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 "src/plugins/vm/VirtualMachineImpl.hpp"
9 #include "src/surf/cpu_interface.hpp"
10 #include "simgrid/s4u/Engine.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 getCurrentWattsValue();
128 double getCurrentWattsValue(double cpu_load);
129 double getConsumedEnergy();
130 double getWattMinAt(int pstate);
131 double getWattMaxAt(int pstate);
135 void initWattsRangeList();
136 simgrid::s4u::Host* host = nullptr;
137 std::vector<PowerRange>
138 power_range_watts_list; /*< List of (min_power,max_power) pairs corresponding to each cpu pstate */
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->getCurrentWattsValue();
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->speed_.peak, 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->getPstate() : pstate_off;
182 HostEnergy::HostEnergy(simgrid::s4u::Host* ptr) : host(ptr), last_updated(surf_get_clock())
184 initWattsRangeList();
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::getWattMinAt(int pstate)
202 xbt_assert(not power_range_watts_list.empty(), "No power range properties specified for host %s", host->get_cname());
203 return power_range_watts_list[pstate].min;
206 double HostEnergy::getWattMaxAt(int pstate)
208 xbt_assert(not power_range_watts_list.empty(), "No power range properties specified for host %s", host->get_cname());
209 return power_range_watts_list[pstate].max;
212 /** @brief Computes the power consumed by the host according to the current situation
214 * - If the host is off, that's the watts_off value
215 * - if it's on, take the current pstate and the current processor load into account */
216 double HostEnergy::getCurrentWattsValue()
218 if (this->pstate == pstate_off) // The host is off (or was off at the beginning of this time interval)
219 return this->watts_off;
221 double current_speed = host->getSpeed();
224 // We may have start == finish if the past consumption was updated since the simcall was started
225 // for example if 2 actors requested to update the same host's consumption in a given scheduling round.
227 // Even in this case, we need to save the pstate for the next call (after this big if),
228 // which may have changed since that recent update.
230 if (current_speed <= 0)
231 // Some users declare a pstate of speed 0 flops (e.g., to model boot time).
232 // We consider that the machine is then fully loaded. That's arbitrary but it avoids a NaN
235 cpu_load = host->pimpl_cpu->get_constraint()->get_usage() / current_speed;
237 /** Divide by the number of cores here **/
238 cpu_load /= host->pimpl_cpu->get_cores_count();
240 if (cpu_load > 1) // A machine with a load > 1 consumes as much as a fully loaded machine, not more
243 /* The problem with this model is that the load is always 0 or 1, never something less.
244 * Another possibility could be to model the total energy as
246 * X/(X+Y)*W_idle + Y/(X+Y)*W_burn
248 * where X is the amount of idling cores, and Y the amount of computing cores.
250 return getCurrentWattsValue(cpu_load);
253 /** @brief Computes the power that the host would consume at the provided processor load
255 * Whether the host is ON or OFF is not taken into account.
257 double HostEnergy::getCurrentWattsValue(double cpu_load)
259 xbt_assert(not power_range_watts_list.empty(), "No power range properties specified for host %s", host->get_cname());
261 /* Return watts_off if pstate == pstate_off (ie, if the host is off) */
262 if (this->pstate == pstate_off) {
266 /* min_power corresponds to the power consumed when only one core is active */
267 /* max_power is the power consumed at 100% cpu load */
268 auto range = power_range_watts_list.at(this->pstate);
269 double current_power = 0;
270 double min_power = 0;
271 double max_power = 0;
272 double power_slope = 0;
274 if (cpu_load > 0) { /* Something is going on, the machine is not idle */
275 double min_power = range.min;
276 double max_power = range.max;
279 * The min_power states how much we consume when only one single
280 * core is working. This means that when cpu_load == 1/coreCount, then
281 * current_power == min_power.
283 * The maximum must be reached when all cores are working (but 1 core was
284 * already accounted for by min_power)
285 * i.e., we need min_power + (maxCpuLoad-1/coreCount)*power_slope == max_power
286 * (maxCpuLoad is by definition 1)
289 int coreCount = host->getCoreCount();
290 double coreReciprocal = static_cast<double>(1) / static_cast<double>(coreCount);
292 power_slope = (max_power - min_power) / (1 - coreReciprocal);
294 power_slope = 0; // Should be 0, since max_power == min_power (in this case)
296 current_power = min_power + (cpu_load - coreReciprocal) * power_slope;
297 } else { /* Our machine is idle, take the dedicated value! */
298 current_power = range.idle;
301 XBT_DEBUG("[get_current_watts] min_power=%f, max_power=%f, slope=%f", min_power, max_power, power_slope);
302 XBT_DEBUG("[get_current_watts] Current power (watts) = %f, load = %f", current_power, cpu_load);
304 return current_power;
307 double HostEnergy::getConsumedEnergy()
309 if (last_updated < surf_get_clock()) // We need to simcall this as it modifies the environment
310 simgrid::simix::simcall(std::bind(&HostEnergy::update, this));
315 void HostEnergy::initWattsRangeList()
317 const char* all_power_values_str = host->get_property("watt_per_state");
318 if (all_power_values_str == nullptr)
321 std::vector<std::string> all_power_values;
322 boost::split(all_power_values, all_power_values_str, boost::is_any_of(","));
323 XBT_DEBUG("%s: profile: %s, cores: %d", host->get_cname(), all_power_values_str, host->getCoreCount());
326 for (auto const& current_power_values_str : all_power_values) {
327 /* retrieve the power values associated with the current pstate */
328 std::vector<std::string> current_power_values;
329 boost::split(current_power_values, current_power_values_str, boost::is_any_of(":"));
330 if (host->getCoreCount() == 1) {
331 xbt_assert(current_power_values.size() == 2 || current_power_values.size() == 3,
332 "Power properties incorrectly defined for host %s."
333 "It should be 'Idle:FullSpeed' power values because you have one core only.",
335 if (current_power_values.size() == 2) {
336 // In this case, 1core == AllCores
337 current_power_values.push_back(current_power_values.at(1));
338 } else { // size == 3
339 xbt_assert((current_power_values.at(1)) == (current_power_values.at(2)),
340 "Power properties incorrectly defined for host %s.\n"
341 "The energy profile of mono-cores should be formatted as 'Idle:FullSpeed' only.\n"
342 "If you go for a 'Idle:OneCore:AllCores' power profile on mono-cores, then OneCore and AllCores "
347 xbt_assert(current_power_values.size() == 3,
348 "Power properties incorrectly defined for host %s."
349 "It should be 'Idle:OneCore:AllCores' power values because you have more than one core.",
353 /* min_power corresponds to the idle power (cpu load = 0) */
354 /* max_power is the power consumed at 100% cpu load */
355 char* msg_idle = bprintf("Invalid idle value for pstate %d on host %s: %%s", i, host->get_cname());
356 char* msg_min = bprintf("Invalid OneCore value for pstate %d on host %s: %%s", i, host->get_cname());
357 char* msg_max = bprintf("Invalid AllCores value for pstate %d on host %s: %%s", i, host->get_cname());
358 PowerRange range(xbt_str_parse_double((current_power_values.at(0)).c_str(), msg_idle),
359 xbt_str_parse_double((current_power_values.at(1)).c_str(), msg_min),
360 xbt_str_parse_double((current_power_values.at(2)).c_str(), msg_max));
361 power_range_watts_list.push_back(range);
371 using simgrid::plugin::HostEnergy;
373 /* **************************** events callback *************************** */
374 static void onCreation(simgrid::s4u::Host& host)
376 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
379 // TODO Trace: set to zero the energy variable associated to host->getName()
381 host.extension_set(new HostEnergy(&host));
384 static void onActionStateChange(simgrid::surf::CpuAction* action, simgrid::kernel::resource::Action::State previous)
386 for (simgrid::surf::Cpu* const& cpu : action->cpus()) {
387 simgrid::s4u::Host* host = cpu->get_host();
388 if (host != nullptr) {
390 // If it's a VM, take the corresponding PM
391 simgrid::s4u::VirtualMachine* vm = dynamic_cast<simgrid::s4u::VirtualMachine*>(host);
392 if (vm) // If it's a VM, take the corresponding PM
395 // Get the host_energy extension for the relevant host
396 HostEnergy* host_energy = host->extension<HostEnergy>();
398 if (host_energy->last_updated < surf_get_clock())
399 host_energy->update();
404 /* This callback is fired either when the host changes its state (on/off) ("onStateChange") or its speed
405 * (because the user changed the pstate, or because of external trace events) ("onSpeedChange") */
406 static void onHostChange(simgrid::s4u::Host& host)
408 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
411 HostEnergy* host_energy = host.extension<HostEnergy>();
413 host_energy->update();
416 static void onHostDestruction(simgrid::s4u::Host& host)
418 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
421 XBT_INFO("Energy consumption of host %s: %f Joules", host.get_cname(),
422 host.extension<HostEnergy>()->getConsumedEnergy());
425 static void onSimulationEnd()
427 std::vector<simgrid::s4u::Host*> hosts = simgrid::s4u::Engine::get_instance()->get_all_hosts();
429 double total_energy = 0.0; // Total energy consumption (whole platform)
430 double used_hosts_energy = 0.0; // Energy consumed by hosts that computed something
431 for (size_t i = 0; i < hosts.size(); i++) {
432 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(hosts[i]) == nullptr) { // Ignore virtual machines
434 bool host_was_used = (sg_host_get_computed_flops(hosts[i]) != 0);
435 double energy = hosts[i]->extension<HostEnergy>()->getConsumedEnergy();
436 total_energy += energy;
438 used_hosts_energy += energy;
441 XBT_INFO("Total energy consumption: %f Joules (used hosts: %f Joules; unused/idle hosts: %f)",
442 total_energy, used_hosts_energy, total_energy - used_hosts_energy);
445 /* **************************** Public interface *************************** */
447 /** \ingroup plugin_energy
448 * \brief Enable host energy plugin
449 * \details Enable energy plugin to get joules consumption of each cpu. Call this function before #MSG_init().
451 void sg_host_energy_plugin_init()
453 if (HostEnergy::EXTENSION_ID.valid())
456 sg_host_load_plugin_init();
458 HostEnergy::EXTENSION_ID = simgrid::s4u::Host::extension_create<HostEnergy>();
460 simgrid::s4u::Host::on_creation.connect(&onCreation);
461 simgrid::s4u::Host::on_state_change.connect(&onHostChange);
462 simgrid::s4u::Host::on_speed_change.connect(&onHostChange);
463 simgrid::s4u::Host::on_destruction.connect(&onHostDestruction);
464 simgrid::s4u::on_simulation_end.connect(&onSimulationEnd);
465 simgrid::surf::CpuAction::onStateChange.connect(&onActionStateChange);
468 /** @ingroup plugin_energy
469 * @brief updates the consumption of all hosts
471 * After this call, sg_host_get_consumed_energy() will not interrupt your process
472 * (until after the next clock update).
474 void sg_host_energy_update_all()
476 simgrid::simix::simcall([]() {
477 std::vector<simgrid::s4u::Host*> list = simgrid::s4u::Engine::get_instance()->get_all_hosts();
478 for (auto const& host : list)
479 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(host) == nullptr) // Ignore virtual machines
480 host->extension<HostEnergy>()->update();
484 /** @ingroup plugin_energy
485 * @brief Returns the total energy consumed by the host so far (in Joules)
487 * Please note that since the consumption is lazily updated, it may require a simcall to update it.
488 * The result is that the actor requesting this value will be interrupted,
489 * the value will be updated in kernel mode before returning the control to the requesting actor.
491 double sg_host_get_consumed_energy(sg_host_t host)
493 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
494 "The Energy plugin is not active. Please call sg_host_energy_plugin_init() during initialization.");
495 return host->extension<HostEnergy>()->getConsumedEnergy();
498 /** @ingroup plugin_energy
499 * @brief Get the amount of watt dissipated at the given pstate when the host is idling
501 double sg_host_get_wattmin_at(sg_host_t host, int pstate)
503 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
504 "The Energy plugin is not active. Please call sg_host_energy_plugin_init() during initialization.");
505 return host->extension<HostEnergy>()->getWattMinAt(pstate);
507 /** @ingroup plugin_energy
508 * @brief Returns the amount of watt dissipated at the given pstate when the host burns CPU at 100%
510 double sg_host_get_wattmax_at(sg_host_t host, int pstate)
512 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
513 "The Energy plugin is not active. Please call sg_host_energy_plugin_init() during initialization.");
514 return host->extension<HostEnergy>()->getWattMaxAt(pstate);
517 /** @ingroup plugin_energy
518 * @brief Returns the current consumption of the host
520 double sg_host_get_current_consumption(sg_host_t host)
522 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
523 "The Energy plugin is not active. Please call sg_host_energy_plugin_init() during initialization.");
524 return host->extension<HostEnergy>()->getCurrentWattsValue();