Check for convergence after final receive too.

Also use opt::load_ratio_threshold when comparing floating point values, when
checking for convergence.

diff --git a/process.cpp b/process.cpp
index 76524fddaec59b27d08c71c2d492a9383e1f15f9..e77217686c25ac7ff6961282fac698c8008d9263 100644 (file)
--- a/process.cpp
+++ b/process.cpp
@@ -22,6 +22,7 @@ double process::total_load_exit = 0.0;
 
 int process::process_counter = 0;
 double process::total_load_average;
 
 int process::process_counter = 0;
 double process::total_load_average;

+double process::load_diff_threshold;

 
 namespace {
 
 
 namespace {
 


@@ -70,6 +71,8 @@ process::process(int argc, char* argv[])
 
     process_counter++;
     total_load_average = total_load_running / process_counter;
 
     process_counter++;
     total_load_average = total_load_running / process_counter;

+    load_diff_threshold = (opt::load_ratio_threshold +
+                           opt::avg_load_ratio * total_load_average) / 100.0;

 
     ctrl_close_pending = data_close_pending = neigh.size();
     close_received = false;
 
     ctrl_close_pending = data_close_pending = neigh.size();
     close_received = false;


@@ -216,21 +219,7 @@ void process::compute_loop()
         if (real_load == 0.0)
             continue;
 
         if (real_load == 0.0)
             continue;
 

-        double load_ratio =
-            100.0 * std::fabs(real_load / total_load_average - 1.0);
-        if (convergence >= 0.0) {
-            if (load_ratio > opt::avg_load_ratio) {
-                XBT_VERB("current load has diverged: %g (%.4g%%)",
-                         real_load, load_ratio);
-                convergence = -1.0;
-            }
-        } else {
-            if (load_ratio <= opt::avg_load_ratio) {
-                XBT_VERB("current load has converged: %g (%.4g%%)",
-                         real_load, load_ratio);
-                convergence = MSG_get_clock();
-            }
-        }
+        convergence_check();

 
         // compute
         ++comp_iter;
 
         // compute
         ++comp_iter;


@@ -259,9 +248,30 @@ void process::compute_loop()
     real_load += received_load;
     received_load = 0.0;
     total_load_running -= real_load;
     real_load += received_load;
     received_load = 0.0;
     total_load_running -= real_load;

+    convergence_check();

     comm.data_flush(true);
 }
 
     comm.data_flush(true);
 }
 

+void process::convergence_check()
+{
+    double load_diff = std::fabs(real_load - total_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) {
+            XBT_VERB("current load has converged: %g (%.4g%%)",
+                     real_load,  100.0 * load_diff / total_load_average);
+            convergence = MSG_get_clock();
+        }
+    }
+}
+

 bool process::still_running()
 {
     static bool last_status = true;
 bool process::still_running()
 {
     static bool last_status = true;

diff --git a/process.h b/process.h
index ff8b7517d5b3d062881388859ec53d4fc4f60882..d2f12292caaf9c62c65115ad5b667133604cd9f4 100644 (file)
--- a/process.h
+++ b/process.h
@@ -87,6 +87,7 @@ private:
 
     static int process_counter;
     static double total_load_average;
 
     static int process_counter;
     static double total_load_average;

+    static double load_diff_threshold;

 
     typedef MAP_TEMPLATE<m_host_t, neighbor*> rev_neigh_type;
     neigh_type neigh;           // list of neighbors (do not alter
 
     typedef MAP_TEMPLATE<m_host_t, neighbor*> rev_neigh_type;
     neigh_type neigh;           // list of neighbors (do not alter


@@ -159,6 +160,9 @@ private:
     // Simulate computation loop
     void compute_loop();
 
     // Simulate computation loop
     void compute_loop();
 

+    // Check for convergence
+    void convergence_check();
+

     // Check if we need to stop
     bool still_running();
 
     // Check if we need to stop
     bool still_running();