1 /* Copyright (c) 2010-2017. 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/simix.hpp"
9 #include "src/plugins/vm/VirtualMachineImpl.hpp"
10 #include "src/surf/cpu_interface.hpp"
12 #include "simgrid/s4u/Engine.hpp"
14 #include <boost/algorithm/string/classification.hpp>
15 #include <boost/algorithm/string/split.hpp>
20 /** @addtogroup plugin_energy
22 This is the energy plugin, enabling to account not only for computation time, but also for the dissipated energy in the
24 To activate this plugin, first call sg_host_energy_plugin_init() before your #MSG_init(), and then use
25 MSG_host_get_consumed_energy() to retrieve the consumption of a given host.
27 When the host is on, this energy consumption naturally depends on both the current CPU load and the host energy profile.
28 According to our measurements, the consumption is somehow linear in the amount of cores at full speed, with an
29 abnormality when all the cores are idle. The full details are in
30 <a href="https://hal.inria.fr/hal-01523608">our scientific paper</a> on that topic.
32 As a result, our energy model takes 4 parameters:
34 - \b Idle: instantaneous consumption (in Watt) when your host is up and running, but without anything to do.
35 - \b OneCore: instantaneous consumption (in Watt) when only one core is active, at 100%.
36 - \b AllCores: instantaneous consumption (in Watt) when all cores of the host are at 100%.
37 - \b Off: instantaneous consumption (in Watt) when the host is turned off.
39 Here is an example of XML declaration:
42 <host id="HostA" power="100.0Mf" cores="4">
43 <prop id="watt_per_state" value="100.0:120.0:200.0" />
44 <prop id="watt_off" value="10" />
48 This example gives the following parameters: \b Off is 10 Watts; \b Idle is 100 Watts; \b OneCore is 120 Watts and \b
49 AllCores is 200 Watts.
50 This is enough to compute the consumption as a function of the amount of loaded cores:
53 <tr><th>#Cores loaded</th><th>Consumption</th><th>Explanation</th></tr>
54 <tr><td>0</td><td> 100 Watts</td><td>Idle value</td></tr>
55 <tr><td>1</td><td> 120 Watts</td><td>OneCore value</td></tr>
56 <tr><td>2</td><td> 147 Watts</td><td>linear extrapolation between OneCore and AllCores</td></tr>
57 <tr><td>3</td><td> 173 Watts</td><td>linear extrapolation between OneCore and AllCores</td></tr>
58 <tr><td>4</td><td> 200 Watts</td><td>AllCores value</td></tr>
61 ### What if a given core is only at load 50%?
63 This is impossible in SimGrid because we recompute everything each time that the CPU starts or stops doing something.
64 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
65 the time, and our model holds.
67 ### What if the host has only one core?
69 In this case, the parameters \b OneCore and \b AllCores are obviously the same.
70 Actually, SimGrid expect an energetic profile formatted as 'Idle:Running' for mono-cores hosts.
71 If you insist on passing 3 parameters in this case, then you must have the same value for \b OneCore and \b AllCores.
74 <host id="HostC" power="100.0Mf" cores="1">
75 <prop id="watt_per_state" value="95.0:200.0" /> <!-- we may have used '95:200:200' instead -->
76 <prop id="watt_off" value="10" />
80 ### How does DVFS interact with the host energy model?
82 If your host has several DVFS levels (several pstates), then you should give the energetic profile of each pstate level:
85 <host id="HostC" power="100.0Mf,50.0Mf,20.0Mf" cores="4">
86 <prop id="watt_per_state" value="95.0:120.0:200.0, 93.0:115.0:170.0, 90.0:110.0:150.0" />
87 <prop id="watt_off" value="10" />
91 This encodes the following values
93 <tr><th>pstate</th><th>Performance</th><th>Idle</th><th>OneCore</th><th>AllCores</th></tr>
94 <tr><td>0</td><td>100 Mflop/s</td><td>95 Watts</td><td>120 Watts</td><td>200 Watts</td></tr>
95 <tr><td>1</td><td>50 Mflop/s</td><td>93 Watts</td><td>115 Watts</td><td>170 Watts</td></tr>
96 <tr><td>2</td><td>20 Mflop/s</td><td>90 Watts</td><td>110 Watts</td><td>150 Watts</td></tr>
99 To change the pstate of a given CPU, use the following functions:
100 #MSG_host_get_nb_pstates(), simgrid#s4u#Host#setPstate(), #MSG_host_get_power_peak_at().
102 ### How accurate are these models?
104 This model cannot be more accurate than your instantiation: with the default values, your result will not be accurate at
105 all. You can still get accurate energy prediction, provided that you carefully instantiate the model.
106 The first step is to ensure that your timing prediction match perfectly. But this is only the first step of the path,
107 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
108 before you can get accurate energy predictions.
111 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_energy, surf, "Logging specific to the SURF energy plugin");
122 PowerRange(double idle, double min, double max) : idle(idle), min(min), max(max) {}
127 static simgrid::xbt::Extension<simgrid::s4u::Host, HostEnergy> EXTENSION_ID;
129 explicit HostEnergy(simgrid::s4u::Host* ptr);
132 double getCurrentWattsValue();
133 double getCurrentWattsValue(double cpu_load);
134 double getConsumedEnergy();
135 double getWattMinAt(int pstate);
136 double getWattMaxAt(int pstate);
140 void initWattsRangeList();
141 simgrid::s4u::Host* host = nullptr;
142 std::vector<PowerRange>
143 power_range_watts_list; /*< List of (min_power,max_power) pairs corresponding to each cpu pstate */
145 /* We need to keep track of what pstate has been used, as we will sometimes be notified only *after* a pstate has been
146 * used (but we need to update the energy consumption with the old pstate!)
149 const int pstate_off = -1;
152 double watts_off = 0.0; /*< Consumption when the machine is turned off (shutdown) */
153 double total_energy = 0.0; /*< Total energy consumed by the host */
154 double last_updated; /*< Timestamp of the last energy update event*/
157 simgrid::xbt::Extension<simgrid::s4u::Host, HostEnergy> HostEnergy::EXTENSION_ID;
159 /* Computes the consumption so far. Called lazily on need. */
160 void HostEnergy::update()
162 double start_time = this->last_updated;
163 double finish_time = surf_get_clock();
165 if (start_time < finish_time) {
166 double previous_energy = this->total_energy;
168 double instantaneous_consumption = this->getCurrentWattsValue();
170 double energy_this_step = instantaneous_consumption * (finish_time - start_time);
172 // TODO Trace: Trace energy_this_step from start_time to finish_time in host->getName()
174 this->total_energy = previous_energy + energy_this_step;
175 this->last_updated = finish_time;
177 XBT_DEBUG("[update_energy of %s] period=[%.2f-%.2f]; current power peak=%.0E flop/s; consumption change: %.2f J -> "
179 host->getCname(), start_time, finish_time, host->pimpl_cpu->speed_.peak, previous_energy,
183 /* Save data for the upcoming time interval: whether it's on/off and the pstate if it's on */
184 this->pstate = host->isOn() ? host->getPstate() : pstate_off;
187 HostEnergy::HostEnergy(simgrid::s4u::Host* ptr) : host(ptr), last_updated(surf_get_clock())
189 initWattsRangeList();
191 const char* off_power_str = host->getProperty("watt_off");
192 if (off_power_str != nullptr) {
194 this->watts_off = std::stod(std::string(off_power_str));
195 } catch (std::invalid_argument& ia) {
196 throw std::invalid_argument(std::string("Invalid value for property watt_off of host ") + host->getCname() +
197 ": " + off_power_str);
200 /* watts_off is 0 by default */
203 HostEnergy::~HostEnergy() = default;
205 double HostEnergy::getWattMinAt(int pstate)
207 xbt_assert(not power_range_watts_list.empty(), "No power range properties specified for host %s", host->getCname());
208 return power_range_watts_list[pstate].min;
211 double HostEnergy::getWattMaxAt(int pstate)
213 xbt_assert(not power_range_watts_list.empty(), "No power range properties specified for host %s", host->getCname());
214 return power_range_watts_list[pstate].max;
217 /** @brief Computes the power consumed by the host according to the current situation
219 * - If the host is off, that's the watts_off value
220 * - if it's on, take the current pstate and the current processor load into account */
221 double HostEnergy::getCurrentWattsValue()
223 if (this->pstate == pstate_off) // The host is off (or was off at the beginning of this time interval)
224 return this->watts_off;
226 double current_speed = host->getSpeed();
229 // We may have start == finish if the past consumption was updated since the simcall was started
230 // for example if 2 actors requested to update the same host's consumption in a given scheduling round.
232 // Even in this case, we need to save the pstate for the next call (after this big if),
233 // which may have changed since that recent update.
235 if (current_speed <= 0)
236 // Some users declare a pstate of speed 0 flops (e.g., to model boot time).
237 // We consider that the machine is then fully loaded. That's arbitrary but it avoids a NaN
240 cpu_load = host->pimpl_cpu->constraint()->get_usage() / current_speed;
242 /** Divide by the number of cores here **/
243 cpu_load /= host->pimpl_cpu->coreCount();
245 if (cpu_load > 1) // A machine with a load > 1 consumes as much as a fully loaded machine, not more
248 /* The problem with this model is that the load is always 0 or 1, never something less.
249 * Another possibility could be to model the total energy as
251 * X/(X+Y)*W_idle + Y/(X+Y)*W_burn
253 * where X is the amount of idling cores, and Y the amount of computing cores.
255 return getCurrentWattsValue(cpu_load);
258 /** @brief Computes the power that the host would consume at the provided processor load
260 * Whether the host is ON or OFF is not taken into account.
262 double HostEnergy::getCurrentWattsValue(double cpu_load)
264 xbt_assert(not power_range_watts_list.empty(), "No power range properties specified for host %s", host->getCname());
267 * * Return watts_off if pstate == pstate_off
268 * * this happens when host is off
270 if (this->pstate == pstate_off) {
274 /* min_power corresponds to the power consumed when only one core is active */
275 /* max_power is the power consumed at 100% cpu load */
276 auto range = power_range_watts_list.at(this->pstate);
277 double current_power = 0;
278 double min_power = 0;
279 double max_power = 0;
280 double power_slope = 0;
282 if (cpu_load > 0) { /* Something is going on, the machine is not idle */
283 double min_power = range.min;
284 double max_power = range.max;
287 * The min_power states how much we consume when only one single
288 * core is working. This means that when cpu_load == 1/coreCount, then
289 * current_power == min_power.
291 * The maximum must be reached when all cores are working (but 1 core was
292 * already accounted for by min_power)
293 * i.e., we need min_power + (maxCpuLoad-1/coreCount)*power_slope == max_power
294 * (maxCpuLoad is by definition 1)
297 int coreCount = host->getCoreCount();
298 double coreReciprocal = static_cast<double>(1) / static_cast<double>(coreCount);
300 power_slope = (max_power - min_power) / (1 - coreReciprocal);
302 power_slope = 0; // Should be 0, since max_power == min_power (in this case)
304 current_power = min_power + (cpu_load - coreReciprocal) * power_slope;
305 } else { /* Our machine is idle, take the dedicated value! */
306 current_power = range.idle;
309 XBT_DEBUG("[get_current_watts] min_power=%f, max_power=%f, slope=%f", min_power, max_power, power_slope);
310 XBT_DEBUG("[get_current_watts] Current power (watts) = %f, load = %f", current_power, cpu_load);
312 return current_power;
315 double HostEnergy::getConsumedEnergy()
317 if (last_updated < surf_get_clock()) // We need to simcall this as it modifies the environment
318 simgrid::simix::kernelImmediate(std::bind(&HostEnergy::update, this));
323 void HostEnergy::initWattsRangeList()
325 const char* all_power_values_str = host->getProperty("watt_per_state");
326 if (all_power_values_str == nullptr)
329 std::vector<std::string> all_power_values;
330 boost::split(all_power_values, all_power_values_str, boost::is_any_of(","));
331 XBT_DEBUG("%s: profile: %s, cores: %d", host->getCname(), all_power_values_str, host->getCoreCount());
334 for (auto const& current_power_values_str : all_power_values) {
335 /* retrieve the power values associated with the current pstate */
336 std::vector<std::string> current_power_values;
337 boost::split(current_power_values, current_power_values_str, boost::is_any_of(":"));
338 if (host->getCoreCount() == 1) {
339 xbt_assert(current_power_values.size() == 2 || current_power_values.size() == 3,
340 "Power properties incorrectly defined for host %s."
341 "It should be 'Idle:FullSpeed' power values because you have one core only.",
343 if (current_power_values.size() == 2) {
344 // In this case, 1core == AllCores
345 current_power_values.push_back(current_power_values.at(1));
346 } else { // size == 3
347 xbt_assert((current_power_values.at(1)) == (current_power_values.at(2)),
348 "Power properties incorrectly defined for host %s.\n"
349 "The energy profile of mono-cores should be formatted as 'Idle:FullSpeed' only.\n"
350 "If you go for a 'Idle:OneCore:AllCores' power profile on mono-cores, then OneCore and AllCores "
355 xbt_assert(current_power_values.size() == 3,
356 "Power properties incorrectly defined for host %s."
357 "It should be 'Idle:OneCore:AllCores' power values because you have more than one core.",
361 /* min_power corresponds to the idle power (cpu load = 0) */
362 /* max_power is the power consumed at 100% cpu load */
363 char* msg_idle = bprintf("Invalid idle value for pstate %d on host %s: %%s", i, host->getCname());
364 char* msg_min = bprintf("Invalid OneCore value for pstate %d on host %s: %%s", i, host->getCname());
365 char* msg_max = bprintf("Invalid AllCores value for pstate %d on host %s: %%s", i, host->getCname());
366 PowerRange range(xbt_str_parse_double((current_power_values.at(0)).c_str(), msg_idle),
367 xbt_str_parse_double((current_power_values.at(1)).c_str(), msg_min),
368 xbt_str_parse_double((current_power_values.at(2)).c_str(), msg_max));
369 power_range_watts_list.push_back(range);
379 using simgrid::plugin::HostEnergy;
381 /* **************************** events callback *************************** */
382 static void onCreation(simgrid::s4u::Host& host)
384 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
387 // TODO Trace: set to zero the energy variable associated to host->getName()
389 host.extension_set(new HostEnergy(&host));
392 static void onActionStateChange(simgrid::surf::CpuAction* action, simgrid::surf::Action::State previous)
394 for (simgrid::surf::Cpu* const& cpu : action->cpus()) {
395 simgrid::s4u::Host* host = cpu->getHost();
396 if (host != nullptr) {
398 // If it's a VM, take the corresponding PM
399 simgrid::s4u::VirtualMachine* vm = dynamic_cast<simgrid::s4u::VirtualMachine*>(host);
400 if (vm) // If it's a VM, take the corresponding PM
403 // Get the host_energy extension for the relevant host
404 HostEnergy* host_energy = host->extension<HostEnergy>();
406 if (host_energy->last_updated < surf_get_clock())
407 host_energy->update();
412 /* This callback is fired either when the host changes its state (on/off) ("onStateChange") or its speed
413 * (because the user changed the pstate, or because of external trace events) ("onSpeedChange") */
414 static void onHostChange(simgrid::s4u::Host& host)
416 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
419 HostEnergy* host_energy = host.extension<HostEnergy>();
421 host_energy->update();
424 static void onHostDestruction(simgrid::s4u::Host& host)
426 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
429 XBT_INFO("Energy consumption of host %s: %f Joules", host.getCname(),
430 host.extension<HostEnergy>()->getConsumedEnergy());
433 static void onSimulationEnd()
435 sg_host_t* host_list = sg_host_list();
436 int host_count = sg_host_count();
437 double total_energy = 0.0; // Total energy consumption (whole platform)
438 double used_hosts_energy = 0.0; // Energy consumed by hosts that computed something
439 for (int i = 0; i < host_count; i++) {
440 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(host_list[i]) == nullptr) { // Ignore virtual machines
442 bool host_was_used = (sg_host_get_computed_flops(host_list[i]) != 0);
443 double energy = host_list[i]->extension<HostEnergy>()->getConsumedEnergy();
444 total_energy += energy;
446 used_hosts_energy += energy;
449 XBT_INFO("Total energy consumption: %f Joules (used hosts: %f Joules; unused/idle hosts: %f)",
450 total_energy, used_hosts_energy, total_energy - used_hosts_energy);
454 /* **************************** Public interface *************************** */
457 /** \ingroup plugin_energy
458 * \brief Enable host energy plugin
459 * \details Enable energy plugin to get joules consumption of each cpu. Call this function before #MSG_init().
461 void sg_host_energy_plugin_init()
463 if (HostEnergy::EXTENSION_ID.valid())
466 sg_host_load_plugin_init();
468 HostEnergy::EXTENSION_ID = simgrid::s4u::Host::extension_create<HostEnergy>();
470 simgrid::s4u::Host::onCreation.connect(&onCreation);
471 simgrid::s4u::Host::onStateChange.connect(&onHostChange);
472 simgrid::s4u::Host::onSpeedChange.connect(&onHostChange);
473 simgrid::s4u::Host::onDestruction.connect(&onHostDestruction);
474 simgrid::s4u::onSimulationEnd.connect(&onSimulationEnd);
475 simgrid::surf::CpuAction::onStateChange.connect(&onActionStateChange);
478 /** @ingroup plugin_energy
479 * @brief updates the consumption of all hosts
481 * After this call, sg_host_get_consumed_energy() will not interrupt your process
482 * (until after the next clock update).
484 void sg_host_energy_update_all()
486 simgrid::simix::kernelImmediate([]() {
487 std::vector<simgrid::s4u::Host*> list;
488 simgrid::s4u::Engine::getInstance()->getHostList(&list);
489 for (auto const& host : list)
490 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(host) == nullptr) // Ignore virtual machines
491 host->extension<HostEnergy>()->update();
495 /** @ingroup plugin_energy
496 * @brief Returns the total energy consumed by the host so far (in Joules)
498 * Please note that since the consumption is lazily updated, it may require a simcall to update it.
499 * The result is that the actor requesting this value will be interrupted,
500 * the value will be updated in kernel mode before returning the control to the requesting actor.
502 double sg_host_get_consumed_energy(sg_host_t host)
504 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
505 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
506 return host->extension<HostEnergy>()->getConsumedEnergy();
509 /** @ingroup plugin_energy
510 * @brief Get the amount of watt dissipated at the given pstate when the host is idling
512 double sg_host_get_wattmin_at(sg_host_t host, int pstate)
514 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
515 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
516 return host->extension<HostEnergy>()->getWattMinAt(pstate);
518 /** @ingroup plugin_energy
519 * @brief Returns the amount of watt dissipated at the given pstate when the host burns CPU at 100%
521 double sg_host_get_wattmax_at(sg_host_t host, int pstate)
523 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
524 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
525 return host->extension<HostEnergy>()->getWattMaxAt(pstate);
528 /** @ingroup plugin_energy
529 * @brief Returns the current consumption of the host
531 double sg_host_get_current_consumption(sg_host_t host)
533 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
534 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
535 return host->extension<HostEnergy>()->getCurrentWattsValue();