Add statistics about idle duration.

[loba.git] / main.cpp

diff --git a/main.cpp b/main.cpp
index 61dfd42519d1b2250ff8e999101aea34a88de04a..45c58cede87dd13f559b1a144cfc42aefa154a70 100644 (file)
--- a/main.cpp
+++ b/main.cpp
@@ -1,4 +1,5 @@
 #include <cerrno>
+#include <cmath>
 #include <csignal>
 #include <cstring>              // strchr
 #include <iostream>
@@ -28,7 +29,7 @@ XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(main);
 #include "tracing.h"
 #include "version.h"
 
-#define DATA_DESCR_WIDTH 37
+#define DATA_DESCR_WIDTH 39
 
 namespace {
     // Failure exit status
@@ -50,6 +51,8 @@ namespace {
 
     statistics loads;
     statistics comps;
+    statistics comp_iterations;
+    statistics all_comp_iterations;
     statistics data_send_amount;
     statistics data_recv_amount;
     statistics data_send_count;
@@ -58,6 +61,8 @@ namespace {
     statistics ctrl_recv_amount;
     statistics ctrl_send_count;
     statistics ctrl_recv_count;
+    statistics idle_duration;
+    statistics convergence;
 
 }
 
@@ -77,6 +82,8 @@ static int simulation_main(int argc, char* argv[])
         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());
@@ -85,6 +92,10 @@ static int simulation_main(int argc, char* argv[])
         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
@@ -169,9 +180,14 @@ static void install_signal_handler()
     }
 }
 
+#define PR_VALUE(descr, format, ...)                                    \
+    XBT_INFO("| %.*s: " format, DATA_DESCR_WIDTH,                       \
+             descr ".................................................", \
+             __VA_ARGS__)
+
 #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[])
@@ -251,6 +267,12 @@ int main(int argc, char* argv[])
             } 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());
@@ -303,9 +325,13 @@ int main(int argc, char* argv[])
         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("Comp. iterations", comp_iterations);
+        PR_STATS("X-Comp. iterations", all_comp_iterations);
+        PR_STATS("Idle duration", idle_duration);
         PR_STATS("Data send amount", data_send_amount);
         PR_STATS("Data recv amount", data_recv_amount);
         PR_STATS("Data send count", data_send_count);
@@ -314,16 +340,27 @@ int main(int argc, char* argv[])
         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());
+        PR_VALUE("Total simulated time", "%g", simulated_time);
+        PR_VALUE("Total simulation time", "%g", simulation_time.duration());
+        PR_VALUE("Elapsed (wall clock) time", "%g", elapsed_time.duration());
         XBT_INFO("`----");
+
+        double load_imbalance = 100.0 * loads.get_stddev() / loads.get_mean();
+        double transfer_amount =
+            data_send_amount.get_sum() / opt::comm_cost(loads.get_sum());
+
+        XBT_INFO(",----[ Useful metrics ]");
+        PR_VALUE("Final load imbalance", "%g %s", load_imbalance,
+                 "percent of the load average");
+        PR_VALUE("Data transfer amount", "%g %s", transfer_amount,
+                 "times the total amount of data");
+        PR_VALUE("Number of hosts that converged", "%u / %u",
+                 convergence.get_count(), loads.get_count());
+        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());
+        XBT_INFO("`----");
+
     }
     if (exit_status)
         XBT_ERROR("Simulation failed (%#x).", exit_status);