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

[loba.git] / main.cpp

diff --git a/main.cpp b/main.cpp
index 5da797a4a02ebdca019b03eb0522a478235cad0d..fc42d750f9c992b8fe9a27b17f5aac7ec232b48c 100644 (file)
--- a/main.cpp
+++ b/main.cpp
@@ -1,11 +1,6 @@
-#include <algorithm>
-#include <cmath>
 #include <cstring>
-#include <tr1/functional>
 #include <iostream>
-#include <numeric>
 #include <stdexcept>
-#include <vector>
 #include <msg/msg.h>
 #include <xbt/log.h>
 
@@ -24,6 +19,7 @@ XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(main);
 #include "misc.h"
 #include "options.h"
 #include "process.h"
+#include "statistics.h"
 #include "timer.h"
 #include "tracing.h"
 #include "version.h"
@@ -38,14 +34,13 @@ namespace {
         EXIT_FAILURE_CLEAN = 0x08,  // error at cleanup
     };
 
-    std::vector<double> comps;
-    double comp_total;
-    double comp_avg;
-    double comp_stddev;
+    xbt_mutex_t proc_mutex;
+    xbt_cond_t proc_cond;
+    unsigned proc_counter;
+
+    struct statistics comps;
+    struct statistics loads;
 
-    std::vector<double> loads;
-    double load_avg;
-    double load_stddev;
 }
 
 static int simulation_main(int argc, char* argv[])
@@ -54,9 +49,28 @@ static int simulation_main(int argc, char* argv[])
     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();
-        comps.push_back(proc->get_comp());
-        loads.push_back(proc->get_load());
+
+        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) {
@@ -74,7 +88,7 @@ static void check_for_lost_load()
     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);
+                  -lost, -lost_ratio);
     else if (lost_ratio > opt::load_ratio_threshold)
         CRITICAL2("Lost load at exit! %g (%g%%) <============",
                   lost, lost_ratio);
@@ -94,38 +108,10 @@ static void check_for_lost_load()
                total_running, running_ratio);
 }
 
-static void statistics(const std::vector<double>& vec,
-                       double* sum, double* avg, double* stddev)
-{
-    using std::tr1::bind;
-    using std::tr1::placeholders::_1;
-
-    unsigned n = vec.size();
-    double vec_sum = std::accumulate(vec.begin(), vec.end(), 0.0);
-    double vec_avg = vec_sum / n;
-
-    if (sum)
-        *sum = vec_sum;
-    if (avg)
-        *avg = vec_avg;
-
-    if (stddev) {
-        std::vector<double> diff(vec);
-        std::transform(diff.begin(), diff.end(), diff.begin(),
-                       bind(std::minus<double>(), _1, vec_avg));
-        double epsilon = std::accumulate(diff.begin(), diff.end(), 0.0);
-        double square_sum = std::inner_product(diff.begin(), diff.end(),
-                                               diff.begin(), 0.0);
-        double variance = (square_sum - (epsilon * epsilon) / n) / n;
-        *stddev = sqrt(variance);
-    }
-}
-
-static void compute_metrics()
-{
-    statistics(comps, &comp_total, &comp_avg, &comp_stddev);
-    statistics(loads, NULL, &load_avg, &load_stddev);
-}
+#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[])
 {
@@ -204,11 +190,19 @@ int main(int argc, char* argv[])
 
         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);
+
+        xbt_cond_destroy(proc_cond);
+        xbt_mutex_destroy(proc_mutex);
+
         if (res != MSG_OK)
             THROW1(0, 0, "MSG_main() failed with status %#x", res);
 
@@ -235,12 +229,9 @@ int main(int argc, char* argv[])
     if (simulated_time >= 0.0) {
         simulation_time.stop();
         check_for_lost_load();
-        compute_metrics();
         INFO0(",----[ Results ]");
-        INFO2("| Load avg./stddev. at exit..............: %g / %g",
-              load_avg, load_stddev);
-        INFO3("| Computation total/avg./stddev. at exit.: %g / %g / %g",
-              comp_total, comp_avg, comp_stddev);
+        PR_STATS("Load", loads);
+        PR_STATS("Computation", comps);
         INFO1("| Total simulated time...................: %g", simulated_time);
         INFO1("| Total simulation time..................: %g",
               simulation_time.duration());