Move min_lb_iter_duration in load-balancing parameters.

[loba.git] / process.cpp

diff --git a/process.cpp b/process.cpp
index fe08b338b03120b39844a49e5f8ccaca0bfb6d52..65496385f08f45baf7d4e1b8015c78e7f54cf0e4 100644 (file)
--- a/process.cpp
+++ b/process.cpp
@@ -18,6 +18,18 @@ double process::total_load_init = 0.0;
 double process::total_load_running = 0.0;
 double process::total_load_exit = 0.0;
 
+namespace {
+
+    void sleep_until_date(double& date, double duration = 0.0)
+    {
+        double sleep_duration = date - MSG_get_clock();
+        if (sleep_duration > 0.0)
+            MSG_process_sleep(sleep_duration);
+        date = MSG_get_clock() + duration;
+    }
+
+}
+
 process::process(int argc, char* argv[])
 {
     if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))
@@ -46,9 +58,9 @@ process::process(int argc, char* argv[])
 
     comp_iter = lb_iter = 0;
 
-    compute_thread = new_msg_thread("compute",
-                                    std::tr1::bind(&process::compute_loop,
-                                                   this));
+    lb_thread = new_msg_thread("loba",
+                               std::tr1::bind(&process::load_balance_loop,
+                                              this));
 
     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
     if (!LOG_ISENABLED(logp))
@@ -68,28 +80,28 @@ process::process(int argc, char* argv[])
 
 process::~process()
 {
-    delete compute_thread;
+    delete lb_thread;
     total_load_exit += real_load;
+    if (opt::log_rate < 0)
+        return;
     if (opt::bookkeeping) {
         XBT_INFO("Final load after %d:%d iterations: %g ; expected: %g",
                  lb_iter, comp_iter, real_load, expected_load);
     } else {
-        XBT_INFO("Final load after %d iterations: %g",
-                 lb_iter, real_load);
-        if (lb_iter != comp_iter)
-            XBT_WARN("lb_iter (%d) and comp_iter (%d) differ!",
-                     lb_iter, comp_iter);
+        XBT_INFO("Final load after %d:%d iterations: %g",
+                 lb_iter, comp_iter, real_load);
     }
     XBT_VERB("Total computation for this process: %g", comp);
 }
 
 int process::run()
 {
-    XBT_INFO("Initial load: %g", real_load);
+    if (opt::log_rate >= 0)
+        XBT_INFO("Initial load: %g", real_load);
     XBT_VERB("Starting...");
-    compute_thread->start();
-    load_balance_loop();
-    compute_thread->wait();
+    lb_thread->start();
+    compute_loop();
+    lb_thread->wait();
     XBT_VERB("Done.");
     return 0;
 }
@@ -99,6 +111,7 @@ void process::load_balance_loop()
     using std::tr1::bind;
     using std::tr1::placeholders::_1;
 
+    double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
     while (still_running()) {
         ++lb_iter;
 
@@ -119,10 +132,10 @@ void process::load_balance_loop()
         // send
         std::for_each(neigh.begin(), neigh.end(),
                       bind(&process::ctrl_send, this, _1));
+        prev_load_broadcast = get_load();
 
-        // block on receiving unless there is something to compute or
-        // to send
-        ctrl_receive(opt::min_lb_iter_duration);
+        sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
+        ctrl_receive(0.0);
 
         comm.ctrl_flush(false);
     }
@@ -145,13 +158,13 @@ void process::compute_loop()
     using std::tr1::bind;
     using std::tr1::placeholders::_1;
 
-    double next_iter_after_date = 0.0;
+    double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
     while (still_running()) {
         // receive
-        double sleep_duration = real_load
-            ? std::max(MSG_get_clock() - next_iter_after_date, 0.0)
-            : opt::min_comp_iter_duration;
-        data_receive(sleep_duration);
+        if (real_load > 0.0)
+            data_receive(0.0);
+        else
+            data_receive(opt::min_comp_iter_duration);
 
         comm.data_flush(false);
 
@@ -172,10 +185,15 @@ void process::compute_loop()
         comp += flops;
         MSG_task_destroy(task);
 
-        next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
+        sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
     }
 
     XBT_VERB("Going to finalize for %s...", __func__);
+    // last send, for not losing load scheduled to be sent
+    std::for_each(neigh.begin(), neigh.end(),
+                  bind(&process::data_send, this, _1));
+    finalizing = true;
+    total_load_running -= real_load;
     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
               neigh.size(), ESSE(neigh.size()));
     std::for_each(neigh.begin(), neigh.end(),
@@ -219,7 +237,6 @@ bool process::still_running()
                opt::load_ratio_threshold) {
         // fixme: this check should be implemented with a distributed
         // algorithm, and not a shared global variable!
-        // fixme: should this chunk be moved before call to receive() ?
         XBT_VERB("No more load to balance in system.");
         last_status = false;
     }
@@ -298,6 +315,7 @@ void process::data_send(neighbor& nb)
     } else {
         load_to_send = nb.get_to_send();
         nb.set_to_send(0.0);
+        // do not update real_load here
     }
     if (load_to_send > 0.0)
         comm.data_send(nb.get_data_mbox(),