Document a bug with parallel executions.

[loba.git] / process.cpp

diff --git a/process.cpp b/process.cpp
index e77217686c25ac7ff6961282fac698c8008d9263..c85b6051a6a4224452f22ddce086740f310ed61e 100644 (file)
--- a/process.cpp
+++ b/process.cpp
@@ -67,6 +67,7 @@ process::process(int argc, char* argv[])
     total_load_init += real_load;
     received_load = 0.0;
 
     total_load_init += real_load;
     received_load = 0.0;
 

+    idle_duration = 0.0;

     convergence = -1.0;
 
     process_counter++;
     convergence = -1.0;
 
     process_counter++;


@@ -120,6 +121,21 @@ process::~process()
     print_loads(true, xbt_log_priority_debug);
 }
 
     print_loads(true, xbt_log_priority_debug);
 }
 

+double process::get_iter_deviation() const
+{
+    double average_cost = opt::comp_cost(total_load_average);
+    // Do not count idle periods
+    double comp_iter_opt = acc.comp_amount / average_cost;
+/*
+    // Add iterations that could have been achieved while beeing idle
+    // (kept for documentation)
+    double self_speed = MSG_get_host_speed(MSG_host_self());
+    double average_duration = average_cost / self_speed;
+    comp_iter_opt += idle_duration / average_duration;
+*/
+    return comp_iter - comp_iter_opt;
+}
+

 int process::run()
 {
     if (opt::log_rate >= 0) {
 int process::run()
 {
     if (opt::log_rate >= 0) {


@@ -202,9 +218,12 @@ void process::compute_loop()
     using std::placeholders::_1;
 
     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
     using std::placeholders::_1;
 
     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;

+    double idle_since_date = 0.0;

     while (still_running()) {
     while (still_running()) {

-        // receive (do not block if there is something to compute)
-        data_receive(real_load > 0.0 ? 0.0 : opt::min_comp_iter_duration);
+        // receive
+        // if there is something to compute, do not block
+        // else, block the duration of an *lb* iteration
+        data_receive(real_load > 0.0 ? 0.0 : opt::min_lb_iter_duration);

 
         // send
         comm.data_flush(false);
 
         // send
         comm.data_flush(false);


@@ -222,6 +241,7 @@ void process::compute_loop()
         convergence_check();
 
         // compute
         convergence_check();
 
         // compute

+        idle_duration += MSG_get_clock() - idle_since_date;

         ++comp_iter;
         double flops = opt::comp_cost(real_load);
         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
         ++comp_iter;
         double flops = opt::comp_cost(real_load);
         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);


@@ -231,10 +251,13 @@ void process::compute_loop()
         add_comp_amount(flops);
         MSG_task_destroy(task);
 
         add_comp_amount(flops);
         MSG_task_destroy(task);
 

+        idle_since_date = MSG_get_clock();
+

         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
     }
 
     XBT_VERB("Going to finalize for %s...", __func__);
         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
     }
 
     XBT_VERB("Going to finalize for %s...", __func__);

+    // Note: idle duration is not counted during finalization

     finalizing = true;
     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
               neigh.size(), ESSE(neigh.size()));
     finalizing = true;
     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
               neigh.size(), ESSE(neigh.size()));