]> AND Private Git Repository - loba.git/blobdiff - main.cpp
Logo AND Algorithmique Numérique Distribuée

Private GIT Repository
Document a bug with parallel executions.
[loba.git] / main.cpp
index 12f1ae245140db23fa59c415f148edd5e34c7893..8a1d5c8408f74812a78f0e079eb913bfe33b13a4 100644 (file)
--- a/main.cpp
+++ b/main.cpp
@@ -53,6 +53,7 @@ namespace {
     statistics comps;
     statistics comp_iterations;
     statistics all_comp_iterations;
+    statistics iter_deviation;
     statistics data_send_amount;
     statistics data_recv_amount;
     statistics data_send_count;
@@ -61,6 +62,7 @@ namespace {
     statistics ctrl_recv_amount;
     statistics ctrl_send_count;
     statistics ctrl_recv_count;
+    statistics idle_duration;
     statistics convergence;
 
 }
@@ -69,51 +71,47 @@ static int simulation_main(int argc, char* argv[])
 {
     int result;
     process* proc;
-    try {
-        proc = opt::loba_algorithms.new_instance(opt::loba_algo, argc, argv);
-
-        proc_mutex->acquire();
-        ++proc_counter;
-        proc_mutex->release();
-
-        result = proc->run();
-
-        proc_mutex->acquire();
-        loads.push(proc->get_real_load());
-        comps.push(proc->get_comp_amount());
-        comp_iterations.push(proc->get_comp_iter());
-        all_comp_iterations.push(proc->get_all_comp_iter());
-        data_send_amount.push(proc->get_data_send_amount());
-        data_recv_amount.push(proc->get_data_recv_amount());
-        data_send_count.push(proc->get_data_send_count());
-        data_recv_count.push(proc->get_data_recv_count());
-        ctrl_send_amount.push(proc->get_ctrl_send_amount());
-        ctrl_recv_amount.push(proc->get_ctrl_recv_amount());
-        ctrl_send_count.push(proc->get_ctrl_send_count());
-        ctrl_recv_count.push(proc->get_ctrl_recv_count());
-        double c = proc->get_convergence();
-        if (c != -1)
-            convergence.push(c);
-
-        // 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;
-        proc_cond->broadcast();
-        while (proc_counter > 0)
-            proc_cond->wait(*proc_mutex);
-        proc_mutex->release();
-
-        delete proc;
-    }
-    catch (const std::invalid_argument& e) {
-        THROWF(arg_error, 0, "%s", e.what());
-    }
-    catch (const std::exception& e) {
-        THROWF(0, 0, "%s", e.what());
-    }
+
+    proc = opt::loba_algorithms.new_instance(opt::loba_algo, argc, argv);
+
+    proc_mutex->acquire();
+    ++proc_counter;
+    proc_mutex->release();
+
+    result = proc->run();
+
+    proc_mutex->acquire();
+    loads.push(proc->get_real_load());
+    comps.push(proc->get_comp_amount());
+    comp_iterations.push(proc->get_comp_iter());
+    all_comp_iterations.push(proc->get_all_comp_iter());
+    iter_deviation.push(proc->get_iter_deviation());
+    data_send_amount.push(proc->get_data_send_amount());
+    data_recv_amount.push(proc->get_data_recv_amount());
+    data_send_count.push(proc->get_data_send_count());
+    data_recv_count.push(proc->get_data_recv_count());
+    ctrl_send_amount.push(proc->get_ctrl_send_amount());
+    ctrl_recv_amount.push(proc->get_ctrl_recv_amount());
+    ctrl_send_count.push(proc->get_ctrl_send_count());
+    ctrl_recv_count.push(proc->get_ctrl_recv_count());
+    idle_duration.push(proc->get_idle_duration());
+    double c = proc->get_convergence();
+    if (c != -1.0)
+        convergence.push(c);
+
+    // 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;
+    proc_cond->broadcast();
+    while (proc_counter > 0)
+        proc_cond->wait(*proc_mutex);
+    proc_mutex->release();
+
+    delete proc;
+
     return result;
 }
 
@@ -185,17 +183,15 @@ static void install_signal_handler()
 
 #define PR_STATS(descr, st)                                             \
     XBT_INFO("| %.*s: %g / %g / %g", DATA_DESCR_WIDTH,                  \
-             descr " (total/avg./stddev).............................", \
+             descr " (sum/avg/dev)...................................", \
              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.
-    volatile int exit_status = 0;   // global exit status
-    volatile double simulated_time = -1.0;
+    int exit_status = 0;   // global exit status
+    double simulated_time = -1.0;
     timestamp elapsed_time(timestamp::wallclock_time);
     timestamp simulation_time(timestamp::cpu_time);
-    xbt_ex_t ex;
     MSG_error_t res;
 
     elapsed_time.start();
@@ -213,8 +209,6 @@ int main(int argc, char* argv[])
     }
 
     // Initialize some MSG internal data.
-    // Note: MSG_global_init() may throw an exception, but it seems
-    // impossible to catch it correctly :-(
     MSG_global_init(&argc, argv);
     install_signal_handler();
 
@@ -236,77 +230,65 @@ int main(int argc, char* argv[])
           version::date.c_str());
     opt::print();
 
-    TRY {
-        exit_status = EXIT_FAILURE_INIT; // =====
-
-        // 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.
-        XBT_DEBUG("Loading platform file...");
-        MSG_create_environment(opt::platform_file.c_str());
-        XBT_DEBUG("Creating hostdata...");
-        hostdata::create();
-        XBT_INFO("Loaded description of %zd hosts.", hostdata::size());
-        XBT_DEBUG("Deploying processes...");
-        if (opt::auto_depl::enabled) {
-            if (!opt::auto_depl::nhosts)
-                opt::auto_depl::nhosts = hostdata::size();
-            if (opt::auto_depl::nhosts > hostdata::size()) {
-                XBT_WARN("%u hosts is too much: limiting to %zu",
-                         opt::auto_depl::nhosts, hostdata::size());
-                opt::auto_depl::nhosts = hostdata::size();
-            }
-            if (opt::auto_depl::load == 0.0) {
-                XBT_WARN("Initial load is zero!  "
-                         "Falling back on old behaviour (load = nhosts).");
-                opt::auto_depl::load = opt::auto_depl::nhosts;
-            } else if (opt::auto_depl::load < 0.0)
-                opt::auto_depl::load =
-                    -opt::auto_depl::load * opt::auto_depl::nhosts;
-            double iload = std::trunc(opt::auto_depl::load);
-            if (opt::integer_transfer && opt::auto_depl::load != iload) {
-                XBT_WARN("Total load %g is not an integer.  Truncate it.",
-                         opt::auto_depl::load);
-                opt::auto_depl::load = iload;
-            }
-            MY_launch_application(); // it is already opt::* aware...
-        } else {
-            MSG_launch_application(opt::deployment_file.c_str());
+    // 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.
+    XBT_DEBUG("Loading platform file...");
+    MSG_create_environment(opt::platform_file.c_str());
+    XBT_DEBUG("Creating hostdata...");
+    hostdata::create();
+    XBT_INFO("Loaded description of %zd hosts.", hostdata::size());
+    XBT_DEBUG("Deploying processes...");
+    if (opt::auto_depl::enabled) {
+        if (!opt::auto_depl::nhosts)
+            opt::auto_depl::nhosts = hostdata::size();
+        if (opt::auto_depl::nhosts > hostdata::size()) {
+            XBT_WARN("%u hosts is too much: limiting to %zu",
+                     opt::auto_depl::nhosts, hostdata::size());
+            opt::auto_depl::nhosts = hostdata::size();
         }
+        if (opt::auto_depl::load == 0.0) {
+            XBT_WARN("Initial load is zero!  "
+                     "Falling back on old behaviour (load = nhosts).");
+            opt::auto_depl::load = opt::auto_depl::nhosts;
+        } else if (opt::auto_depl::load < 0.0)
+            opt::auto_depl::load =
+                -opt::auto_depl::load * opt::auto_depl::nhosts;
+        double iload = std::trunc(opt::auto_depl::load);
+        if (opt::integer_transfer && opt::auto_depl::load != iload) {
+            XBT_WARN("Total load %g is not an integer.  Truncate it.",
+                     opt::auto_depl::load);
+            opt::auto_depl::load = iload;
+        }
+        MY_launch_application(); // it is already opt::* aware...
+    } else {
+        MSG_launch_application(opt::deployment_file.c_str());
+    }
 
-        // Register tracing categories
-        TRACE_category_with_color(TRACE_CAT_COMP, TRACE_COLOR_COMP);
-        TRACE_category_with_color(TRACE_CAT_CTRL, TRACE_COLOR_CTRL);
-        TRACE_category_with_color(TRACE_CAT_DATA, TRACE_COLOR_DATA);
-
-        exit_status = EXIT_FAILURE_SIMU; // =====
-
-        proc_mutex = new mutex_t();
-        proc_cond = new condition_t();
+    // Register tracing categories
+    TRACE_category_with_color(TRACE_CAT_COMP, TRACE_COLOR_COMP);
+    TRACE_category_with_color(TRACE_CAT_CTRL, TRACE_COLOR_CTRL);
+    TRACE_category_with_color(TRACE_CAT_DATA, TRACE_COLOR_DATA);
 
-        // Launch the MSG simulation.
-        XBT_INFO("Starting simulation at %f...", MSG_get_clock());
-        res = MSG_main();
-        simulated_time = MSG_get_clock();
-        XBT_INFO("Simulation ended at %f.", simulated_time);
+    proc_mutex = new mutex_t();
+    proc_cond = new condition_t();
 
-        delete proc_cond;
-        delete proc_mutex;
+    // Launch the MSG simulation.
+    XBT_INFO("Starting simulation at %f...", MSG_get_clock());
+    res = MSG_main();
+    simulated_time = MSG_get_clock();
+    XBT_INFO("Simulation ended at %f.", simulated_time);
 
-        if (res != MSG_OK)
-            THROWF(0, 0, "MSG_main() failed with status %#x", res);
+    delete proc_cond;
+    delete proc_mutex;
 
-        exit_status = EXIT_NO_FAILURE; // =====
-    }
-    CATCH (ex) {
-        int len = strlen(ex.msg);
-        if (len > 0 && ex.msg[len - 1] == '\n')
-            ex.msg[len - 1] = '\0'; // strip the ending '\n'
-        XBT_ERROR("%s", ex.msg);
-        XBT_DEBUG("Error from %s() in %s:%d", ex.func, ex.file, ex.line);
-        xbt_ex_free(ex);
+    if (res == MSG_OK) {
+        exit_status = EXIT_NO_FAILURE;
+    } else {
+        XBT_ERROR("MSG_main() failed with status %#x", res);
+        exit_status = EXIT_FAILURE_SIMU;
     }
 
     // Clean the MSG simulation.
@@ -353,8 +335,11 @@ int main(int argc, char* argv[])
                  "times the total amount of data");
         PR_VALUE("Number of hosts that converged", "%u / %u",
                  convergence.get_count(), loads.get_count());
-        PR_VALUE("Date of first host convergence", "%g", convergence.get_min());
-        PR_VALUE("Date of last host convergence", "%g", convergence.get_max());
+        PR_VALUE("Times of convergence (min/max/avg/dev)", "%g / %g / %g / %g",
+                 convergence.get_min(), convergence.get_max(),
+                 convergence.get_mean(), convergence.get_stddev());
+        PR_STATS("Idle duration", idle_duration);
+        PR_STATS("Supernumer. comp. iter.", iter_deviation);
         XBT_INFO("`----");
 
     }