 host_energy->update();
447 static void on_host_destruction(simgrid::s4u::Host const& host)
449 if (dynamic_cast<simgrid::s4u::VirtualMachine const*>(&host)) // Ignore virtual machines
452 XBT_INFO("Energy consumption of host %s: %f Joules", host.get_cname(),
453 host.extension<HostEnergy>()->get_consumed_energy());
456 static void on_simulation_end()
458 std::vector<simgrid::s4u::Host*> hosts = simgrid::s4u::Engine::get_instance()->get_all_hosts();
460 double total_energy = 0.0; // Total energy consumption (whole platform)
461 double used_hosts_energy = 0.0; // Energy consumed by hosts that computed something
462 for (size_t i = 0; i < hosts.size(); i++) {
463 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(hosts[i]) == nullptr) { // Ignore virtual machines
465 double energy = hosts[i]->extension<HostEnergy>()->get_consumed_energy();
466 total_energy += energy;
467 if (hosts[i]->extension<HostEnergy>()->host_was_used_)
468 used_hosts_energy += energy;
471 XBT_INFO("Total energy consumption: %f Joules (used hosts: %f Joules; unused/idle hosts: %f)", total_energy,
472 used_hosts_energy, total_energy - used_hosts_energy);
475 /* **************************** Public interface *************************** */
477 /** @ingroup plugin_energy
478 * @brief Enable host energy plugin
479 * @details Enable energy plugin to get joules consumption of each cpu. Call this function before #MSG_init().
481 void sg_host_energy_plugin_init()
483 if (HostEnergy::EXTENSION_ID.valid())
486 HostEnergy::EXTENSION_ID = simgrid::s4u::Host::extension_create<HostEnergy>();
488 simgrid::s4u::Host::on_creation.connect(&on_creation);
489 simgrid::s4u::Host::on_state_change.connect(&on_host_change);
490 simgrid::s4u::Host::on_speed_change.connect(&on_host_change);
491 simgrid::s4u::Host::on_destruction.connect(&on_host_destruction);
492 simgrid::s4u::on_simulation_end.connect(&on_simulation_end);
493 simgrid::kernel::resource::CpuAction::on_state_change.connect(&on_action_state_change);
494 // We may only have one actor on a node. If that actor executes something like
495 // compute -> recv -> compute
496 // the recv operation will not trigger a "CpuAction::on_state_change". This means
497 // that the next trigger would be the 2nd compute, hence ignoring the idle time
498 // during the recv call. By updating at the beginning of a compute, we can
499 // fix that. (If the cpu is not idle, this is not required.)
500 simgrid::kernel::activity::ExecImpl::on_creation.connect([](simgrid::kernel::activity::ExecImpl const& activity) {
501 if (activity.get_host_number() == 1) { // We only run on one host
502 simgrid::s4u::Host* host = activity.get_host();
503 simgrid::s4u::VirtualMachine* vm = dynamic_cast<simgrid::s4u::VirtualMachine*>(host);
506 xbt_assert(host != nullptr);
507 host->extension<HostEnergy>()->update();
512 /** @ingroup plugin_energy
513 * @brief updates the consumption of all hosts
515 * After this call, sg_host_get_consumed_energy() will not interrupt your process
516 * (until after the next clock update).
518 void sg_host_energy_update_all()
520 simgrid::simix::simcall([]() {
521 std::vector<simgrid::s4u::Host*> list = simgrid::s4u::Engine::get_instance()->get_all_hosts();
522 for (auto const& host : list)
523 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(host) == nullptr) { // Ignore virtual machines
524 xbt_assert(host != nullptr);
525 host->extension<HostEnergy>()->update();
530 /** @ingroup plugin_energy
531 * @brief Returns the total energy consumed by the host so far (in Joules)
533 * Please note that since the consumption is lazily updated, it may require a simcall to update it.
534 * The result is that the actor requesting this value will be interrupted,
535 * the value will be updated in kernel mode before returning the control to the requesting actor.
537 double sg_host_get_consumed_energy(sg_host_t host)
539 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
540 "The Energy plugin is not active. Please call sg_host_energy_plugin_init() during initialization.");
541 return host->extension<HostEnergy>()->get_consumed_energy();
544 /** @ingroup plugin_energy
545 * @brief Get the amount of watt dissipated when the host is idling
547 double sg_host_get_idle_consumption(sg_host_t host)
549 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
550 "The Energy plugin is not active. Please call sg_host_energy_plugin_init() during initialization.");
551 return host->extension<HostEnergy>()->get_idle_consumption();
554 /** @ingroup plugin_energy
555 * @brief Get the amount of watt dissipated at the given pstate when the host is idling
557 double sg_host_get_wattmin_at(sg_host_t host, int pstate)
559 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
560 "The Energy plugin is not active. Please call sg_host_energy_plugin_init() during initialization.");
561 return host->extension<HostEnergy>()->get_watt_min_at(pstate);
563 /** @ingroup plugin_energy
564 * @brief Returns the amount of watt dissipated at the given pstate when the host burns CPU at 100%
566 double sg_host_get_wattmax_at(sg_host_t host, int pstate)
568 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
569 "The Energy plugin is not active. Please call sg_host_energy_plugin_init() during initialization.");
570 return host->extension<HostEnergy>()->get_watt_max_at(pstate);
573 /** @ingroup plugin_energy
574 * @brief Returns the current consumption of the host
576 double sg_host_get_current_consumption(sg_host_t host)
578 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
579 "The Energy plugin is not active. Please call sg_host_energy_plugin_init() during initialization.");
580 return host->extension<HostEnergy>()->get_current_watts_value();