Use MSG_comm_waitall for communicator::flush(true).

[loba.git] / main.cpp

diff --git a/main.cpp b/main.cpp
index a40bb9fa1e11fe26a7ec4898898bcbd8ab655633..fc42d750f9c992b8fe9a27b17f5aac7ec232b48c 100644 (file)
--- a/main.cpp
+++ b/main.cpp
@@ -1,40 +1,118 @@
 #include <cstring>
 #include <iostream>
 #include <cstring>
 #include <iostream>

+#include <stdexcept>

 #include <msg/msg.h>
 #include <xbt/log.h>
 #include <msg/msg.h>
 #include <xbt/log.h>

-#include "hostdata.h"
-#include "misc.h"
-#include "options.h"
-#include "process.h"
-#include "timer.h"
-#include "version.h"

 
 // Creates log categories
 
 // Creates log categories

-XBT_LOG_NEW_CATEGORY(simu, "Simulation messages");
+XBT_LOG_NEW_CATEGORY(simu, "Root of simulation messages");

 XBT_LOG_NEW_SUBCATEGORY(main, simu, "Messages from global infrastructure");
 XBT_LOG_NEW_SUBCATEGORY(main, simu, "Messages from global infrastructure");

+XBT_LOG_NEW_SUBCATEGORY(depl, main, "Messages from auto deployment");

 XBT_LOG_NEW_SUBCATEGORY(comm, simu, "Messages from asynchronous pipes");
 XBT_LOG_NEW_SUBCATEGORY(proc, simu, "Messages from base process class");
 XBT_LOG_NEW_SUBCATEGORY(loba, simu, "Messages from load-balancer");
 
 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(main);
 
 XBT_LOG_NEW_SUBCATEGORY(comm, simu, "Messages from asynchronous pipes");
 XBT_LOG_NEW_SUBCATEGORY(proc, simu, "Messages from base process class");
 XBT_LOG_NEW_SUBCATEGORY(loba, simu, "Messages from load-balancer");
 
 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(main);
 

-// Failure exit status
-enum {
-    EXIT_NO_FAILURE    = 0x00,  // no error
-    EXIT_FAILURE_ARGS  = 0x01,  // bad arguments
-    EXIT_FAILURE_INIT  = 0x02,  // failed to initialize simulator
-    EXIT_FAILURE_SIMU  = 0x04,  // simulation failed
-    EXIT_FAILURE_CLEAN = 0x08,  // error at cleanup
-};
-
-#include "loba_least_loaded.h"
-int simulation_main(int argc, char* argv[])
+#include "deployment.h"
+#include "hostdata.h"
+#include "misc.h"
+#include "options.h"
+#include "process.h"
+#include "statistics.h"
+#include "timer.h"
+#include "tracing.h"
+#include "version.h"
+
+namespace {
+    // Failure exit status
+    enum {
+        EXIT_NO_FAILURE    = 0x00,  // no error
+        EXIT_FAILURE_ARGS  = 0x01,  // bad arguments
+        EXIT_FAILURE_INIT  = 0x02,  // failed to initialize simulator
+        EXIT_FAILURE_SIMU  = 0x04,  // simulation failed
+        EXIT_FAILURE_CLEAN = 0x08,  // error at cleanup
+    };
+
+    xbt_mutex_t proc_mutex;
+    xbt_cond_t proc_cond;
+    unsigned proc_counter;
+
+    struct statistics comps;
+    struct statistics loads;
+
+}
+
+static int simulation_main(int argc, char* argv[])

 {
 {

-    //    process proc(argc, argv);
-    loba_least_loaded proc(argc, argv);
-    return proc.run();
+    int result;
+    process* proc;
+    try {
+        proc = opt::loba_algorithms.new_instance(opt::loba_algo, argc, argv);
+
+        xbt_mutex_acquire(proc_mutex);
+        ++proc_counter;
+        xbt_mutex_release(proc_mutex);
+
+        result = proc->run();
+
+        xbt_mutex_acquire(proc_mutex);
+        comps.push(proc->get_comp());
+        loads.push(proc->get_real_load());
+
+        // Synchronization barrier...
+        // The goal is to circumvent a limitation in SimGrid (at least
+        // in version 3.5): a process must be alive when another one
+        // destroys a communication they had together.
+
+        --proc_counter;
+        xbt_cond_broadcast(proc_cond);
+        while (proc_counter > 0)
+            xbt_cond_wait(proc_cond, proc_mutex);
+        xbt_mutex_release(proc_mutex);
+
+        delete proc;
+    }
+    catch (std::invalid_argument& e) {
+        THROW1(arg_error, 0, "%s", e.what());
+    }
+    return result;
+}
+
+static void check_for_lost_load()
+{
+    double total_init = process::get_total_load_init();
+
+    double total_exit = process::get_total_load_exit();
+    double lost = total_init - total_exit;
+    double lost_ratio = 100.0 * lost / total_init;
+    if (lost_ratio < -opt::load_ratio_threshold)
+        CRITICAL2("Gained load at exit! %g (%g%%) <============",
+                  -lost, -lost_ratio);
+    else if (lost_ratio > opt::load_ratio_threshold)
+        CRITICAL2("Lost load at exit! %g (%g%%) <============",
+                  lost, lost_ratio);
+    else
+        VERB2("Total load at exit looks good: %g (%g%%)", lost, lost_ratio);
+
+    double total_running = process::get_total_load_running();
+    double running_ratio = 100.0 * total_running / total_init;
+    if (running_ratio < -opt::load_ratio_threshold)
+        CRITICAL2("Negative running load at exit! %g (%g%%) <============",
+                  total_running, running_ratio);
+    else if (running_ratio > opt::load_ratio_threshold)
+        CRITICAL2("Remaining running load at exit! %g (%g%%) <============",
+                  total_running, running_ratio);
+    else
+        VERB2("Running load at exit looks good: %g (%g%%)",
+               total_running, running_ratio);

 }
 
 }
 

+#define PR_STATS(descr, st)                                             \
+    INFO5("| %.*s: %g / %g / %g", 39,                                   \
+          descr " total/avg./stddev. at exit.........................", \
+          st.get_sum(), st.get_mean(), st.get_stddev())
+

 int main(int argc, char* argv[])
 {
     // Note: variables used after THROW must be declared as volatile.
 int main(int argc, char* argv[])
 {
     // Note: variables used after THROW must be declared as volatile.


@@ -47,8 +125,15 @@ int main(int argc, char* argv[])
     simulation_time.start();
 
     // Set default logging parameters
     simulation_time.start();
 
     // Set default logging parameters

-    // xbt_log_control_set("simu.thres:verbose");
-    xbt_log_control_set("simu.fmt:'[%h %r] [%c/%p] %m%n'");
+    bool do_log_control_set = true;
+    for (int i = 1 ; do_log_control_set && i < argc ; i++)
+        do_log_control_set = !(argv[i][0] == '-' && argv[i][1] != '-' &&
+                               strchr(argv[i] + 1, 'v'));
+    if (do_log_control_set) {
+        // xbt_log_control_set("simu.thres:verbose");
+        xbt_log_control_set("simu.fmt:'[%h %r] [%c/%p] %m%n'");
+        xbt_log_control_set("main.fmt:'[%c/%p] %m%n'");
+    }

 
     // Initialize some MSG internal data.
     // Note: MSG_global_init() may throw an exception, but it seems
 
     // Initialize some MSG internal data.
     // Note: MSG_global_init() may throw an exception, but it seems


@@ -56,11 +141,12 @@ int main(int argc, char* argv[])
     MSG_global_init(&argc, argv);
 
     // Parse global parameters
     MSG_global_init(&argc, argv);
 
     // Parse global parameters

-    int parse_res = opt::parse_args(&argc, argv);
+    bool parse_res = opt::parse_args(&argc, argv);

     if (!parse_res
         || opt::version_requested || opt::help_requested) {
         if (opt::version_requested)
     if (!parse_res
         || opt::version_requested || opt::help_requested) {
         if (opt::version_requested)

-            std::clog << version::name << " version " << version::num << "\n"
+            std::clog << version::name << " (" << opt::program_name << ")"
+                      << " version " << version::num << "\n"

                       << version::copyright << "\n"
                 "Compiled on " << version::date << "\n\n";
         if (!parse_res || opt::help_requested)
                       << version::copyright << "\n"
                 "Compiled on " << version::date << "\n\n";
         if (!parse_res || opt::help_requested)


@@ -68,28 +154,55 @@ int main(int argc, char* argv[])
         MSG_clean();
         exit(parse_res ? EXIT_NO_FAILURE : EXIT_FAILURE_ARGS);
     }
         MSG_clean();
         exit(parse_res ? EXIT_NO_FAILURE : EXIT_FAILURE_ARGS);
     }

+    INFO3("%s v%s (%s)", opt::program_name.c_str(), version::num.c_str(),
+          version::date.c_str());

     opt::print();
 
     TRY {
         exit_status = EXIT_FAILURE_INIT; // =====
 
     opt::print();
 
     TRY {
         exit_status = EXIT_FAILURE_INIT; // =====
 

-        // Register the main function of an agent in a global table.
-        MSG_function_register("simulation_main", simulation_main);
-        // Preserve some compatibility with old code...
-        MSG_function_register("Calculs", simulation_main);
+        // Register the default function of an agent
+        // MSG_function_register("simulation_main", simulation_main);
+        MSG_function_register_default(simulation_main);

 
         // Create the platform and the application.
 
         // Create the platform and the application.

-        MSG_create_environment(opt::platform_file);
+        MSG_create_environment(opt::platform_file.c_str());

         hostdata::create();
         hostdata::create();

-        MSG_launch_application(opt::application_file);
+        if (opt::auto_depl::enabled) {
+            if (!opt::auto_depl::nhosts)
+                opt::auto_depl::nhosts = hostdata::size();
+            if (opt::auto_depl::nhosts > hostdata::size()) {
+                WARN2("%u hosts is too much: limiting to %u",
+                      opt::auto_depl::nhosts, (unsigned )hostdata::size());
+                opt::auto_depl::nhosts = hostdata::size();
+            }
+            if (!opt::auto_depl::load)
+                opt::auto_depl::load = opt::auto_depl::nhosts;
+            MY_launch_application(); // it is already opt::* aware...
+        } else {
+            MSG_launch_application(opt::deployment_file.c_str());
+        }
+
+        // Register tracing categories
+        TRACE_category(TRACE_CAT_COMP);
+        TRACE_category(TRACE_CAT_CTRL);
+        TRACE_category(TRACE_CAT_DATA);

 
         exit_status = EXIT_FAILURE_SIMU; // =====
 
 
         exit_status = EXIT_FAILURE_SIMU; // =====
 

+        proc_mutex = xbt_mutex_init();
+        proc_cond = xbt_cond_init();
+        proc_counter = 0;
+

         // Launch the MSG simulation.
         // Launch the MSG simulation.

-        INFO0("Starting simulation...");
+        INFO1("Starting simulation at %f...", MSG_get_clock());

         res = MSG_main();
         res = MSG_main();

-        INFO0("Simulation ended.");

         simulated_time = MSG_get_clock();
         simulated_time = MSG_get_clock();

+        INFO1("Simulation ended at %f.", simulated_time);
+
+        xbt_cond_destroy(proc_cond);
+        xbt_mutex_destroy(proc_mutex);
+

         if (res != MSG_OK)
             THROW1(0, 0, "MSG_main() failed with status %#x", res);
 
         if (res != MSG_OK)
             THROW1(0, 0, "MSG_main() failed with status %#x", res);
 


@@ -115,9 +228,13 @@ int main(int argc, char* argv[])
     // Report final simulation status.
     if (simulated_time >= 0.0) {
         simulation_time.stop();
     // Report final simulation status.
     if (simulated_time >= 0.0) {
         simulation_time.stop();

+        check_for_lost_load();

         INFO0(",----[ Results ]");
         INFO0(",----[ Results ]");

-        INFO1("| Total simulated time...: %g", simulated_time);
-        INFO1("| Total simulation time..: %g", simulation_time.duration());
+        PR_STATS("Load", loads);
+        PR_STATS("Computation", comps);
+        INFO1("| Total simulated time...................: %g", simulated_time);
+        INFO1("| Total simulation time..................: %g",
+              simulation_time.duration());

         INFO0("`----");
     }
     if (exit_status)
         INFO0("`----");
     }
     if (exit_status)