Use a lock-free synchronized queue for the message queue.

[loba.git] / main.cpp

diff --git a/main.cpp b/main.cpp
index dcefafcf3c2fdc83f8091e1a63ac38a91e34df56..bce9749dd203324b21ac2985f39a97427d7d4068 100644 (file)
--- a/main.cpp
+++ b/main.cpp
@@ -1,4 +1,6 @@
-#include <cstring>
+#include <cerrno>
+#include <csignal>
+#include <cstring>              // strchr

 #include <iostream>
 #include <stdexcept>
 #include <msg/msg.h>
 #include <iostream>
 #include <stdexcept>
 #include <msg/msg.h>


@@ -11,6 +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(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);
 
 
 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(main);
 


@@ -20,10 +23,13 @@ XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(main);
 #include "options.h"
 #include "process.h"
 #include "statistics.h"
 #include "options.h"
 #include "process.h"
 #include "statistics.h"

+#include "synchro.h"

 #include "timer.h"
 #include "tracing.h"
 #include "version.h"
 
 #include "timer.h"
 #include "tracing.h"
 #include "version.h"
 

+#define DATA_DESCR_WIDTH 37
+

 namespace {
     // Failure exit status
     enum {
 namespace {
     // Failure exit status
     enum {


@@ -32,14 +38,25 @@ namespace {
         EXIT_FAILURE_INIT  = 0x02,  // failed to initialize simulator
         EXIT_FAILURE_SIMU  = 0x04,  // simulation failed
         EXIT_FAILURE_CLEAN = 0x08,  // error at cleanup
         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_OTHER = 0x20,  // other error

     };
 
     };
 

-    xbt_mutex_t proc_mutex;
-    xbt_cond_t proc_cond;
-    unsigned proc_counter;
-
-    struct statistics comps;
-    struct statistics loads;
+    // 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;

 
 }
 
 
 }
 


@@ -50,15 +67,23 @@ static int simulation_main(int argc, char* argv[])
     try {
         proc = opt::loba_algorithms.new_instance(opt::loba_algo, argc, argv);
 
     try {
         proc = opt::loba_algorithms.new_instance(opt::loba_algo, argc, argv);
 

-        xbt_mutex_acquire(proc_mutex);
+        proc_mutex->acquire();

         ++proc_counter;
         ++proc_counter;

-        xbt_mutex_release(proc_mutex);
+        proc_mutex->release();

 
         result = proc->run();
 
 
         result = proc->run();
 

-        xbt_mutex_acquire(proc_mutex);
-        comps.push(proc->get_comp());
+        proc_mutex->acquire();

         loads.push(proc->get_real_load());
         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
 
         // Synchronization barrier...
         // The goal is to circumvent a limitation in SimGrid (at least


@@ -66,15 +91,18 @@ static int simulation_main(int argc, char* argv[])
         // destroys a communication they had together.
 
         --proc_counter;
         // destroys a communication they had together.
 
         --proc_counter;

-        xbt_cond_broadcast(proc_cond);
+        proc_cond->broadcast();

         while (proc_counter > 0)
         while (proc_counter > 0)

-            xbt_cond_wait(proc_cond, proc_mutex);
-        xbt_mutex_release(proc_mutex);
+            proc_cond->wait(*proc_mutex);
+        proc_mutex->release();

 
         delete proc;
     }
 
         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;
 }
     }
     return result;
 }


@@ -82,35 +110,61 @@ static int simulation_main(int argc, char* argv[])
 static void check_for_lost_load()
 {
     double total_init = process::get_total_load_init();
 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)
     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)

-        XBT_CRITICAL("Gained load at exit! %g (%g%%) <============",
-                     -lost, -lost_ratio);
+        XBT_ERROR("Gained load at exit! %g (%g%%) <============",
+                  -lost, -lost_ratio);

     else if (lost_ratio > opt::load_ratio_threshold)
     else if (lost_ratio > opt::load_ratio_threshold)

-        XBT_CRITICAL("Lost load at exit! %g (%g%%) <============",
-                     lost, lost_ratio);
+        XBT_ERROR("Lost load at exit! %g (%g%%) <============",
+                  lost, lost_ratio);

     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)
     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)

-        XBT_CRITICAL("Negative running load at exit! %g (%g%%) <============",
-                     total_running, running_ratio);
+        XBT_ERROR("Negative running load at exit! %g (%g%%) <============",
+                  total_running, running_ratio);

     else if (running_ratio > opt::load_ratio_threshold)
     else if (running_ratio > opt::load_ratio_threshold)

-        XBT_CRITICAL("Remaining running load at exit! %g (%g%%) <============",
-                     total_running, running_ratio);
+        XBT_ERROR("Remaining running load at exit! %g (%g%%) <============",
+                  total_running, running_ratio);

     else
         XBT_VERB("Running load at exit looks good: %g (%g%%)",
                  total_running, running_ratio);
 }
 
     else
         XBT_VERB("Running load at exit looks good: %g (%g%%)",
                  total_running, running_ratio);
 }
 

+static void signal_handler(int /*sig*/)
+{
+    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 install_signal_handler()
+{
+    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)                                             \
 #define PR_STATS(descr, st)                                             \

-    XBT_INFO("| %.*s: %g / %g / %g", 39,                                \
-             descr " total/avg./stddev. at exit.........................", \
+    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[])
              st.get_sum(), st.get_mean(), st.get_stddev())
 
 int main(int argc, char* argv[])


@@ -118,10 +172,12 @@ 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;
     // 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;
 
     xbt_ex_t ex;
     MSG_error_t res;
 

+    elapsed_time.start();

     simulation_time.start();
 
     // Set default logging parameters
     simulation_time.start();
 
     // Set default logging parameters


@@ -139,6 +195,7 @@ 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);
     // 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);
 
     // Parse global parameters
     bool parse_res = opt::parse_args(&argc, argv);


@@ -166,14 +223,17 @@ int main(int argc, char* argv[])
         MSG_function_register_default(simulation_main);
 
         // Create the platform and the application.
         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());
         MSG_create_environment(opt::platform_file.c_str());

+        XBT_DEBUG("Creating hostdata...");

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

+        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()) {
         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 %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)
                 opt::auto_depl::nhosts = hostdata::size();
             }
             if (!opt::auto_depl::load)


@@ -184,15 +244,14 @@ int main(int argc, char* argv[])
         }
 
         // Register tracing categories
         }
 
         // 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; // =====
 
 
         exit_status = EXIT_FAILURE_SIMU; // =====
 

-        proc_mutex = xbt_mutex_init();
-        proc_cond = xbt_cond_init();
-        proc_counter = 0;
+        proc_mutex = new mutex_t();
+        proc_cond = new condition_t();

 
         // Launch the MSG simulation.
         XBT_INFO("Starting simulation at %f...", MSG_get_clock());
 
         // Launch the MSG simulation.
         XBT_INFO("Starting simulation at %f...", MSG_get_clock());


@@ -200,11 +259,11 @@ int main(int argc, char* argv[])
         simulated_time = MSG_get_clock();
         XBT_INFO("Simulation ended at %f.", simulated_time);
 
         simulated_time = MSG_get_clock();
         XBT_INFO("Simulation ended at %f.", simulated_time);
 

-        xbt_cond_destroy(proc_cond);
-        xbt_mutex_destroy(proc_mutex);
+        delete proc_cond;
+        delete proc_mutex;

 
         if (res != MSG_OK)
 
         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; // =====
     }
 
         exit_status = EXIT_NO_FAILURE; // =====
     }


@@ -228,14 +287,28 @@ 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();

+        elapsed_time.stop();

         check_for_lost_load();
         XBT_INFO(",----[ Results ]");
         PR_STATS("Load", loads);
         PR_STATS("Computation", comps);
         check_for_lost_load();
         XBT_INFO(",----[ Results ]");
         PR_STATS("Load", loads);
         PR_STATS("Computation", comps);

-        XBT_INFO("| Total simulated time...................: %g",
+        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);
                  simulated_time);

-        XBT_INFO("| Total simulation time..................: %g",
+        XBT_INFO("| %.*s: %g", DATA_DESCR_WIDTH,
+                 "Total simulation time.................................",

                  simulation_time.duration());
                  simulation_time.duration());

+        XBT_INFO("| %.*s: %g", DATA_DESCR_WIDTH,
+                 "Elapsed (wall clock) time.............................",
+                 elapsed_time.duration());

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