Display load average and standard deviation at the end.

[loba.git] / process.cpp

diff --git a/process.cpp b/process.cpp
index 214b9d05a785d80edeb436f14466a6d8140545d9..8b316787b8ae1d8654e0f941337a64c993573f24 100644 (file)
--- a/process.cpp
+++ b/process.cpp
@@ -15,6 +15,7 @@ XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
 #include "process.h"
 
 double process::total_load_init = 0.0;
 #include "process.h"
 
 double process::total_load_init = 0.0;

+double process::total_load_running = 0.0;

 double process::total_load_exit = 0.0;
 
 process::process(int argc, char* argv[])
 double process::total_load_exit = 0.0;
 
 process::process(int argc, char* argv[])


@@ -32,7 +33,9 @@ process::process(int argc, char* argv[])
         rev_neigh.insert(std::make_pair(host, ptr));
     }
 
         rev_neigh.insert(std::make_pair(host, ptr));
     }
 

+    prev_load_broadcast = -1;   // force sending of load on first send()

     expected_load = load;
     expected_load = load;

+    total_load_running += load;

     total_load_init += load;
 
     ctrl_close_pending = data_close_pending = neigh.size();
     total_load_init += load;
 
     ctrl_close_pending = data_close_pending = neigh.size();


@@ -40,7 +43,11 @@ process::process(int argc, char* argv[])
         comm.next_close_on_ctrl_is_last();
         comm.next_close_on_data_is_last();
     }
         comm.next_close_on_ctrl_is_last();
         comm.next_close_on_data_is_last();
     }

-    if (neigh.size() > 0)
+    close_received = false;
+    may_receive =  (neigh.size() > 0); // the same as (ctrl_close_pending ||
+                                       //              data_close_pending)
+    finalizing = false;
+    if (may_receive)

         comm.listen();
 
     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
         comm.listen();
 
     e_xbt_log_priority_t logp = xbt_log_priority_verbose;


@@ -56,7 +63,7 @@ process::process(int argc, char* argv[])
         oss << neigh.back().get_name();
     }
     LOG1(logp, "Got %s.", oss.str().c_str());
         oss << neigh.back().get_name();
     }
     LOG1(logp, "Got %s.", oss.str().c_str());

-    print_loads(true, logp);
+    print_loads(false, logp);

 }
 
 process::~process()
 }
 
 process::~process()


@@ -68,54 +75,61 @@ int process::run()
 {
     INFO1("Initial load: %g", load);
     VERB0("Starting...");
 {
     INFO1("Initial load: %g", load);
     VERB0("Starting...");

-    // first send() to inform neighbors about our load (force it)
-    prev_load_broadcast = -1;

     iter = 0;
     iter = 0;

-    bool one_more = true;
-    do {
-        ++iter;
+    while (true) {
+        if (load > 0.0) {
+            ++iter;
+            if (opt::log_rate && iter % opt::log_rate == 0) {
+                if (opt::bookkeeping)
+                    INFO3("(%u) current load: %g ; expected: %g",
+                          iter, load, expected_load);
+                else
+                    INFO2("(%u) current load: %g",
+                          iter, load);
+            }

 
 

-        if (opt::log_rate && iter % opt::log_rate == 0) {

             if (opt::bookkeeping)
             if (opt::bookkeeping)

-                INFO3("(%u) current load: %g ; expected: %g",
-                      iter, load, expected_load);
+                expected_load -= load_balance(expected_load);

             else
             else

-                INFO2("(%u) current load: %g",
-                      iter, load);
-        }
-        print_loads(true, xbt_log_priority_debug);
+                load -= load_balance(load);

 
 

-        if (opt::bookkeeping)
-            expected_load -= load_balance(expected_load);
-        else
-            load -= load_balance(load);
+            print_loads(true, xbt_log_priority_debug);

 
 

-        send();
-        compute();
+            send();
+            compute();

 
 

-// NDS for Need To Send
-#define NDS ((opt::bookkeeping ? expected_load : load) != prev_load_broadcast)
-        bool can_recv;
-        do {
-            // General idea: do not iterate if there is nothing to
-            // compute, nor to send.
+            if (opt::maxiter && iter >= opt::maxiter)
+                break;
+        } else {
+            // send load information, and load when bookkeeping
+            send();
+        }

 
 

-            // fixme: review this chunk, and remove this NDS macro!
+        // block on receiving unless there is something to compute or
+        // to send
+        bool recv_wait = (load == 0 &&
+                          ((opt::bookkeeping ? expected_load : load)
+                           == prev_load_broadcast));
+        DEBUG1("CALL RECEIVE(%s)", recv_wait? "WAIT": "NO_WAIT");
+        receive(recv_wait? WAIT: NO_WAIT);

 
 

-            bool recv_wait = (load == 0 && !NDS);
-            bool close_received = !receive(recv_wait? WAIT: NO_WAIT);
+        // one of our neighbor is finalizing
+        if (opt::exit_on_close && close_received)
+            break;

 
 

-            if (opt::exit_on_close && close_received)
-                one_more = false;
-            else if (opt::maxiter && iter >= opt::maxiter)
-                one_more = false;
-            
-            can_recv = (ctrl_close_pending || data_close_pending);
+        // have no load and cannot receive anything
+        if (load == 0.0 && !may_receive)
+            break;

 
 

-        } while (one_more && can_recv && load == 0 && !NDS);
-#undef NDS
+        // fixme: this check should be implemented with a distributed
+        // algorithm, and not a shared global variable!
+        if (100.0 * total_load_running / total_load_init <=
+            opt::load_ratio_threshold) {
+            VERB0("No more load to balance in system, stopping.");
+            break;
+        }

 
 

-    } while (one_more);
+    }

     VERB0("Going to finalize...");
     finalize();
 
     VERB0("Going to finalize...");
     finalize();
 


@@ -218,15 +232,14 @@ void process::send()
     }
 }
 
     }
 }
 

-bool process::receive(recv_wait_mode wait)
+void process::receive(recv_wait_mode wait)

 {
     // DEBUG1("go for receive(%s)",
     //        "NO_WAIT\0WAIT\0\0\0\0WAIT_FOR_CLOSE" + 8 * wait);
 {
     // DEBUG1("go for receive(%s)",
     //        "NO_WAIT\0WAIT\0\0\0\0WAIT_FOR_CLOSE" + 8 * wait);

-    bool result = true;

     message* msg;
     m_host_t from;
     message* msg;
     m_host_t from;

-    bool do_recv = ctrl_close_pending || data_close_pending;
-    while (do_recv && comm.recv(msg, from, wait)) {
+    bool do_wait = (wait != NO_WAIT);
+    while (may_receive && comm.recv(msg, from, do_wait)) {

         switch (msg->get_type()) {
         case message::INFO: {
             neighbor* n = rev_neigh[from];
         switch (msg->get_type()) {
         case message::INFO: {
             neighbor* n = rev_neigh[from];


@@ -236,33 +249,36 @@ bool process::receive(recv_wait_mode wait)
         case message::CREDIT:
             expected_load += msg->get_amount();
             break;
         case message::CREDIT:
             expected_load += msg->get_amount();
             break;

-        case message::LOAD:
-            load += msg->get_amount();
+        case message::LOAD: {
+            double ld = msg->get_amount();
+            load += ld;
+            if (finalizing)
+                total_load_running -= ld;

             break;
             break;

+        }

         case message::CTRL_CLOSE:
             if (--ctrl_close_pending == 1)
                 comm.next_close_on_ctrl_is_last();
         case message::CTRL_CLOSE:
             if (--ctrl_close_pending == 1)
                 comm.next_close_on_ctrl_is_last();

-            DEBUG1("ctrl_close_pending = %d", ctrl_close_pending);
-            result = false;
+            // DEBUG1("ctrl_close_pending = %d", ctrl_close_pending);
+            close_received = true;

             break;
         case message::DATA_CLOSE:
             if (--data_close_pending == 1)
                 comm.next_close_on_data_is_last();
             break;
         case message::DATA_CLOSE:
             if (--data_close_pending == 1)
                 comm.next_close_on_data_is_last();

-            DEBUG1("data_close_pending = %d", data_close_pending);
-            result = false;
+            // DEBUG1("data_close_pending = %d", data_close_pending);
+            close_received = true;

             break;
         }
         delete msg;
             break;
         }
         delete msg;

-        do_recv = (wait == WAIT_FOR_CLOSE) &&
-            (ctrl_close_pending || data_close_pending);
+        may_receive = (ctrl_close_pending || data_close_pending);
+        do_wait = (wait == WAIT_FOR_CLOSE);

     }
     }

-    return result;

 }
 
 void process::finalize1(neighbor& nb)
 {
     comm.send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 }
 
 void process::finalize1(neighbor& nb)
 {
     comm.send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));

-    comm.send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));    
+    comm.send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));

 }
 
 void process::finalize()
 }
 
 void process::finalize()


@@ -270,6 +286,9 @@ void process::finalize()
     using namespace std::tr1;
     using namespace std::tr1::placeholders;
 
     using namespace std::tr1;
     using namespace std::tr1::placeholders;
 

+    finalizing = true;
+    total_load_running -= load;
+

     DEBUG2("send CLOSE to %d neighbor%s.",
            (int )neigh.size(), ESSE(neigh.size()));
     std::for_each(neigh.begin(), neigh.end(),
     DEBUG2("send CLOSE to %d neighbor%s.",
            (int )neigh.size(), ESSE(neigh.size()));
     std::for_each(neigh.begin(), neigh.end(),