X-Git-Url: https://bilbo.iut-bm.univ-fcomte.fr/and/gitweb/loba.git/blobdiff_plain/94b462631204e69491f6cb175ca210830bb11088..c3d3baabd62916d605d97758bf0f08ccfe479e01:/process.cpp

diff --git a/process.cpp b/process.cpp
index c85b605..97f2a69 100644
--- a/process.cpp
+++ b/process.cpp
@@ -16,14 +16,19 @@ XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
 
 #include "process.h"
 
+mutex_t *process::proc_mutex;
+
 double process::total_load_init = 0.0;
 double process::total_load_running = 0.0;
 double process::total_load_exit = 0.0;
 
 int process::process_counter = 0;
 double process::total_load_average;
+double process::average_load_ratio;
 double process::load_diff_threshold;
 
+std::atomic<int> process::convergence_counter(0);
+
 namespace {
 
     void sleep_until_date(double& date, double duration)
@@ -58,22 +63,27 @@ process::process(int argc, char* argv[])
         rev_neigh.insert(std::make_pair(host, ptr));
     }
 
-    // Note: there should not be race condition with the current
-    // version of Simgrid, when updating the global variables.
-
     prev_load_broadcast = -1;   // force sending of load on first send_all()
     expected_load = real_load;
-    total_load_running += real_load;
-    total_load_init += real_load;
     received_load = 0.0;
 
     idle_duration = 0.0;
     convergence = -1.0;
 
+    proc_mutex->acquire();
     process_counter++;
+    convergence_counter++;
+    total_load_init += real_load;
+    total_load_running += real_load;
     total_load_average = total_load_running / process_counter;
+    if (opt::avg_load_ratio >= 0.0)
+        average_load_ratio = opt::avg_load_ratio;
+    else
+        average_load_ratio = 100.0 *
+            (process_counter / -opt::avg_load_ratio) / total_load_average;
     load_diff_threshold = (opt::load_ratio_threshold +
-                           opt::avg_load_ratio * total_load_average) / 100.0;
+                           average_load_ratio * total_load_average) / 100.0;
+    proc_mutex->release();
 
     ctrl_close_pending = data_close_pending = neigh.size();
     close_received = false;
@@ -103,7 +113,9 @@ process::process(int argc, char* argv[])
 process::~process()
 {
     delete lb_thread;
+    proc_mutex->acquire();
     total_load_exit += real_load;
+    proc_mutex->release();
     xbt_assert(received_load == 0.0,
                "received_load is %g, but should be 0.0 !", received_load);
     if (opt::log_rate < 0)
@@ -112,10 +124,10 @@ process::~process()
              lb_iter, comp_iter, all_comp_iter, real_load);
     if (convergence >= 0.0)
         XBT_INFO("Convergence within %g%% was achieved at time %g",
-                 opt::avg_load_ratio, convergence);
+                 average_load_ratio, convergence);
     else
         XBT_INFO("Convergence within %g%% was not achieved",
-                 opt::avg_load_ratio);
+                 average_load_ratio);
     XBT_VERB("Expected load was: %g", expected_load);
     XBT_VERB("Total computation for this process: %g", get_comp_amount());
     print_loads(true, xbt_log_priority_debug);
@@ -123,7 +135,7 @@ process::~process()
 
 double process::get_iter_deviation() const
 {
-    double average_cost = opt::comp_cost(total_load_average);
+    double average_cost = opt::comp_cost(total_load_average); // fixme: get locked?
     // Do not count idle periods
     double comp_iter_opt = acc.comp_amount / average_cost;
 /*
@@ -270,27 +282,35 @@ void process::compute_loop()
     }
     real_load += received_load;
     received_load = 0.0;
+    proc_mutex->acquire();
     total_load_running -= real_load;
+    proc_mutex->release();
     convergence_check();
     comm.data_flush(true);
 }
 
 void process::convergence_check()
 {
-    double load_diff = std::fabs(real_load - total_load_average);
+    double average = total_load_average; // fixme: get locked?
+    double load_diff = std::fabs(real_load - average);
     bool converged = load_diff <= load_diff_threshold;
 
-    if (convergence >= 0.0) {
-        if (!converged) {
-            XBT_VERB("current load has diverged: %g (%.4g%%)",
-                     real_load, 100.0 * load_diff / total_load_average);
-            convergence = -1.0;
-        }
-    } else {
-        if (converged) {
+    if (converged) {
+        if (convergence < 0) {
             XBT_VERB("current load has converged: %g (%.4g%%)",
-                     real_load,  100.0 * load_diff / total_load_average);
+                     real_load,  100.0 * load_diff / average);
             convergence = MSG_get_clock();
+            local_convergence_counter = opt::exit_on_convergence;
+        }
+        if (local_convergence_counter > 0 && --local_convergence_counter == 0)
+                --convergence_counter;
+    } else {
+        if (convergence >= 0.0) {
+            XBT_VERB("current load has diverged: %g (%.4g%%)",
+                     real_load, 100.0 * load_diff / average);
+            convergence = -1.0;
+            if (local_convergence_counter == 0)
+                ++convergence_counter;
         }
     }
 }
@@ -318,6 +338,10 @@ bool process::still_running()
         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
         last_status = false;
 
+    } else if (opt::exit_on_convergence && convergence_counter == 0) {
+        XBT_VERB("Global convergence detected");
+        last_status = false;
+
     } else if (opt::exit_on_close && close_received) {
         XBT_VERB("Close received");
         last_status = false;
@@ -327,7 +351,7 @@ bool process::still_running()
         last_status = false;
 
     } else if (100.0 * total_load_running / total_load_init <=
-               opt::load_ratio_threshold) {
+               opt::load_ratio_threshold) { // fixme: get locked?
         // fixme: this check should be implemented with a distributed
         // algorithm, and not a shared global variable!
         XBT_VERB("No more load to balance in system.");