Document SG 3.5 bugs, and workarounds taken to circumvent them.

[loba.git] / main.cpp

diff --git a/main.cpp b/main.cpp
index 3ee3ae1544ea3152b8ce4b8de5a32247442c80d7..fc42d750f9c992b8fe9a27b17f5aac7ec232b48c 100644 (file)
--- a/main.cpp
+++ b/main.cpp
@@ -3,13 +3,6 @@
 #include <stdexcept>
 #include <msg/msg.h>
 #include <xbt/log.h>
 #include <stdexcept>
 #include <msg/msg.h>
 #include <xbt/log.h>

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

 
 // Creates log categories
 XBT_LOG_NEW_CATEGORY(simu, "Root of simulation messages");
 
 // Creates log categories
 XBT_LOG_NEW_CATEGORY(simu, "Root of simulation messages");


@@ -21,22 +14,63 @@ XBT_LOG_NEW_SUBCATEGORY(loba, simu, "Messages from load-balancer");
 
 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(main);
 
 
 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 "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;
+
+}

 
 

-int simulation_main(int argc, char* argv[])
+static int simulation_main(int argc, char* argv[])

 {
     int result;
     process* proc;
     try {
         proc = opt::loba_algorithms.new_instance(opt::loba_algo, argc, argv);
 {
     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();
         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) {
         delete proc;
     }
     catch (std::invalid_argument& e) {


@@ -45,25 +79,40 @@ int simulation_main(int argc, char* argv[])
     return result;
 }
 
     return result;
 }
 

-void check_for_lost_load()
+static void check_for_lost_load()

 {
 {

-    const double threshold = 1e-4;

     double total_init = process::get_total_load_init();
     double total_init = process::get_total_load_init();

+

     double total_exit = process::get_total_load_exit();
     double lost = total_init - total_exit;
     double total_exit = process::get_total_load_exit();
     double lost = total_init - total_exit;

-    double lost_ratio = 100 * lost / total_init;
-    if (lost_ratio < -threshold) {
+    double lost_ratio = 100.0 * lost / total_init;
+    if (lost_ratio < -opt::load_ratio_threshold)

         CRITICAL2("Gained load at exit! %g (%g%%) <============",
         CRITICAL2("Gained load at exit! %g (%g%%) <============",

-                  lost, lost_ratio);
-    } else if (lost_ratio > threshold) {
+                  -lost, -lost_ratio);
+    else if (lost_ratio > opt::load_ratio_threshold)

         CRITICAL2("Lost load at exit! %g (%g%%) <============",
                   lost, lost_ratio);
         CRITICAL2("Lost load at exit! %g (%g%%) <============",
                   lost, lost_ratio);

-    } else {
-        DEBUG2("Total load at exit looks good: %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.


@@ -92,7 +141,7 @@ 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)


@@ -134,14 +183,26 @@ int main(int argc, char* argv[])
             MSG_launch_application(opt::deployment_file.c_str());
         }
 
             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.
         INFO1("Starting simulation at %f...", MSG_get_clock());
         res = MSG_main();
         simulated_time = MSG_get_clock();
         INFO1("Simulation ended at %f.", simulated_time);
         // Launch the MSG simulation.
         INFO1("Starting simulation at %f...", MSG_get_clock());
         res = MSG_main();
         simulated_time = MSG_get_clock();
         INFO1("Simulation ended at %f.", simulated_time);

-        check_for_lost_load();
+
+        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);
 


@@ -167,9 +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)