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/simix.hpp"
8 #include "src/plugins/vm/VirtualMachineImpl.hpp"
9 #include "src/surf/cpu_interface.hpp"
11 #include "simgrid/s4u/Engine.hpp"
13 #include <boost/algorithm/string/classification.hpp>
14 #include <boost/algorithm/string/split.hpp>
19 /** @addtogroup plugin_energy
21 This is the energy plugin, enabling to account not only for computation time, but also for the dissipated energy in the
23 To activate this plugin, first call sg_host_energy_plugin_init() before your #MSG_init(), and then use
24 MSG_host_get_consumed_energy() to retrieve the consumption of a given host.
26 When the host is on, this energy consumption naturally depends on both the current CPU load and the host energy profile.
27 According to our measurements, the consumption is somehow linear in the amount of cores at full speed, with an
28 abnormality when all the cores are idle. The full details are in
29 <a href="https://hal.inria.fr/hal-01523608">our scientific paper</a> on that topic.
31 As a result, our energy model takes 4 parameters:
33 - \b Idle: instantaneous consumption (in Watt) when your host is up and running, but without anything to do.
34 - \b OneCore: instantaneous consumption (in Watt) when only one core is active, at 100%.
35 - \b AllCores: instantaneous consumption (in Watt) when all cores of the host are at 100%.
36 - \b Off: instantaneous consumption (in Watt) when the host is turned off.
38 Here is an example of XML declaration:
41 <host id="HostA" power="100.0Mf" cores="4">
42 <prop id="watt_per_state" value="100.0:120.0:200.0" />
43 <prop id="watt_off" value="10" />
47 This example gives the following parameters: \b Off is 10 Watts; \b Idle is 100 Watts; \b OneCore is 120 Watts and \b
48 AllCores is 200 Watts.
49 This is enough to compute the consumption as a function of the amount of loaded cores:
52 <tr><th>#Cores loaded</th><th>Consumption</th><th>Explanation</th></tr>
53 <tr><td>0</td><td> 100 Watts</td><td>Idle value</td></tr>
54 <tr><td>1</td><td> 120 Watts</td><td>OneCore value</td></tr>
55 <tr><td>2</td><td> 147 Watts</td><td>linear extrapolation between OneCore and AllCores</td></tr>
56 <tr><td>3</td><td> 173 Watts</td><td>linear extrapolation between OneCore and AllCores</td></tr>
57 <tr><td>4</td><td> 200 Watts</td><td>AllCores value</td></tr>
60 ### What if a given core is only at load 50%?
62 This is impossible in SimGrid because we recompute everything each time that the CPU starts or stops doing something.
63 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
64 the time, and our model holds.
66 ### What if the host has only one core?
68 In this case, the parameters \b OneCore and \b AllCores are obviously the same.
69 Actually, SimGrid expect an energetic profile formatted as 'Idle:Running' for mono-cores hosts.
70 If you insist on passing 3 parameters in this case, then you must have the same value for \b OneCore and \b AllCores.
73 <host id="HostC" power="100.0Mf" cores="1">
74 <prop id="watt_per_state" value="95.0:200.0" /> <!-- we may have used '95:200:200' instead -->
75 <prop id="watt_off" value="10" />
79 ### How does DVFS interact with the host energy model?
81 If your host has several DVFS levels (several pstates), then you should give the energetic profile of each pstate level:
84 <host id="HostC" power="100.0Mf,50.0Mf,20.0Mf" cores="4">
85 <prop id="watt_per_state" value="95.0:120.0:200.0, 93.0:115.0:170.0, 90.0:110.0:150.0" />
86 <prop id="watt_off" value="10" />
90 This encodes the following values
92 <tr><th>pstate</th><th>Performance</th><th>Idle</th><th>OneCore</th><th>AllCores</th></tr>
93 <tr><td>0</td><td>100 Mflop/s</td><td>95 Watts</td><td>120 Watts</td><td>200 Watts</td></tr>
94 <tr><td>1</td><td>50 Mflop/s</td><td>93 Watts</td><td>115 Watts</td><td>170 Watts</td></tr>
95 <tr><td>2</td><td>20 Mflop/s</td><td>90 Watts</td><td>110 Watts</td><td>150 Watts</td></tr>
98 To change the pstate of a given CPU, use the following functions:
99 #MSG_host_get_nb_pstates(), simgrid#s4u#Host#setPstate(), #MSG_host_get_power_peak_at().
101 ### How accurate are these models?
103 This model cannot be more accurate than your instantiation: with the default values, your result will not be accurate at
104 all. You can still get accurate energy prediction, provided that you carefully instantiate the model.
105 The first step is to ensure that your timing prediction match perfectly. But this is only the first step of the path,
106 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
107 before you can get accurate energy predictions.
110 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_energy, surf, "Logging specific to the SURF energy plugin");
121 PowerRange(double idle, double min, double max) : idle(idle), min(min), max(max) {}
126 static simgrid::xbt::Extension<simgrid::s4u::Host, HostEnergy> EXTENSION_ID;
128 explicit HostEnergy(simgrid::s4u::Host* ptr);
131 double getCurrentWattsValue(double cpu_load);
132 double getConsumedEnergy();
133 double getWattMinAt(int pstate);
134 double getWattMaxAt(int pstate);
138 void initWattsRangeList();
139 simgrid::s4u::Host* host = nullptr;
140 std::vector<PowerRange>
141 power_range_watts_list; /*< List of (min_power,max_power) pairs corresponding to each cpu pstate */
143 /* We need to keep track of what pstate has been used, as we will sometimes be notified only *after* a pstate has been
144 * used (but we need to update the energy consumption with the old pstate!)
147 const int pstate_off = -1;
150 double watts_off = 0.0; /*< Consumption when the machine is turned off (shutdown) */
151 double total_energy = 0.0; /*< Total energy consumed by the host */
152 double last_updated; /*< Timestamp of the last energy update event*/
155 simgrid::xbt::Extension<simgrid::s4u::Host, HostEnergy> HostEnergy::EXTENSION_ID;
157 /* Computes the consumption so far. Called lazily on need. */
158 void HostEnergy::update()
160 double start_time = this->last_updated;
161 double finish_time = surf_get_clock();
162 double current_speed = host->getSpeed();
164 if (start_time < finish_time) {
166 // We may have start == finish if the past consumption was updated since the simcall was started
167 // for example if 2 actors requested to update the same host's consumption in a given scheduling round.
169 // Even in this case, we need to save the pstate for the next call (after this big if),
170 // which may have changed since that recent update.
172 if (current_speed <= 0)
173 // Some users declare a pstate of speed 0 flops (e.g., to model boot time).
174 // We consider that the machine is then fully loaded. That's arbitrary but it avoids a NaN
177 cpu_load = lmm_constraint_get_usage(host->pimpl_cpu->constraint()) / current_speed;
179 /** Divide by the number of cores here **/
180 cpu_load /= host->pimpl_cpu->coreCount();
182 if (cpu_load > 1) // A machine with a load > 1 consumes as much as a fully loaded machine, not more
185 /* The problem with this model is that the load is always 0 or 1, never something less.
186 * Another possibility could be to model the total energy as
188 * X/(X+Y)*W_idle + Y/(X+Y)*W_burn
190 * where X is the amount of idling cores, and Y the amount of computing cores.
193 double previous_energy = this->total_energy;
195 double instantaneous_consumption;
196 if (this->pstate == pstate_off) // The host was off at the beginning of this time interval
197 instantaneous_consumption = this->watts_off;
199 instantaneous_consumption = this->getCurrentWattsValue(cpu_load);
201 double energy_this_step = instantaneous_consumption * (finish_time - start_time);
203 // TODO Trace: Trace energy_this_step from start_time to finish_time in host->name()
205 this->total_energy = previous_energy + energy_this_step;
206 this->last_updated = finish_time;
208 XBT_DEBUG("[update_energy of %s] period=[%.2f-%.2f]; current power peak=%.0E flop/s; consumption change: %.2f J -> "
210 host->getCname(), start_time, finish_time, host->pimpl_cpu->speed_.peak, previous_energy,
214 /* Save data for the upcoming time interval: whether it's on/off and the pstate if it's on */
215 this->pstate = host->isOn() ? host->getPstate() : pstate_off;
218 HostEnergy::HostEnergy(simgrid::s4u::Host* ptr) : host(ptr), last_updated(surf_get_clock())
220 initWattsRangeList();
222 const char* off_power_str = host->getProperty("watt_off");
223 if (off_power_str != nullptr) {
225 this->watts_off = std::stod(std::string(off_power_str));
226 } catch (std::invalid_argument& ia) {
227 throw std::invalid_argument(std::string("Invalid value for property watt_off of host ") + host->getCname() +
228 ": " + off_power_str);
231 /* watts_off is 0 by default */
234 HostEnergy::~HostEnergy() = default;
236 double HostEnergy::getWattMinAt(int pstate)
238 xbt_assert(not power_range_watts_list.empty(), "No power range properties specified for host %s", host->getCname());
239 return power_range_watts_list[pstate].min;
242 double HostEnergy::getWattMaxAt(int pstate)
244 xbt_assert(not power_range_watts_list.empty(), "No power range properties specified for host %s", host->getCname());
245 return power_range_watts_list[pstate].max;
248 /** @brief Computes the power consumed by the host according to the current pstate and processor load */
249 double HostEnergy::getCurrentWattsValue(double cpu_load)
251 xbt_assert(not power_range_watts_list.empty(), "No power range properties specified for host %s", host->getCname());
253 /* min_power corresponds to the power consumed when only one core is active */
254 /* max_power is the power consumed at 100% cpu load */
255 auto range = power_range_watts_list.at(this->pstate);
256 double current_power = 0;
257 double min_power = 0;
258 double max_power = 0;
259 double power_slope = 0;
261 if (cpu_load > 0) { /* Something is going on, the machine is not idle */
262 double min_power = range.min;
263 double max_power = range.max;
266 * The min_power states how much we consume when only one single
267 * core is working. This means that when cpu_load == 1/coreCount, then
268 * current_power == min_power.
270 * The maximum must be reached when all cores are working (but 1 core was
271 * already accounted for by min_power)
272 * i.e., we need min_power + (maxCpuLoad-1/coreCount)*power_slope == max_power
273 * (maxCpuLoad is by definition 1)
276 int coreCount = host->getCoreCount();
277 double coreReciprocal = static_cast<double>(1) / static_cast<double>(coreCount);
279 power_slope = (max_power - min_power) / (1 - coreReciprocal);
281 power_slope = 0; // Should be 0, since max_power == min_power (in this case)
283 current_power = min_power + (cpu_load - coreReciprocal) * power_slope;
284 } else { /* Our machine is idle, take the dedicated value! */
285 current_power = range.idle;
288 XBT_DEBUG("[get_current_watts] min_power=%f, max_power=%f, slope=%f", min_power, max_power, power_slope);
289 XBT_DEBUG("[get_current_watts] Current power (watts) = %f, load = %f", current_power, cpu_load);
291 return current_power;
294 double HostEnergy::getConsumedEnergy()
296 if (last_updated < surf_get_clock()) // We need to simcall this as it modifies the environment
297 simgrid::simix::kernelImmediate(std::bind(&HostEnergy::update, this));
302 void HostEnergy::initWattsRangeList()
304 const char* all_power_values_str = host->getProperty("watt_per_state");
305 if (all_power_values_str == nullptr)
308 std::vector<std::string> all_power_values;
309 boost::split(all_power_values, all_power_values_str, boost::is_any_of(","));
310 XBT_DEBUG("%s: profile: %s, cores: %d", host->getCname(), all_power_values_str, host->getCoreCount());
313 for (auto const& current_power_values_str : all_power_values) {
314 /* retrieve the power values associated with the current pstate */
315 std::vector<std::string> current_power_values;
316 boost::split(current_power_values, current_power_values_str, boost::is_any_of(":"));
317 if (host->getCoreCount() == 1) {
318 xbt_assert(current_power_values.size() == 2 || current_power_values.size() == 3,
319 "Power properties incorrectly defined for host %s."
320 "It should be 'Idle:FullSpeed' power values because you have one core only.",
322 if (current_power_values.size() == 2) {
323 // In this case, 1core == AllCores
324 current_power_values.push_back(current_power_values.at(1));
325 } else { // size == 3
326 xbt_assert((current_power_values.at(1)) == (current_power_values.at(2)),
327 "Power properties incorrectly defined for host %s.\n"
328 "The energy profile of mono-cores should be formatted as 'Idle:FullSpeed' only.\n"
329 "If you go for a 'Idle:OneCore:AllCores' power profile on mono-cores, then OneCore and AllCores "
334 xbt_assert(current_power_values.size() == 3,
335 "Power properties incorrectly defined for host %s."
336 "It should be 'Idle:OneCore:AllCores' power values because you have more than one core.",
340 /* min_power corresponds to the idle power (cpu load = 0) */
341 /* max_power is the power consumed at 100% cpu load */
342 char* msg_idle = bprintf("Invalid idle value for pstate %d on host %s: %%s", i, host->getCname());
343 char* msg_min = bprintf("Invalid OneCore value for pstate %d on host %s: %%s", i, host->getCname());
344 char* msg_max = bprintf("Invalid AllCores value for pstate %d on host %s: %%s", i, host->getCname());
345 PowerRange range(xbt_str_parse_double((current_power_values.at(0)).c_str(), msg_idle),
346 xbt_str_parse_double((current_power_values.at(1)).c_str(), msg_min),
347 xbt_str_parse_double((current_power_values.at(2)).c_str(), msg_max));
348 power_range_watts_list.push_back(range);
358 using simgrid::energy::HostEnergy;
360 /* **************************** events callback *************************** */
361 static void onCreation(simgrid::s4u::Host& host)
363 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
366 //TODO Trace: set to zero the energy variable associated to host->name()
368 host.extension_set(new HostEnergy(&host));
371 static void onActionStateChange(simgrid::surf::CpuAction* action, simgrid::surf::Action::State previous)
373 for (simgrid::surf::Cpu* const& cpu : action->cpus()) {
374 simgrid::s4u::Host* host = cpu->getHost();
375 if (host != nullptr) {
377 // If it's a VM, take the corresponding PM
378 simgrid::s4u::VirtualMachine* vm = dynamic_cast<simgrid::s4u::VirtualMachine*>(host);
379 if (vm) // If it's a VM, take the corresponding PM
380 host = vm->pimpl_vm_->getPm();
382 // Get the host_energy extension for the relevant host
383 HostEnergy* host_energy = host->extension<HostEnergy>();
385 if (host_energy->last_updated < surf_get_clock())
386 host_energy->update();
391 /* This callback is fired either when the host changes its state (on/off) ("onStateChange") or its speed
392 * (because the user changed the pstate, or because of external trace events) ("onSpeedChange") */
393 static void onHostChange(simgrid::s4u::Host& host)
395 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
398 HostEnergy* host_energy = host.extension<HostEnergy>();
400 host_energy->update();
403 static void onHostDestruction(simgrid::s4u::Host& host)
405 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
408 HostEnergy* host_energy = host.extension<HostEnergy>();
409 host_energy->update();
410 XBT_INFO("Energy consumption of host %s: %f Joules", host.getCname(), host_energy->getConsumedEnergy());
413 static void onSimulationEnd()
415 sg_host_t* host_list = sg_host_list();
416 int host_count = sg_host_count();
417 double total_energy = 0.0; // Total energy consumption (whole platform)
418 double used_hosts_energy = 0.0; // Energy consumed by hosts that computed something
419 for (int i = 0; i < host_count; i++) {
420 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(host_list[i]) == nullptr) { // Ignore virtual machines
422 bool host_was_used = (host_list[i]->extension<HostEnergy>()->last_updated != 0);
423 double energy = host_list[i]->extension<HostEnergy>()->getConsumedEnergy();
424 total_energy += energy;
426 used_hosts_energy += energy;
429 XBT_INFO("Total energy consumption: %f Joules (used hosts: %f Joules; unused/idle hosts: %f)",
430 total_energy, used_hosts_energy, total_energy - used_hosts_energy);
434 /* **************************** Public interface *************************** */
437 /** \ingroup plugin_energy
438 * \brief Enable host energy plugin
439 * \details Enable energy plugin to get joules consumption of each cpu. Call this function before #MSG_init().
441 void sg_host_energy_plugin_init()
443 if (HostEnergy::EXTENSION_ID.valid())
446 HostEnergy::EXTENSION_ID = simgrid::s4u::Host::extension_create<HostEnergy>();
448 simgrid::s4u::Host::onCreation.connect(&onCreation);
449 simgrid::s4u::Host::onStateChange.connect(&onHostChange);
450 simgrid::s4u::Host::onSpeedChange.connect(&onHostChange);
451 simgrid::s4u::Host::onDestruction.connect(&onHostDestruction);
452 simgrid::s4u::onSimulationEnd.connect(&onSimulationEnd);
453 simgrid::surf::CpuAction::onStateChange.connect(&onActionStateChange);
456 /** @ingroup plugin_energy
457 * @brief updates the consumption of all hosts
459 * After this call, sg_host_get_consumed_energy() will not interrupt your process
460 * (until after the next clock update).
462 void sg_host_energy_update_all()
464 simgrid::simix::kernelImmediate([]() {
465 std::vector<simgrid::s4u::Host*> list;
466 simgrid::s4u::Engine::getInstance()->getHostList(&list);
467 for (auto const& host : list)
468 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(host) == nullptr) // Ignore virtual machines
469 host->extension<HostEnergy>()->update();
473 /** @ingroup plugin_energy
474 * @brief Returns the total energy consumed by the host so far (in Joules)
476 * Please note that since the consumption is lazily updated, it may require a simcall to update it.
477 * The result is that the actor requesting this value will be interrupted,
478 * the value will be updated in kernel mode before returning the control to the requesting actor.
480 double sg_host_get_consumed_energy(sg_host_t host)
482 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
483 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
484 return host->extension<HostEnergy>()->getConsumedEnergy();
487 /** @ingroup plugin_energy
488 * @brief Get the amount of watt dissipated at the given pstate when the host is idling
490 double sg_host_get_wattmin_at(sg_host_t host, int pstate)
492 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
493 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
494 return host->extension<HostEnergy>()->getWattMinAt(pstate);
496 /** @ingroup plugin_energy
497 * @brief Returns the amount of watt dissipated at the given pstate when the host burns CPU at 100%
499 double sg_host_get_wattmax_at(sg_host_t host, int pstate)
501 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
502 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
503 return host->extension<HostEnergy>()->getWattMaxAt(pstate);
506 /** @ingroup plugin_energy
507 * @brief Returns the current consumption of the host
509 double sg_host_get_current_consumption(sg_host_t host)
511 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
512 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
513 double cpu_load = lmm_constraint_get_usage(host->pimpl_cpu->constraint()) / host->getSpeed();
514 return host->extension<HostEnergy>()->getCurrentWattsValue(cpu_load);