X-Git-Url: https://bilbo.iut-bm.univ-fcomte.fr/and/gitweb/loba.git/blobdiff_plain/69ea6636935b8fae72f3915d521eeb2b12aabd34..c2e3e8bc788aa97edc116a151feb705cf77fce57:/main.cpp?ds=inline

diff --git a/main.cpp b/main.cpp
index 5da797a..aa964bb 100644
--- a/main.cpp
+++ b/main.cpp
@@ -1,11 +1,8 @@
-#include <algorithm>
-#include <cmath>
-#include <cstring>
-#include <tr1/functional>
+#include <cerrno>
+#include <csignal>
+#include <cstring>              // strchr
 #include <iostream>
-#include <numeric>
 #include <stdexcept>
-#include <vector>
 #include <msg/msg.h>
 #include <xbt/log.h>
 
@@ -16,6 +13,7 @@ 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_NEW_SUBCATEGORY(thrd, simu, "Messages from thread wrapper class");
 
 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(main);
 
@@ -24,10 +22,14 @@ XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(main);
 #include "misc.h"
 #include "options.h"
 #include "process.h"
+#include "statistics.h"
+#include "synchro.h"
 #include "timer.h"
 #include "tracing.h"
 #include "version.h"
 
+#define DATA_DESCR_WIDTH 37
+
 namespace {
     // Failure exit status
     enum {
@@ -36,16 +38,27 @@ namespace {
         EXIT_FAILURE_INIT  = 0x02,  // failed to initialize simulator
         EXIT_FAILURE_SIMU  = 0x04,  // simulation failed
         EXIT_FAILURE_CLEAN = 0x08,  // error at cleanup
+        EXIT_FAILURE_INTR  = 0x10,  // interrupted by user
+        EXIT_FAILURE_LOAD  = 0x20,  // lost load on exit
+        EXIT_FAILURE_OTHER = 0x40,  // other error
     };
 
-    std::vector<double> comps;
-    double comp_total;
-    double comp_avg;
-    double comp_stddev;
+    // Cannot be globally initialized...
+    mutex_t* proc_mutex;
+    condition_t* proc_cond;
+    unsigned proc_counter = 0;
+
+    statistics loads;
+    statistics comps;
+    statistics data_send_amount;
+    statistics data_recv_amount;
+    statistics data_send_count;
+    statistics data_recv_count;
+    statistics ctrl_send_amount;
+    statistics ctrl_recv_amount;
+    statistics ctrl_send_count;
+    statistics ctrl_recv_count;
 
-    std::vector<double> loads;
-    double load_avg;
-    double load_stddev;
 }
 
 static int simulation_main(int argc, char* argv[])
@@ -54,88 +67,124 @@ static int simulation_main(int argc, char* argv[])
     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();
-        comps.push_back(proc->get_comp());
-        loads.push_back(proc->get_load());
+
+        proc_mutex->acquire();
+        loads.push(proc->get_real_load());
+        comps.push(proc->get_comp_amount());
+        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());
+
+        // 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 (std::invalid_argument& e) {
-        THROW1(arg_error, 0, "%s", e.what());
+    catch (const std::invalid_argument& e) {
+        THROWF(arg_error, 0, "%s", e.what());
+    }
+    catch (const std::exception& e) {
+        THROWF(0, 0, "%s", e.what());
     }
     return result;
 }
 
-static void check_for_lost_load()
+static bool check_for_lost_load()
 {
+    bool res = true;
     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%%) <============",
+    if (lost_ratio < -opt::load_ratio_threshold) {
+        XBT_ERROR("Gained load at exit! %g (%g%%) <============",
+                  -lost, -lost_ratio);
+        res = false;
+    } else if (lost_ratio > opt::load_ratio_threshold) {
+        XBT_ERROR("Lost load at exit! %g (%g%%) <============",
                   lost, lost_ratio);
-    else
-        VERB2("Total load at exit looks good: %g (%g%%)", lost, lost_ratio);
+        res = false;
+    } else
+        XBT_VERB("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%%) <============",
+    if (running_ratio < -opt::load_ratio_threshold) {
+        XBT_ERROR("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%%) <============",
+        res = false;
+    } else if (running_ratio > opt::load_ratio_threshold) {
+        XBT_ERROR("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);
+        res = false;
+    } else
+        XBT_VERB("Running load at exit looks good: %g (%g%%)",
+                 total_running, running_ratio);
+    return res;
 }
 
-static void statistics(const std::vector<double>& vec,
-                       double* sum, double* avg, double* stddev)
+static void signal_handler(int /*sig*/)
 {
-    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);
+    if (!opt::exit_request) {
+        XBT_CRITICAL(">>>>>>>>>>"
+                     " caught CTRL-C: global exit requested "
+                     "<<<<<<<<<<");
+        opt::exit_request = 1;
+    } else {
+        XBT_CRITICAL(">>>>>>>>>>"
+                     " caught CTRL-C for the 2nd time: exit immediately "
+                     "<<<<<<<<<<");
+        exit(EXIT_FAILURE_INTR);
     }
 }
 
-static void compute_metrics()
+static void install_signal_handler()
 {
-    statistics(comps, &comp_total, &comp_avg, &comp_stddev);
-    statistics(loads, NULL, &load_avg, &load_stddev);
+    struct sigaction action;
+    action.sa_handler = signal_handler;
+    sigemptyset(&action.sa_mask);
+    action.sa_flags = SA_RESTART;
+    if (sigaction(SIGINT, &action, NULL) == -1) {
+        std::cerr << "sigaction: " << strerror(errno) << "\n";
+        exit(EXIT_FAILURE_OTHER);
+    }
 }
 
+#define PR_STATS(descr, st)                                             \
+    XBT_INFO("| %.*s: %g / %g / %g", DATA_DESCR_WIDTH,                  \
+             descr " (total/avg./stddev)................................", \
+             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;
-    timestamp simulation_time;
+    timestamp elapsed_time(timestamp::wallclock_time);
+    timestamp simulation_time(timestamp::cpu_time);
     xbt_ex_t ex;
     MSG_error_t res;
 
+    elapsed_time.start();
     simulation_time.start();
 
     // Set default logging parameters
@@ -153,6 +202,7 @@ int main(int argc, char* argv[])
     // Note: MSG_global_init() may throw an exception, but it seems
     // impossible to catch it correctly :-(
     MSG_global_init(&argc, argv);
+    install_signal_handler();
 
     // Parse global parameters
     bool parse_res = opt::parse_args(&argc, argv);
@@ -168,7 +218,7 @@ int main(int argc, char* argv[])
         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(),
+    XBT_INFO("%s v%s (%s)", opt::program_name.c_str(), version::num.c_str(),
           version::date.c_str());
     opt::print();
 
@@ -180,37 +230,52 @@ int main(int argc, char* argv[])
         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()) {
-                WARN2("%u hosts is too much: limiting to %u",
-                      opt::auto_depl::nhosts, (unsigned )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)
+            if (opt::auto_depl::load == 0.0) {
+                XBT_WARN("Initial load is zero!  Falling back on old behaviour.");
                 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;
             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);
+        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();
+
         // Launch the MSG simulation.
-        INFO1("Starting simulation at %f...", MSG_get_clock());
+        XBT_INFO("Starting simulation at %f...", MSG_get_clock());
         res = MSG_main();
         simulated_time = MSG_get_clock();
-        INFO1("Simulation ended at %f.", simulated_time);
+        XBT_INFO("Simulation ended at %f.", simulated_time);
+
+        delete proc_cond;
+        delete proc_mutex;
+
         if (res != MSG_OK)
-            THROW1(0, 0, "MSG_main() failed with status %#x", res);
+            THROWF(0, 0, "MSG_main() failed with status %#x", res);
 
         exit_status = EXIT_NO_FAILURE; // =====
     }
@@ -218,8 +283,8 @@ int main(int argc, char* argv[])
         int len = strlen(ex.msg);
         if (len > 0 && ex.msg[len - 1] == '\n')
             ex.msg[len - 1] = '\0'; // strip the ending '\n'
-        ERROR1("%s", ex.msg);
-        DEBUG3("Error from %s() in %s:%d", ex.func, ex.file, ex.line);
+        XBT_ERROR("%s", ex.msg);
+        XBT_DEBUG("Error from %s() in %s:%d", ex.func, ex.file, ex.line);
         xbt_ex_free(ex);
     }
 
@@ -227,29 +292,42 @@ int main(int argc, char* argv[])
     hostdata::destroy();
     res = MSG_clean();
     if (res != MSG_OK) {
-        ERROR1("MSG_clean() failed with status %#x", res);
+        XBT_ERROR("MSG_clean() failed with status %#x", res);
         exit_status |= EXIT_FAILURE_CLEAN;
     }
 
     // Report final simulation status.
     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);
-        INFO1("| Total simulated time...................: %g", simulated_time);
-        INFO1("| Total simulation time..................: %g",
-              simulation_time.duration());
-        INFO0("`----");
+        elapsed_time.stop();
+        if (!check_for_lost_load())
+            exit_status |= EXIT_FAILURE_LOAD;
+        XBT_INFO(",----[ Results ]");
+        PR_STATS("Load", loads);
+        PR_STATS("Computation", comps);
+        PR_STATS("Data send amount", data_send_amount);
+        PR_STATS("Data recv amount", data_recv_amount);
+        PR_STATS("Data send count", data_send_count);
+        PR_STATS("Data recv count", data_recv_count);
+        PR_STATS("Ctrl send amount", ctrl_send_amount);
+        PR_STATS("Ctrl recv amount", ctrl_recv_amount);
+        PR_STATS("Ctrl send count", ctrl_send_count);
+        PR_STATS("Ctrl recv count", ctrl_recv_count);
+        XBT_INFO("| %.*s: %g", DATA_DESCR_WIDTH,
+                 "Total simulated time..................................",
+                 simulated_time);
+        XBT_INFO("| %.*s: %g", DATA_DESCR_WIDTH,
+                 "Total simulation time.................................",
+                 simulation_time.duration());
+        XBT_INFO("| %.*s: %g", DATA_DESCR_WIDTH,
+                 "Elapsed (wall clock) time.............................",
+                 elapsed_time.duration());
+        XBT_INFO("`----");
     }
     if (exit_status)
-        ERROR1("Simulation failed (%#x).", exit_status);
+        XBT_ERROR("Simulation failed (%#x).", exit_status);
     else
-        INFO0("Simulation succeeded.");
+        XBT_INFO("Simulation succeeded.");
 
     return exit_status;
 }