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
22 This is the energy plugin, enabling to account not only for computation time,
23 but also for the dissipated energy in the simulated platform.
24 To activate this plugin, first call sg_host_energy_plugin_init() before your #MSG_init(),
25 and then use 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
28 current CPU load and the host energy profile. According to our measurements,
29 the consumption is somehow linear in the amount of cores at full speed,
30 with an abnormality when all the cores are idle. The full details are in
31 <a href="https://hal.inria.fr/hal-01523608">our scientific paper</a> on that topic.
33 As a result, our energy model takes 4 parameters:
35 - \b Idle: instantaneous consumption (in Watt) when your host is up and running, but without anything to do.
36 - \b OneCore: instantaneous consumption (in Watt) when only one core is active, at 100%.
37 - \b AllCores: instantaneous consumption (in Watt) when all cores of the host are at 100%.
38 - \b Off: instantaneous consumption (in Watt) when the host is turned off.
40 Here is an example of XML declaration:
43 <host id="HostA" power="100.0Mf" cores="4">
44 <prop id="watt_per_state" value="100.0:120.0:200.0" />
45 <prop id="watt_off" value="10" />
49 This example gives the following parameters: \b Off is 10 Watts; \b Idle is 100 Watts; \b OneCore is 120 Watts and \b
50 AllCores is 200 Watts.
51 This is enough to compute the consumption as a function of the amount of loaded cores:
54 <tr><th>#Cores loaded</th><th>Consumption</th><th>Explanation</th></tr>
55 <tr><td>0</td><td> 100 Watts</td><td>Idle value</td></tr>
56 <tr><td>1</td><td> 120 Watts</td><td>OneCore value</td></tr>
57 <tr><td>2</td><td> 147 Watts</td><td>linear extrapolation between OneCore and AllCores</td></tr>
58 <tr><td>3</td><td> 173 Watts</td><td>linear extrapolation between OneCore and AllCores</td></tr>
59 <tr><td>4</td><td> 200 Watts</td><td>AllCores value</td></tr>
62 ### What if a given core is only at load 50%?
64 This is impossible in SimGrid because we recompute everything each time
65 that the CPU starts or stops doing something. So if a core is at load 50% over
66 a period, it means that it is at load 100% half of the time and at load 0% the
67 rest of the time, and our model holds.
69 ### What if the host has only one core?
71 In this case, the parameters \b OneCore and \b AllCores are obviously the same.
72 Actually, SimGrid expect an energetic profile formated as 'Idle:Running' for mono-cores hosts.
73 If you insist on passing 3 parameters in this case, then you must have the same value for \b OneCore and \b AllCores.
76 <host id="HostC" power="100.0Mf" cores="1">
77 <prop id="watt_per_state" value="95.0:200.0" /> <!-- we may have used '95:200:200' instead -->
78 <prop id="watt_off" value="10" />
82 ### How does DVFS interact with the host energy model?
84 If your host has several DVFS levels (several pstates), then you should
85 give the energetic profile of each pstate level:
88 <host id="HostC" power="100.0Mf,50.0Mf,20.0Mf" cores="4">
89 <prop id="watt_per_state" value="95.0:120.0:200.0, 93.0:115.0:170.0, 90.0:110.0:150.0" />
90 <prop id="watt_off" value="10" />
94 This encodes the following values
96 <tr><th>pstate</th><th>Performance</th><th>Idle</th><th>OneCore</th><th>AllCores</th></tr>
97 <tr><td>0</td><td>100 Mflop/s</td><td>95 Watts</td><td>120 Watts</td><td>200 Watts</td></tr>
98 <tr><td>1</td><td>50 Mflop/s</td><td>93 Watts</td><td>115 Watts</td><td>170 Watts</td></tr>
99 <tr><td>2</td><td>20 Mflop/s</td><td>90 Watts</td><td>110 Watts</td><td>150 Watts</td></tr>
102 To change the pstate of a given CPU, use the following functions:
103 #MSG_host_get_nb_pstates(), simgrid#s4u#Host#setPstate(), #MSG_host_get_power_peak_at().
105 ### How accurate are these models?
107 This model cannot be more accurate than your instantiation:
108 with the default values, your result will not be accurate at all. You can still get
109 accurate energy prediction, provided that you carefully instantiate the model.
110 The first step is to ensure that your timing prediction match perfectly. But this
111 is only the first step of the path, and you really want to read
112 <a href="https://hal.inria.fr/hal-01523608">this paper</a> to see all what you need
113 to do before you can get accurate energy predictions.
117 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_energy, surf, "Logging specific to the SURF energy plugin");
128 PowerRange(double idle, double min, double max) : idle(idle), min(min), max(max) {}
133 static simgrid::xbt::Extension<simgrid::s4u::Host, HostEnergy> EXTENSION_ID;
135 explicit HostEnergy(simgrid::s4u::Host* ptr);
138 double getCurrentWattsValue(double cpu_load);
139 double getConsumedEnergy();
140 double getWattMinAt(int pstate);
141 double getWattMaxAt(int pstate);
145 void initWattsRangeList();
146 simgrid::s4u::Host* host = nullptr;
147 std::vector<PowerRange>
148 power_range_watts_list; /*< List of (min_power,max_power) pairs corresponding to each cpu pstate */
150 /* We need to keep track of what pstate has been used, as we will sometimes
151 * be notified only *after* a pstate has been used (but we need to update the energy consumption
152 * with the old pstate!)
157 double watts_off = 0.0; /*< Consumption when the machine is turned off (shutdown) */
158 double total_energy = 0.0; /*< Total energy consumed by the host */
159 double last_updated; /*< Timestamp of the last energy update event*/
162 simgrid::xbt::Extension<simgrid::s4u::Host, HostEnergy> HostEnergy::EXTENSION_ID;
164 /* Computes the consumption so far. Called lazily on need. */
165 void HostEnergy::update()
167 double start_time = this->last_updated;
168 double finish_time = surf_get_clock();
169 double current_speed = host->getSpeed();
171 if (start_time < finish_time) {
173 // We may have start == finish if the past consumption was updated since the simcall was started
174 // for example if 2 actors requested to update the same host's consumption in a given scheduling round.
176 // Even in this case, we need to save the pstate for the next call (after this big if),
177 // which may have changed since that recent update.
179 if (current_speed <= 0)
180 // Some users declare a pstate of speed 0 flops (e.g., to model boot time).
181 // We consider that the machine is then fully loaded. That's arbitrary but it avoids a NaN
184 cpu_load = lmm_constraint_get_usage(host->pimpl_cpu->constraint()) / current_speed;
186 /** Divide by the number of cores here **/
187 cpu_load /= host->pimpl_cpu->coreCount();
189 if (cpu_load > 1) // A machine with a load > 1 consumes as much as a fully loaded machine, not more
192 /* The problem with this model is that the load is always 0 or 1, never something less.
193 * Another possibility could be to model the total energy as
195 * X/(X+Y)*W_idle + Y/(X+Y)*W_burn
197 * where X is the amount of idling cores, and Y the amount of computing cores.
200 double previous_energy = this->total_energy;
202 double instantaneous_consumption;
203 if (this->pstate == -1) // The host was off at the beginning of this time interval
204 instantaneous_consumption = this->watts_off;
206 instantaneous_consumption = this->getCurrentWattsValue(cpu_load);
208 double energy_this_step = instantaneous_consumption * (finish_time - start_time);
210 // TODO Trace: Trace energy_this_step from start_time to finish_time in host->name()
212 this->total_energy = previous_energy + energy_this_step;
213 this->last_updated = finish_time;
215 XBT_DEBUG("[update_energy of %s] period=[%.2f-%.2f]; current power peak=%.0E flop/s; consumption change: %.2f J -> "
217 host->getCname(), start_time, finish_time, host->pimpl_cpu->speed_.peak, previous_energy,
221 /* Save data for the upcoming time interval: whether it's on/off and the pstate if it's on */
222 this->pstate = host->isOn() ? host->getPstate() : -1;
225 HostEnergy::HostEnergy(simgrid::s4u::Host* ptr) : host(ptr), last_updated(surf_get_clock())
227 initWattsRangeList();
229 const char* off_power_str = host->getProperty("watt_off");
230 if (off_power_str != nullptr) {
231 char* msg = bprintf("Invalid value for property watt_off of host %s: %%s", host->getCname());
232 this->watts_off = xbt_str_parse_double(off_power_str, msg);
235 /* watts_off is 0 by default */
238 HostEnergy::~HostEnergy() = default;
240 double HostEnergy::getWattMinAt(int pstate)
242 xbt_assert(not power_range_watts_list.empty(), "No power range properties specified for host %s", host->getCname());
243 return power_range_watts_list[pstate].min;
246 double HostEnergy::getWattMaxAt(int pstate)
248 xbt_assert(not power_range_watts_list.empty(), "No power range properties specified for host %s", host->getCname());
249 return power_range_watts_list[pstate].max;
252 /** @brief Computes the power consumed by the host according to the current pstate and processor load */
253 double HostEnergy::getCurrentWattsValue(double cpu_load)
255 xbt_assert(not power_range_watts_list.empty(), "No power range properties specified for host %s", host->getCname());
257 /* min_power corresponds to the power consumed when only one core is active */
258 /* max_power is the power consumed at 100% cpu load */
259 auto range = power_range_watts_list.at(this->pstate);
260 double current_power = 0;
261 double min_power = 0;
262 double max_power = 0;
263 double power_slope = 0;
265 if (cpu_load > 0) { /* Something is going on, the machine is not idle */
266 double min_power = range.min;
267 double max_power = range.max;
270 * The min_power states how much we consume when only one single
271 * core is working. This means that when cpu_load == 1/coreCount, then
272 * current_power == min_power.
274 * The maximum must be reached when all cores are working (but 1 core was
275 * already accounted for by min_power)
276 * i.e., we need min_power + (maxCpuLoad-1/coreCount)*power_slope == max_power
277 * (maxCpuLoad is by definition 1)
280 int coreCount = host->getCoreCount();
281 double coreReciprocal = static_cast<double>(1) / static_cast<double>(coreCount);
283 power_slope = (max_power - min_power) / (1 - coreReciprocal);
285 power_slope = 0; // Should be 0, since max_power == min_power (in this case)
287 current_power = min_power + (cpu_load - coreReciprocal) * power_slope;
288 } else { /* Our machine is idle, take the dedicated value! */
289 current_power = range.idle;
292 XBT_DEBUG("[get_current_watts] min_power=%f, max_power=%f, slope=%f", min_power, max_power, power_slope);
293 XBT_DEBUG("[get_current_watts] Current power (watts) = %f, load = %f", current_power, cpu_load);
295 return current_power;
298 double HostEnergy::getConsumedEnergy()
300 if (last_updated < surf_get_clock()) // We need to simcall this as it modifies the environment
301 simgrid::simix::kernelImmediate(std::bind(&HostEnergy::update, this));
306 void HostEnergy::initWattsRangeList()
308 const char* all_power_values_str = host->getProperty("watt_per_state");
309 if (all_power_values_str == nullptr)
312 std::vector<std::string> all_power_values;
313 boost::split(all_power_values, all_power_values_str, boost::is_any_of(","));
314 XBT_DEBUG("%s: profile: %s, cores: %d", host->getCname(), all_power_values_str, host->getCoreCount());
317 for (auto current_power_values_str : all_power_values) {
318 /* retrieve the power values associated with the current pstate */
319 std::vector<std::string> current_power_values;
320 boost::split(current_power_values, current_power_values_str, boost::is_any_of(":"));
321 if (host->getCoreCount() == 1) {
322 xbt_assert(current_power_values.size() == 2 || current_power_values.size() == 3,
323 "Power properties incorrectly defined for host %s."
324 "It should be 'Idle:FullSpeed' power values because you have one core only.",
326 if (current_power_values.size() == 2) {
327 // In this case, 1core == AllCores
328 current_power_values.push_back(current_power_values.at(1));
329 } else { // size == 3
330 xbt_assert((current_power_values.at(1)) == (current_power_values.at(2)),
331 "Power properties incorrectly defined for host %s.\n"
332 "The energy profile of mono-cores should be formated as 'Idle:FullSpeed' only.\n"
333 "If you go for a 'Idle:OneCore:AllCores' power profile on mono-cores, then OneCore and AllCores "
338 xbt_assert(current_power_values.size() == 3,
339 "Power properties incorrectly defined for host %s."
340 "It should be 'Idle:OneCore:AllCores' power values because you have more than one core.",
344 /* min_power corresponds to the idle power (cpu load = 0) */
345 /* max_power is the power consumed at 100% cpu load */
346 char* msg_idle = bprintf("Invalid idle value for pstate %d on host %s: %%s", i, host->getCname());
347 char* msg_min = bprintf("Invalid OneCore value for pstate %d on host %s: %%s", i, host->getCname());
348 char* msg_max = bprintf("Invalid AllCores value for pstate %d on host %s: %%s", i, host->getCname());
349 PowerRange range(xbt_str_parse_double((current_power_values.at(0)).c_str(), msg_idle),
350 xbt_str_parse_double((current_power_values.at(1)).c_str(), msg_min),
351 xbt_str_parse_double((current_power_values.at(2)).c_str(), msg_max));
352 power_range_watts_list.push_back(range);
362 using simgrid::energy::HostEnergy;
364 /* **************************** events callback *************************** */
365 static void onCreation(simgrid::s4u::Host& host)
367 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
370 //TODO Trace: set to zero the energy variable associated to host->name()
372 host.extension_set(new HostEnergy(&host));
375 static void onActionStateChange(simgrid::surf::CpuAction* action, simgrid::surf::Action::State previous)
377 for (simgrid::surf::Cpu* cpu : action->cpus()) {
378 simgrid::s4u::Host* host = cpu->getHost();
379 if (host != nullptr) {
381 // If it's a VM, take the corresponding PM
382 simgrid::s4u::VirtualMachine* vm = dynamic_cast<simgrid::s4u::VirtualMachine*>(host);
383 if (vm) // If it's a VM, take the corresponding PM
384 host = vm->pimpl_vm_->getPm();
386 // Get the host_energy extension for the relevant host
387 HostEnergy* host_energy = host->extension<HostEnergy>();
389 if (host_energy->last_updated < surf_get_clock())
390 host_energy->update();
395 /* This callback is fired either when the host changes its state (on/off) ("onStateChange") or its speed
396 * (because the user changed the pstate, or because of external trace events) ("onSpeedChange") */
397 static void onHostChange(simgrid::s4u::Host& host)
399 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
402 HostEnergy* host_energy = host.extension<HostEnergy>();
404 host_energy->update();
407 static void onHostDestruction(simgrid::s4u::Host& host)
409 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(&host)) // Ignore virtual machines
412 HostEnergy* host_energy = host.extension<HostEnergy>();
413 host_energy->update();
414 XBT_INFO("Energy consumption of host %s: %f Joules", host.getCname(), host_energy->getConsumedEnergy());
417 static void onSimulationEnd()
419 sg_host_t* host_list = sg_host_list();
420 int host_count = sg_host_count();
421 double total_energy = 0.0; // Total energy consumption (whole platform)
422 double used_hosts_energy = 0.0; // Energy consumed by hosts that computed something
423 for (int i = 0; i < host_count; i++) {
424 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(host_list[i]) == nullptr) { // Ignore virtual machines
426 bool host_was_used = (host_list[i]->extension<HostEnergy>()->last_updated != 0);
427 double energy = host_list[i]->extension<HostEnergy>()->getConsumedEnergy();
428 total_energy += energy;
430 used_hosts_energy += energy;
433 XBT_INFO("Total energy consumption: %f Joules (used hosts: %f Joules; unused/idle hosts: %f)",
434 total_energy, used_hosts_energy, total_energy - used_hosts_energy);
438 /* **************************** Public interface *************************** */
441 /** \ingroup plugin_energy
442 * \brief Enable host energy plugin
443 * \details Enable energy plugin to get joules consumption of each cpu. Call this function before #MSG_init().
445 void sg_host_energy_plugin_init()
447 if (HostEnergy::EXTENSION_ID.valid())
450 HostEnergy::EXTENSION_ID = simgrid::s4u::Host::extension_create<HostEnergy>();
452 simgrid::s4u::Host::onCreation.connect(&onCreation);
453 simgrid::s4u::Host::onStateChange.connect(&onHostChange);
454 simgrid::s4u::Host::onSpeedChange.connect(&onHostChange);
455 simgrid::s4u::Host::onDestruction.connect(&onHostDestruction);
456 simgrid::s4u::onSimulationEnd.connect(&onSimulationEnd);
457 simgrid::surf::CpuAction::onStateChange.connect(&onActionStateChange);
460 /** @ingroup plugin_energy
461 * @brief updates the consumption of all hosts
463 * After this call, sg_host_get_consumed_energy() will not interrupt your process
464 * (until after the next clock update).
466 void sg_host_energy_update_all()
468 simgrid::simix::kernelImmediate([]() {
469 std::vector<simgrid::s4u::Host*> list;
470 simgrid::s4u::Engine::getInstance()->getHostList(&list);
471 for (auto host : list)
472 if (dynamic_cast<simgrid::s4u::VirtualMachine*>(host) == nullptr) // Ignore virtual machines
473 host->extension<HostEnergy>()->update();
477 /** @ingroup plugin_energy
478 * @brief Returns the total energy consumed by the host so far (in Joules)
480 * Please note that since the consumption is lazily updated, it may require a simcall to update it.
481 * The result is that the actor requesting this value will be interrupted,
482 * the value will be updated in kernel mode before returning the control to the requesting actor.
484 double sg_host_get_consumed_energy(sg_host_t host)
486 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
487 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
488 return host->extension<HostEnergy>()->getConsumedEnergy();
491 /** @ingroup plugin_energy
492 * @brief Get the amount of watt dissipated at the given pstate when the host is idling
494 double sg_host_get_wattmin_at(sg_host_t host, int pstate)
496 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
497 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
498 return host->extension<HostEnergy>()->getWattMinAt(pstate);
500 /** @ingroup plugin_energy
501 * @brief Returns the amount of watt dissipated at the given pstate when the host burns CPU at 100%
503 double sg_host_get_wattmax_at(sg_host_t host, int pstate)
505 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
506 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
507 return host->extension<HostEnergy>()->getWattMaxAt(pstate);
510 /** @ingroup plugin_energy
511 * @brief Returns the current consumption of the host
513 double sg_host_get_current_consumption(sg_host_t host)
515 xbt_assert(HostEnergy::EXTENSION_ID.valid(),
516 "The Energy plugin is not active. Please call sg_energy_plugin_init() during initialization.");
517 double cpu_load = lmm_constraint_get_usage(host->pimpl_cpu->constraint()) / host->getSpeed();
518 return host->extension<HostEnergy>()->getCurrentWattsValue(cpu_load);