New field in output file OUT_jobs.csv: the consumed_energy field computes, for...

New field in output file OUT_jobs.csv: the consumed_energy field computes, for each job, the consumed energy by the allocated machines during the job execution. Beware:this metric might be invalid if space-sharing is enabled!

export.cpp

@@ -512,7 +512,7 @@ void exportJobsToCSV(const string &filename, BatsimContext *context)
     xbt_assert(f.is_open(), "Cannot write file '%s'", filename.c_str());
 
     // write headers
-    f << "jobID,submission_time,requested_number_of_processors,requested_time,success,starting_time,execution_time,finish_time,waiting_time,turnaround_time,stretch,allocated_processors\n";
+    f << "jobID,submission_time,requested_number_of_processors,requested_time,success,starting_time,execution_time,finish_time,waiting_time,turnaround_time,stretch,consumed_energy,allocated_processors\n";
 
     const auto & jobs = context->jobs.jobs();
     for (const auto & mit : jobs)
@@ -523,7 +523,8 @@ void exportJobsToCSV(const string &filename, BatsimContext *context)
         {
             char * buf;
             int success = job->state == JobState::JOB_STATE_COMPLETED_SUCCESSFULLY;
-            int ret = asprintf(&buf, "%d,%lf,%d,%lf,%d,%lf,%lf,%lf,%lf,%lf,%lf,", job->id,
+            int ret = asprintf(&buf, "%d,%lf,%d,%lf,%d,%lf,%lf,%lf,%lf,%lf,%lf,%Lf,", // finished by a ',' because the next part is written after asprintf
+                     job->id,
                      job->submission_time,
                      job->required_nb_res,
                      job->walltime,
@@ -533,12 +534,15 @@ void exportJobsToCSV(const string &filename, BatsimContext *context)
                      job->starting_time + job->runtime, // finish_time
                      job->starting_time - job->submission_time, // waiting_time
                      job->starting_time + job->runtime - job->submission_time, // turnaround_time
-                     (job->starting_time + job->runtime - job->submission_time) / job->runtime // stretch
+                     (job->starting_time + job->runtime - job->submission_time) / job->runtime, // stretch
+                     job->consumed_energy
                      );
             xbt_assert(ret != -1, "asprintf failed (not enough memory?)");
             f << buf;
             free(buf);
 
+            // TODO: use boost::join for the following if
+
             if (job->required_nb_res > 0)
             {
                 int ret = asprintf(&buf, "%d", job->allocation[0]);

jobs.hpp

@@ -29,6 +29,8 @@ struct Job
     double walltime;
     int required_nb_res;
 
+    long double consumed_energy; //! The sum, for each machine on which the job has been allocated, of the consumed energy (in Joules) during the job execution time (consumed_energy_after_job_completion - consumed_energy_before_job_start)
+
     double starting_time;
     double runtime;
     std::vector<int> allocation;

jobs_execution.cpp

@@ -227,6 +227,7 @@ int execute_job_process(int argc, char *argv[])
     (void) argc;
     (void) argv;
 
+    // Retrieving input parameters
     ExecuteJobProcessArguments * args = (ExecuteJobProcessArguments *) MSG_process_get_data(MSG_process_self());
 
     Job * job = args->context->jobs[args->allocation.job_id];
@@ -234,6 +235,19 @@ int execute_job_process(int argc, char *argv[])
     job->allocation = args->allocation.machine_ids;
     double remaining_time = job->walltime;
 
+    // If energy is enabled, let us compute the energy used by the machines before running the job
+    if (args->context->energy_used)
+    {
+        job->consumed_energy = 0;
+
+        for(const int & machine_id : job->allocation)
+        {
+            Machine * machine = args->context->machines[machine_id];
+            job->consumed_energy += MSG_host_get_consumed_energy(machine->host);
+        }
+    }
+
+    // Job computation
     args->context->machines.updateMachinesOnJobRun(job->id, args->allocation.machine_ids);
     if (execute_profile(args->context, job->profile, &args->allocation, &remaining_time) == 1)
     {
@@ -250,6 +264,23 @@ int execute_job_process(int argc, char *argv[])
     args->context->machines.updateMachinesOnJobEnd(job->id, args->allocation.machine_ids);
     job->runtime = MSG_get_clock() - job->starting_time;
 
+    // If energy is enabled, let us compute the energy used by the machines after running the job
+    if (args->context->energy_used)
+    {
+        long double consumed_energy_before = job->consumed_energy;
+        job->consumed_energy = 0;
+
+        for(const int & machine_id : job->allocation)
+        {
+            Machine * machine = args->context->machines[machine_id];
+            job->consumed_energy += MSG_host_get_consumed_energy(machine->host);
+        }
+
+        // The consumed energy is the difference (consumed_energy_after_job - consumed_energy_before_job)
+        job->consumed_energy = job->consumed_energy - consumed_energy_before;
+    }
+
+    // Let us tell the server that the job completed
     JobCompletedMessage * message = new JobCompletedMessage;
     message->job_id = job->id;