More updates to METRICS.

[loba.git] / main.cpp

diff --git a/main.cpp b/main.cpp
index a88378daa584a7321a48f71e4c7f0ad8b812dfe1..6417250d128315df81fa935b377f40037ee0a1c1 100644 (file)
--- a/main.cpp
+++ b/main.cpp
@@ -1,7 +1,8 @@
 #include <cerrno>
 #include <cerrno>

+#include <cmath>
+#include <csignal>

 #include <cstring>              // strchr
 #include <iostream>
 #include <cstring>              // strchr
 #include <iostream>

-#include <signal.h>

 #include <stdexcept>
 #include <msg/msg.h>
 #include <xbt/log.h>
 #include <stdexcept>
 #include <msg/msg.h>
 #include <xbt/log.h>


@@ -28,7 +29,7 @@ XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(main);
 #include "tracing.h"
 #include "version.h"
 
 #include "tracing.h"
 #include "version.h"
 

-#define DATA_DESCR_WIDTH 37
+#define DATA_DESCR_WIDTH 39

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


@@ -38,7 +39,9 @@ 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_OTHER = 0x10,  // other error
+        EXIT_FAILURE_INTR  = 0x10,  // interrupted by user
+        EXIT_FAILURE_LOAD  = 0x20,  // lost load on exit
+        EXIT_FAILURE_OTHER = 0x40,  // other error

     };
 
     // Cannot be globally initialized...
     };
 
     // Cannot be globally initialized...


@@ -48,6 +51,8 @@ namespace {
 
     statistics loads;
     statistics comps;
 
     statistics loads;
     statistics comps;

+    statistics comp_iterations;
+    statistics all_comp_iterations;

     statistics data_send_amount;
     statistics data_recv_amount;
     statistics data_send_count;
     statistics data_send_amount;
     statistics data_recv_amount;
     statistics data_send_count;


@@ -75,6 +80,8 @@ static int simulation_main(int argc, char* argv[])
         proc_mutex->acquire();
         loads.push(proc->get_real_load());
         comps.push(proc->get_comp_amount());
         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_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());


@@ -106,32 +113,38 @@ static int simulation_main(int argc, char* argv[])
     return result;
 }
 
     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;
     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)
+    if (lost_ratio < -opt::load_ratio_threshold) {

         XBT_ERROR("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)
+        res = false;
+    } else if (lost_ratio > opt::load_ratio_threshold) {

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

-    else
+        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;
         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)
+    if (running_ratio < -opt::load_ratio_threshold) {

         XBT_ERROR("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)
+        res = false;
+    } else if (running_ratio > opt::load_ratio_threshold) {

         XBT_ERROR("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
+        res = false;
+    } else

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

+    return res;

 }
 
 static void signal_handler(int /*sig*/)
 }
 
 static void signal_handler(int /*sig*/)


@@ -140,7 +153,12 @@ static void signal_handler(int /*sig*/)
         XBT_CRITICAL(">>>>>>>>>>"
                      " caught CTRL-C: global exit requested "
                      "<<<<<<<<<<");
         XBT_CRITICAL(">>>>>>>>>>"
                      " caught CTRL-C: global exit requested "
                      "<<<<<<<<<<");

-        opt::exit_request = true;
+        opt::exit_request = 1;
+    } else {
+        XBT_CRITICAL(">>>>>>>>>>"
+                     " caught CTRL-C for the 2nd time: exit immediately "
+                     "<<<<<<<<<<");
+        exit(EXIT_FAILURE_INTR);

     }
 }
 
     }
 }
 


@@ -150,7 +168,7 @@ static void install_signal_handler()
     action.sa_handler = signal_handler;
     sigemptyset(&action.sa_mask);
     action.sa_flags = SA_RESTART;
     action.sa_handler = signal_handler;
     sigemptyset(&action.sa_mask);
     action.sa_flags = SA_RESTART;

-    if (sigaction (SIGINT, &action, NULL) == -1) {
+    if (sigaction(SIGINT, &action, NULL) == -1) {

         std::cerr << "sigaction: " << strerror(errno) << "\n";
         exit(EXIT_FAILURE_OTHER);
     }
         std::cerr << "sigaction: " << strerror(errno) << "\n";
         exit(EXIT_FAILURE_OTHER);
     }


@@ -158,7 +176,7 @@ static void install_signal_handler()
 
 #define PR_STATS(descr, st)                                             \
     XBT_INFO("| %.*s: %g / %g / %g", DATA_DESCR_WIDTH,                  \
 
 #define PR_STATS(descr, st)                                             \
     XBT_INFO("| %.*s: %g / %g / %g", DATA_DESCR_WIDTH,                  \

-             descr " (total/avg./stddev)................................", \
+             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[])


@@ -221,6 +239,7 @@ int main(int argc, char* argv[])
         MSG_create_environment(opt::platform_file.c_str());
         XBT_DEBUG("Creating hostdata...");
         hostdata::create();
         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)
         XBT_DEBUG("Deploying processes...");
         if (opt::auto_depl::enabled) {
             if (!opt::auto_depl::nhosts)


@@ -230,8 +249,19 @@ int main(int argc, char* argv[])
                          opt::auto_depl::nhosts, hostdata::size());
                 opt::auto_depl::nhosts = hostdata::size();
             }
                          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 (load = nhosts).");

                 opt::auto_depl::load = opt::auto_depl::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());
             MY_launch_application(); // it is already opt::* aware...
         } else {
             MSG_launch_application(opt::deployment_file.c_str());


@@ -282,10 +312,13 @@ int main(int argc, char* argv[])
     if (simulated_time >= 0.0) {
         simulation_time.stop();
         elapsed_time.stop();
     if (simulated_time >= 0.0) {
         simulation_time.stop();
         elapsed_time.stop();

-        check_for_lost_load();
+        if (!check_for_lost_load())
+            exit_status |= EXIT_FAILURE_LOAD;

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

+        PR_STATS("Comp. iterations", comp_iterations);
+        PR_STATS("X-Comp. iterations", all_comp_iterations);

         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 send amount", data_send_amount);
         PR_STATS("Data recv amount", data_recv_amount);
         PR_STATS("Data send count", data_send_count);