#include <boost/algorithm/string.hpp>
-#include <simgrid/msg.h>
#include <simgrid/s4u.hpp>
#include <smpi/smpi.h>
#include <xbt/file.hpp>
};
// ugly globals to avoid creating structures for giving args to processes
-std::vector<simgrid::s4u::Host*> hosts;
-int noise_between_jobs;
+static std::vector<simgrid::s4u::Host*> hosts;
+static int noise_between_jobs;
static bool job_comparator(const Job* j1, const Job* j2)
{
return j1->starting_time < j2->starting_time;
}
-struct s_smpi_replay_process_args {
- Job* job;
- simgrid::s4u::BarrierPtr barrier;
- int rank;
-};
-
-static int smpi_replay_process(int argc, char* argv[])
+static void smpi_replay_process(Job* job, simgrid::s4u::BarrierPtr barrier, int rank)
{
- s_smpi_replay_process_args* args = static_cast<s_smpi_replay_process_args*>(MSG_process_get_data(MSG_process_self()));
-
- XBT_INFO("Replaying rank %d of job %d (smpi_app '%s')", args->rank, args->job->unique_job_number,
- args->job->smpi_app_name.c_str());
-
+ // Prepare data for smpi_replay_run
+ int argc = 5;
+ char** argv = xbt_new(char*, argc);
+ argv[0] = xbt_strdup("1"); // log only?
+ argv[1] = xbt_strdup(job->smpi_app_name.c_str()); // application instance
+ argv[2] = bprintf("%d", rank); // rank
+ argv[3] = xbt_strdup(job->traces_filenames[rank].c_str()); // smpi trace file for this rank
+ argv[4] = xbt_strdup("0"); // ?
+
+ XBT_INFO("Replaying rank %d of job %d (smpi_app '%s')", rank, job->unique_job_number, job->smpi_app_name.c_str());
smpi_replay_run(&argc, &argv);
- XBT_INFO("Finished replaying rank %d of job %d (smpi_app '%s')", args->rank, args->job->unique_job_number,
- args->job->smpi_app_name.c_str());
+ XBT_INFO("Finished replaying rank %d of job %d (smpi_app '%s')", rank, job->unique_job_number,
+ job->smpi_app_name.c_str());
- args->barrier->wait();
+ barrier->wait();
- delete args;
- return 0;
+ // Memory clean-up — leaks can come from argc/argv modifications from smpi_replay_run
+ for (int i = 0; i < argc; ++i)
+ xbt_free(argv[i]);
+ xbt_free(argv);
}
// Sleeps for a given amount of time
simgrid::s4u::BarrierPtr barrier = simgrid::s4u::Barrier::create(job->app_size + 1);
for (int i = 0; i < job->app_size; ++i) {
- char** argv = xbt_new(char*, 5);
- argv[0] = xbt_strdup("1"); // log only?
- argv[1] = xbt_strdup(job->smpi_app_name.c_str()); // application instance
- argv[2] = bprintf("%d", i); // rank
- argv[3] = xbt_strdup(job->traces_filenames[i].c_str()); // smpi trace file for this rank
- argv[4] = xbt_strdup("0"); // ?
-
- s_smpi_replay_process_args* args = new s_smpi_replay_process_args;
- args->job = job;
- args->barrier = barrier;
- args->rank = i;
-
- char* str_pname = bprintf("%d_%d", job->unique_job_number, i);
- MSG_process_create_with_arguments(str_pname, smpi_replay_process, (void*)args, hosts[job->allocation[i]], 5, argv);
+ char* str_pname = bprintf("rank_%d_%d", job->unique_job_number, i);
+ simgrid::s4u::Actor::create(str_pname, hosts[job->allocation[i]], smpi_replay_process, job, barrier, i);
xbt_free(str_pname);
}
barrier->wait();
+ simgrid::s4u::this_actor::sleep_for(1);
XBT_INFO("Finished job %d (smpi_app '%s')", job->unique_job_number, job->smpi_app_name.c_str());
return 0;
}
// Let's finally run the job executor
- std::string job_process_name = "job_" + job->smpi_app_name;
+ char* str_pname = bprintf("job_%04d", job->unique_job_number);
XBT_INFO("Launching the job executor of job %d (app '%s')", job->unique_job_number, job->smpi_app_name.c_str());
- simgrid::s4u::Actor::create(job_process_name.c_str(), hosts[job->allocation[0]], job_executor_process, job);
+ simgrid::s4u::Actor::create(str_pname, hosts[job->allocation[0]], job_executor_process, job);
+ free(str_pname);
}
return 0;
argv[0], argv[0]);
// Simulation setting
- MSG_init(&argc, argv);
simgrid::s4u::Engine e(&argc, argv);
e.load_platform(argv[1]);
hosts = e.get_all_hosts();
// Let's register them
for (const Job* job : jobs)
- SMPI_app_instance_register(job->smpi_app_name.c_str(), smpi_replay_process, job->app_size);
+ SMPI_app_instance_register(job->smpi_app_name.c_str(), nullptr, job->app_size);
SMPI_init();
}
// Let's execute the workload
- simgrid::s4u::Actor::create("workload_executor", hosts[0], workload_executor_process, &jobs);
+ simgrid::s4u::Actor::create("workload", hosts[0], workload_executor_process, &jobs);
e.run();
XBT_INFO("Simulation finished! Final time: %g", e.get_clock());