TODO: add some notes for future work.

[loba.git] / process.cpp

diff --git a/process.cpp b/process.cpp
index db84ff82704e4292284af6520e1986f838ee57f9..e77217686c25ac7ff6961282fac698c8008d9263 100644 (file)
--- a/process.cpp
+++ b/process.cpp
@@ -20,6 +20,10 @@ 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::load_diff_threshold;
+
 namespace {
 
     void sleep_until_date(double& date, double duration)
@@ -54,17 +58,27 @@ 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;
 
+    convergence = -1.0;
+
+    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;
     finalizing = false;
 
-    comp_iter = lb_iter = 0;
+    all_comp_iter = comp_iter = lb_iter = 0;
 
     lb_thread = new_msg_thread("loba",
                                std::bind(&process::load_balance_loop, this));
@@ -93,8 +107,14 @@ process::~process()
                "received_load is %g, but should be 0.0 !", received_load);
     if (opt::log_rate < 0)
         return;
-    XBT_INFO("Final load after %d:%d iterations: %g",
-             lb_iter, comp_iter, real_load);
+    XBT_INFO("Final load after %d:%d:%d iterations: %g",
+             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);
+    else
+        XBT_INFO("Convergence within %g%% was not achieved",
+                 opt::avg_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);
@@ -144,7 +164,8 @@ void process::load_balance_loop()
         // nothing to do with opt::bookkeeping
 
         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
-            XBT_INFO("(%u:%u) current load: %g", lb_iter, comp_iter, real_load);
+            XBT_INFO("(%u:%u:%u) current load: %g",
+                     lb_iter, comp_iter, all_comp_iter, real_load);
             XBT_VERB("... expected load: %g", expected_load);
         }
 
@@ -194,9 +215,12 @@ void process::compute_loop()
                       std::bind(&process::data_send, this, _1));
         mutex.release();
 
+        ++all_comp_iter;
         if (real_load == 0.0)
             continue;
 
+        convergence_check();
+
         // compute
         ++comp_iter;
         double flops = opt::comp_cost(real_load);
@@ -224,9 +248,30 @@ void process::compute_loop()
     real_load += received_load;
     received_load = 0.0;
     total_load_running -= real_load;
+    convergence_check();
     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;
@@ -323,13 +368,18 @@ void process::data_send(neighbor& nb)
 {
     double load_to_send;
     if (opt::bookkeeping) {     // bookkeeping
-        double excess_load;
+        double excess_load;     // load amount we are able to send
         if (opt::egocentric)
             excess_load = std::max(0.0, real_load - expected_load);
         else
             excess_load = real_load;
-        load_to_send = compute_load_to_send(std::min(excess_load,
-                                                     nb.get_debt()));
+
+        double balance = nb.get_debt() - nb.get_credit();
+        load_to_send = std::min(excess_load,
+                                std::max(0.0, balance));
+
+        // adjust load to send (rounding, truncation, etc.)
+        load_to_send = compute_load_to_send(load_to_send);
         if (load_to_send > 0.0)
             nb.set_debt(nb.get_debt() - load_to_send);
     } else {                    // !bookkeeping
@@ -395,12 +445,18 @@ void process::handle_message(message* msg, m_host_t from)
     case message::CTRL: {
         neighbor* n = rev_neigh[from];
         n->set_load(msg->get_amount() + n->get_to_send());
-        expected_load += msg->get_credit(); // may be 0.0 if !opt::bookkeeping
+        if (opt::bookkeeping) {
+            double credit = msg->get_credit();
+            expected_load += credit;
+            n->set_credit(n->get_credit() + credit);
+        }
         break;
     }
     case message::DATA: {
+        neighbor* n = rev_neigh[from];
         double ld = msg->get_amount();
         received_load += ld;
+        n->set_credit(n->get_credit() - ld);
         break;
     }
     case message::CTRL_CLOSE: