Add platforms with 1024 nodes.

[loba.git] / main.cpp

diff --git a/main.cpp b/main.cpp
index 61dfd42519d1b2250ff8e999101aea34a88de04a..f31e847d13244248afeccc28aef47a5b00c17510 100644 (file)
--- a/main.cpp
+++ b/main.cpp
@@ -1,4 +1,5 @@
 #include <cerrno>
 #include <cerrno>

+#include <cmath>

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


@@ -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


@@ -50,6 +51,9 @@ namespace {
 
     statistics loads;
     statistics comps;
 
     statistics loads;
     statistics comps;

+    statistics comp_iterations;
+    statistics all_comp_iterations;
+    statistics iter_deviation;

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


@@ -58,6 +62,8 @@ namespace {
     statistics ctrl_recv_amount;
     statistics ctrl_send_count;
     statistics ctrl_recv_count;
     statistics ctrl_recv_amount;
     statistics ctrl_send_count;
     statistics ctrl_recv_count;

+    statistics idle_duration;
+    statistics convergence;

 
 }
 
 
 }
 


@@ -65,46 +71,47 @@ static int simulation_main(int argc, char* argv[])
 {
     int result;
     process* proc;
 {
     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());
-        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 (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;
 }
 
     return result;
 }
 


@@ -169,19 +176,22 @@ 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,                  \
 #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[])
 {
              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);
     timestamp elapsed_time(timestamp::wallclock_time);
     timestamp simulation_time(timestamp::cpu_time);

-    xbt_ex_t ex;

     MSG_error_t res;
 
     elapsed_time.start();
     MSG_error_t res;
 
     elapsed_time.start();


@@ -199,8 +209,6 @@ int main(int argc, char* argv[])
     }
 
     // Initialize some MSG internal data.
     }
 
     // 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();
 
     MSG_global_init(&argc, argv);
     install_signal_handler();
 


@@ -222,71 +230,67 @@ int main(int argc, char* argv[])
           version::date.c_str());
     opt::print();
 
           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;
-            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();
+    process::set_proc_mutex(proc_mutex);

 
 

-        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);
+    process::set_proc_mutex(NULL);
+    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.
     }
 
     // Clean the MSG simulation.


@@ -303,9 +307,12 @@ int main(int argc, char* argv[])
         elapsed_time.stop();
         if (!check_for_lost_load())
             exit_status |= EXIT_FAILURE_LOAD;
         elapsed_time.stop();
         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);


@@ -314,16 +321,29 @@ 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);
         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());
+        PR_STATS("Idle duration", idle_duration);
+        PR_STATS("Supernumer. comp. iter.", iter_deviation);

         XBT_INFO("`----");
         XBT_INFO("`----");

+

     }
     if (exit_status)
         XBT_ERROR("Simulation failed (%#x).", exit_status);
     }
     if (exit_status)
         XBT_ERROR("Simulation failed (%#x).", exit_status);