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->getName()
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());
254 * * Return watts_off if pstate == pstate_off
255 * * this happens when host is off
257 if (this->pstate == pstate_off) {
261 /* min_power corresponds to the power consumed when only one core is active */
262 /* max_power is the power consumed at 100% cpu load */
263 auto range = power_range_watts_list.at(this->pstate);
264 double current_power = 0;
265 double min_power = 0;
266 double max_power = 0;
267 double power_slope = 0;
269 if (cpu_load > 0) { /* Something is going on, the machine is not idle */
270 double min_power = range.min;
271 double max_power = range.max;
274 * The min_power states how much we consume when only one single
275 * core is working. This means that when cpu_load == 1/coreCount, then
276 * current_power == min_power.
278 * The maximum must be reached when all cores are working (but 1 core was
279 * already accounted for by min_power)
280 * i.e., we need min_power + (maxCpuLoad-1/coreCount)*power_slope == max_power
281 * (maxCpuLoad is by definition 1)
284 int coreCount = host->getCoreCount();
285 double coreReciprocal = static_cast<double>(1) / static_cast<double>(coreCount);
287 power_slope = (max_power - min_power) / (1 - coreReciprocal);
289 power_slope = 0; // Should be 0, since max_power == min_power (in this case)
291 current_power = min_power + (cpu_load - coreReciprocal) * power_slope;
292 } else { /* Our machine is idle, take the dedicated value! */
293 current_power = range.idle;
296 XBT_DEBUG("[get_current_watts] min_power=%f, max_power=%f, slope=%f", min_power, max_power, power_slope);
297 XBT_DEBUG("[get_current_watts] Current power (watts) = %f, load = %f", current_power, cpu_load);
299 return current_power;
302 double HostEnergy::getConsumedEnergy()
304 if (last_updated < surf_get_clock()) // We need to simcall this as it modifies the environment
305 simgrid::simix::kernelImmediate(std::bind(&HostEnergy::update, this));
310 void HostEnergy::initWattsRangeList()
312 const char* all_power_values_str = host->getProperty("watt_per_state");
313 if (all_power_values_str == nullptr)
316 std::vector<std::string> all_power_values;
317 boost::split(all_power_values, all_power_values_str, boost::is_any_of(","));
318 XBT_DEBUG("%s: profile: %s, cores: %d", host->getCname(), all_power_values_str, host->getCoreCount());
321 for (auto const& current_power_values_str : all_power_values) {
322 /* retrieve the power values associated with the current pstate */
323 std::vector<std::string> current_power_values;
324 boost::split(current_power_values, current_power_values_str, boost::is_any_of(":"));
325 if (host->getCoreCount() == 1) {
326 xbt_assert(current_power_values.size() == 2 || current_power_values.size() == 3,
327 "Power properties incorrectly defined for host %s."
328 "It should be 'Idle:FullSpeed' power values because you have one core only.",
330 if (current_power_values.size() == 2) {
331 // In this case, 1core == AllCores
332 current_power_values.push_back(current_power_values.at(1));
333 } else { // size == 3
334 xbt_assert((current_power_values.at(1)) == (current_power_values.at(2)),
335 "Power properties incorrectly defined for host %s.\n"
336 "The energy profile of mono-cores should be formatted as 'Idle:FullSpeed' only.\n"
337 "If you go for a 'Idle:OneCore:AllCores' power profile on mono-cores, then OneCore and AllCores "
342 xbt_assert(current_power_values.size() == 3,
343 "Power properties incorrectly defined for host %s."
344 "It should be 'Idle:OneCore:AllCores' power values because you have more than one core.",
348 /* min_power corresponds to the idle power (cpu load = 0) */
349 /* max_power is the power consumed at 100% cpu load */
350 char* msg_idle = bprintf("Invalid idle value for pstate %d on host %s: %%s", i, host->getCname());
351 char* msg_min = bprintf("Invalid OneCore value for pstate %d on host %s: %%s", i, host->getCname());
352 char* msg_max = bprintf("Invalid AllCores value for pstate %d on host %s: %%s", i, host->getCname());
353 PowerRange range(xbt_str_parse_double((current_power_values.at(0)).c_str(), msg_idle),
354 xbt_str_parse_double((current_power_values.at(1)).c_str(), msg_min),
355 xbt_str_parse_double((current_power_values.at(2)).c_str(), msg_max));
356 power_range_watts_list.push_back(range);
366 using simgrid::plugin::HostEnergy;
368 /* **************************** events callback *************************** */
369 static void onCreation(simgrid::s4u::Host& host)
371 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
374 // TODO Trace: set to zero the energy variable associated to host->getName()
376 host.extension_set(new HostEnergy(&host));
379 static void onActionStateChange(simgrid::surf::CpuAction* action, simgrid::surf::Action::State previous)
381 for (simgrid::surf::Cpu* const& cpu : action->cpus()) {
382 simgrid::s4u::Host* host = cpu->getHost();
383 if (host != nullptr) {
385 // If it's a VM, take the corresponding PM
386 simgrid::s4u::VirtualMachine* vm = dynamic_cast<simgrid::s4u::VirtualMachine*>(host);
387 if (vm) // If it's a VM, take the corresponding PM
388 host = vm->pimpl_vm_->getPm();
390 // Get the host_energy extension for the relevant host
391 HostEnergy* host_energy = host->extension<HostEnergy>();
393 if (host_energy->last_updated < surf_get_clock())
394 host_energy->update();
399 /* This callback is fired either when the host changes its state (on/off) ("onStateChange") or its speed
400 * (because the user changed the pstate, or because of external trace events) ("onSpeedChange") */
401 static void onHostChange(simgrid::s4u::Host& host)
403 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
406 HostEnergy* host_energy = host.extension<HostEnergy>();
408 host_energy->update();
411 static void onHostDestruction(simgrid::s4u::Host& host)
413 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
416 HostEnergy* host_energy = host.extension<HostEnergy>();
417 host_energy->update();
418 XBT_INFO("Energy consumption of host %s: %f Joules", host.getCname(), host_energy->getConsumedEnergy());
421 static void onSimulationEnd()
423 sg_host_t* host_list = sg_host_list();
424 int host_count = sg_host_count();
425 double total_energy = 0.0; // Total energy consumption (whole platform)
426 double used_hosts_energy = 0.0; // Energy consumed by hosts that computed something
427 for (int i = 0; i < host_count; i++) {
428 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(host_list[i]) == nullptr) { // Ignore virtual machines
430 bool host_was_used = (host_list[i]->extension<HostEnergy>()->last_updated != 0);
431 double energy = host_list[i]->extension<HostEnergy>()->getConsumedEnergy();
432 total_energy += energy;
434 used_hosts_energy += energy;
437 XBT_INFO("Total energy consumption: %f Joules (used hosts: %f Joules; unused/idle hosts: %f)",
438 total_energy, used_hosts_energy, total_energy - used_hosts_energy);
442 /* **************************** Public interface *************************** */
445 /** \ingroup plugin_energy
446 * \brief Enable host energy plugin
447 * \details Enable energy plugin to get joules consumption of each cpu. Call this function before #MSG_init().
449 void sg_host_energy_plugin_init()
451 if (HostEnergy::EXTENSION_ID.valid())
454 HostEnergy::EXTENSION_ID = simgrid::s4u::Host::extension_create<HostEnergy>();
456 simgrid::s4u::Host::onCreation.connect(&onCreation);
457 simgrid::s4u::Host::onStateChange.connect(&onHostChange);
458 simgrid::s4u::Host::onSpeedChange.connect(&onHostChange);
459 simgrid::s4u::Host::onDestruction.connect(&onHostDestruction);
460 simgrid::s4u::onSimulationEnd.connect(&onSimulationEnd);
461 simgrid::surf::CpuAction::onStateChange.connect(&onActionStateChange);
464 /** @ingroup plugin_energy
465 * @brief updates the consumption of all hosts
467 * After this call, sg_host_get_consumed_energy() will not interrupt your process
468 * (until after the next clock update).
470 void sg_host_energy_update_all()
472 simgrid::simix::kernelImmediate([]() {
473 std::vector<simgrid::s4u::Host*> list;
474 simgrid::s4u::Engine::getInstance()->getHostList(&list);
475 for (auto const& host : list)
476 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(host) == nullptr) // Ignore virtual machines
477 host->extension<HostEnergy>()->update();
481 /** @ingroup plugin_energy
482 * @brief Returns the total energy consumed by the host so far (in Joules)
484 * Please note that since the consumption is lazily updated, it may require a simcall to update it.
485 * The result is that the actor requesting this value will be interrupted,
486 * the value will be updated in kernel mode before returning the control to the requesting actor.
488 double sg_host_get_consumed_energy(sg_host_t host)
490 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
491 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
492 return host->extension<HostEnergy>()->getConsumedEnergy();
495 /** @ingroup plugin_energy
496 * @brief Get the amount of watt dissipated at the given pstate when the host is idling
498 double sg_host_get_wattmin_at(sg_host_t host, int pstate)
500 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
501 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
502 return host->extension<HostEnergy>()->getWattMinAt(pstate);
504 /** @ingroup plugin_energy
505 * @brief Returns the amount of watt dissipated at the given pstate when the host burns CPU at 100%
507 double sg_host_get_wattmax_at(sg_host_t host, int pstate)
509 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
510 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
511 return host->extension<HostEnergy>()->getWattMaxAt(pstate);
514 /** @ingroup plugin_energy
515 * @brief Returns the current consumption of the host
517 double sg_host_get_current_consumption(sg_host_t host)
519 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
520 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
521 double cpu_load = lmm_constraint_get_usage(host->pimpl_cpu->constraint()) / host->getSpeed();
522 return host->extension<HostEnergy>()->getCurrentWattsValue(cpu_load);