Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
[HostLoad] And lastly: Update the hooks used
[simgrid.git] / src / plugins / host_load.cpp
index 17ae65a..4a549d8 100644 (file)
@@ -4,7 +4,12 @@
  * under the terms of the license (GNU LGPL) which comes with this package. */
 
 #include "simgrid/plugins/load.h"
+#include "src/include/surf/surf.hpp"
+#include "src/kernel/activity/ExecImpl.hpp"
 #include "src/plugins/vm/VirtualMachineImpl.hpp"
+#include <simgrid/s4u.hpp>
+
+SIMGRID_REGISTER_PLUGIN(host_load, "Cpu load", &sg_host_load_plugin_init)
 
 /** @addtogroup plugin_load
 
@@ -17,6 +22,8 @@ XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_plugin_load, surf, "Logging specific to the
 namespace simgrid {
 namespace plugin {
 
+static const double activity_uninitialized_remaining_cost = -1;
+
 class HostLoad {
 public:
   static simgrid::xbt::Extension<simgrid::s4u::Host, HostLoad> EXTENSION_ID;
@@ -25,27 +32,34 @@ public:
       : host_(ptr)
       , last_updated_(surf_get_clock())
       , last_reset_(surf_get_clock())
-      , current_speed_(host_->getSpeed())
+      , current_speed_(host_->get_speed())
       , current_flops_(host_->pimpl_cpu->get_constraint()->get_usage())
       , theor_max_flops_(0)
-      , was_prev_idle_(current_flops_ == 0)
   {
   }
   ~HostLoad() = default;
+  HostLoad() = delete;
+  explicit HostLoad(simgrid::s4u::Host& ptr) = delete;
+  explicit HostLoad(simgrid::s4u::Host&& ptr) = delete;
 
   double get_current_load();
-  double get_average_load() { return (theor_max_flops_ == 0) ? 0 : computed_flops_ / theor_max_flops_; };
-  double get_computed_flops() { return computed_flops_; }
-  double get_idle_time() { return idle_time_; } /** Return idle time since last reset */
+  double get_average_load() { update(); return (theor_max_flops_ == 0) ? 0 : computed_flops_ / theor_max_flops_; };
+  double get_computed_flops() { update(); return computed_flops_; }
+  double get_idle_time() { update(); return idle_time_; } /** Return idle time since last reset */
+  double get_total_idle_time() { update(); return total_idle_time_; } /** Return idle time over the whole simulation */
   void update();
+  void add_activity(simgrid::kernel::activity::ExecImplPtr activity);
   void reset();
 
 private:
   simgrid::s4u::Host* host_ = nullptr;
+  /* Stores all currently ongoing activities (computations) on this machine */
+  std::map<simgrid::kernel::activity::ExecImplPtr, /* cost still remaining*/double> current_activities;
   double last_updated_      = 0;
   double last_reset_        = 0;
   /**
-   * current_speed each core is running at right now
+   * current_speed each core is running at; we need to store this as the speed
+   * will already have changed once we get notified
    */
   double current_speed_     = 0;
   /**
@@ -55,31 +69,53 @@ private:
   double current_flops_     = 0;
   double computed_flops_    = 0;
   double idle_time_         = 0;
+  double total_idle_time_   = 0; /* This gets never reset */
   double theor_max_flops_   = 0;
-  bool was_prev_idle_       = true; /* A host is idle at the beginning */
 };
 
 simgrid::xbt::Extension<simgrid::s4u::Host, HostLoad> HostLoad::EXTENSION_ID;
 
+void HostLoad::add_activity(simgrid::kernel::activity::ExecImplPtr activity)
+{
+  current_activities.insert({activity, activity_uninitialized_remaining_cost});
+}
+
 void HostLoad::update()
 {
   double now = surf_get_clock();
 
-  /* Current flop per second computed by the cpu; current_flops = k * pstate_speed_in_flops, k \in {0, 1, ..., cores}
-   * number of active cores */
-  current_flops_ = host_->pimpl_cpu->get_constraint()->get_usage();
+  // This loop updates the flops that the host executed for the ongoing computations
+  for (auto& pair : current_activities) {
+    auto& activity                         = pair.first;  // Just an alias
+    auto& remaining_cost_after_last_update = pair.second; // Just an alias
+
+    if (activity->surf_action_->get_finish_time() != now && activity->state_ == e_smx_state_t::SIMIX_RUNNING) {
+      if (remaining_cost_after_last_update == activity_uninitialized_remaining_cost) {
+        remaining_cost_after_last_update = activity->surf_action_->get_cost();
+      }
+      double computed_flops_since_last_update = remaining_cost_after_last_update - /*remaining now*/activity->get_remaining();
+      computed_flops_                        += computed_flops_since_last_update;
+      remaining_cost_after_last_update        = activity->get_remaining();
+    }
+    else if (activity->state_ == e_smx_state_t::SIMIX_DONE) {
+      computed_flops_ += remaining_cost_after_last_update;
+      current_activities.erase(activity);
+    }
+  }
 
-  /* flops == pstate_speed * cores_being_currently_used */
-  computed_flops_ += (now - last_updated_) * current_flops_;
+  /* Current flop per second computed by the cpu; current_flops = k * pstate_speed_in_flops, k @in {0, 1, ..., cores-1}
+   * designates number of active cores; will be 0 if CPU is currently idle */
+  current_flops_ = host_->pimpl_cpu->get_constraint()->get_usage();
 
-  if (was_prev_idle_) {
+  if (current_flops_ == 0) {
     idle_time_ += (now - last_updated_);
+    total_idle_time_ += (now - last_updated_);
+    XBT_DEBUG("[%s]: Currently idle -> Added %f seconds to idle time (totaling %fs)", host_->get_cname(), (now - last_updated_), idle_time_);
   }
 
   theor_max_flops_ += current_speed_ * host_->get_core_count() * (now - last_updated_);
-  current_speed_ = host_->getSpeed();
+  current_speed_ = host_->get_speed();
   last_updated_  = now;
-  was_prev_idle_ = (current_flops_ == 0);
 }
 
 /**
@@ -94,7 +130,7 @@ double HostLoad::get_current_load()
 {
   // We don't need to call update() here because it is called every time an action terminates or starts
   // FIXME: Can this happen at the same time? stop -> call to getCurrentLoad, load = 0 -> next action starts?
-  return current_flops_ / static_cast<double>(host_->getSpeed() * host_->get_core_count());
+  return current_flops_ / static_cast<double>(host_->get_speed() * host_->get_core_count());
 }
 
 /*
@@ -108,8 +144,7 @@ void HostLoad::reset()
   computed_flops_  = 0;
   theor_max_flops_ = 0;
   current_flops_   = host_->pimpl_cpu->get_constraint()->get_usage();
-  current_speed_   = host_->getSpeed();
-  was_prev_idle_   = (current_flops_ == 0);
+  current_speed_   = host_->get_speed();
 }
 } // namespace plugin
 } // namespace simgrid
@@ -128,7 +163,7 @@ static void on_host_change(simgrid::s4u::Host& host)
 }
 
 /* This callback is called when an action (computation, idle, ...) terminates */
-static void on_action_state_change(simgrid::surf::CpuAction* action)
+static void on_action_state_change(simgrid::surf::CpuAction* action, simgrid::kernel::resource::Action::State /*previous*/)
 {
   for (simgrid::surf::Cpu* const& cpu : action->cpus()) {
     simgrid::s4u::Host* host = cpu->get_host();
@@ -144,9 +179,9 @@ static void on_action_state_change(simgrid::surf::CpuAction* action)
 
 /* **************************** Public interface *************************** */
 
-/** \ingroup plugin_load
- * \brief Initializes the HostLoad plugin
- * \details The HostLoad plugin provides an API to get the current load of each host.
+/** @ingroup plugin_load
+ * @brief Initializes the HostLoad plugin
+ * @details The HostLoad plugin provides an API to get the current load of each host.
  */
 void sg_host_load_plugin_init()
 {
@@ -155,6 +190,14 @@ void sg_host_load_plugin_init()
 
   HostLoad::EXTENSION_ID = simgrid::s4u::Host::extension_create<HostLoad>();
 
+  if (simgrid::s4u::Engine::is_initialized()) { // If not yet initialized, this would create a new instance
+                                                // which would cause seg faults...
+    simgrid::s4u::Engine* e = simgrid::s4u::Engine::get_instance();
+    for (auto& host : e->get_all_hosts()) {
+      host->extension_set(new HostLoad(host));
+    }
+  }
+
   /* When attaching a callback into a signal, you can use a lambda as follows, or a regular function as done below */
 
   simgrid::s4u::Host::on_creation.connect([](simgrid::s4u::Host& host) {
@@ -163,7 +206,33 @@ void sg_host_load_plugin_init()
     host.extension_set(new HostLoad(&host));
   });
 
-  simgrid::surf::CpuAction::on_state_change.connect(&on_action_state_change);
+  simgrid::kernel::activity::ExecImpl::on_creation.connect([](simgrid::kernel::activity::ExecImplPtr activity){
+    if (activity->host_ != nullptr) { // We only run on one host
+      simgrid::s4u::Host* host = activity->host_;
+      if (dynamic_cast<simgrid::s4u::VirtualMachine*>(activity->host_))
+        host = dynamic_cast<simgrid::s4u::VirtualMachine*>(activity->host_)->get_pm();
+
+      host->extension<HostLoad>()->add_activity(activity);
+      host->extension<HostLoad>()->update(); // If the system was idle until now, we need to update *before*
+                                             // this computation starts running so we can keep track of the
+                                             // idle time. (Communication operations don't trigger this hook!)
+    }
+    else { // This runs on multiple hosts
+      XBT_DEBUG("HostLoad plugin currently does not support executions on several hosts");
+    }
+  });
+  simgrid::kernel::activity::ExecImpl::on_completion.connect([](simgrid::kernel::activity::ExecImplPtr activity){
+    if (activity->host_ != nullptr) { // We only run on one host
+      simgrid::s4u::Host* host = activity->host_;
+      if (dynamic_cast<simgrid::s4u::VirtualMachine*>(activity->host_))
+        host = dynamic_cast<simgrid::s4u::VirtualMachine*>(activity->host_)->get_pm();
+
+      host->extension<HostLoad>()->update();
+    }
+    else { // This runs on multiple hosts
+      XBT_DEBUG("HostLoad plugin currently does not support executions on several hosts");
+    }
+  });
   simgrid::s4u::Host::on_state_change.connect(&on_host_change);
   simgrid::s4u::Host::on_speed_change.connect(&on_host_change);
 }
@@ -204,6 +273,14 @@ double sg_host_get_idle_time(sg_host_t host)
   return host->extension<HostLoad>()->get_idle_time();
 }
 
+double sg_host_get_total_idle_time(sg_host_t host)
+{
+  xbt_assert(HostLoad::EXTENSION_ID.valid(),
+             "The Load plugin is not active. Please call sg_host_load_plugin_init() during initialization.");
+
+  return host->extension<HostLoad>()->get_total_idle_time();
+}
+
 double sg_host_get_computed_flops(sg_host_t host)
 {
   xbt_assert(HostLoad::EXTENSION_ID.valid(),