Bug fix: we really need to consume all messages in proc::receive!

[loba.git] / process.cpp

diff --git a/process.cpp b/process.cpp
index 214b9d05a785d80edeb436f14466a6d8140545d9..53cbb59c7434da4a84beb5d2971e8c98ea7ae910 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[])


@@ -32,16 +33,14 @@ 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();

-    if (neigh.size() == 1) {
-        comm.next_close_on_ctrl_is_last();
-        comm.next_close_on_data_is_last();
-    }
-    if (neigh.size() > 0)
-        comm.listen();
+    close_received = false;
+    finalizing = false;

 
     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
     if (!LOG_ISENABLED(logp))
 
     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
     if (!LOG_ISENABLED(logp))


@@ -56,7 +55,7 @@ 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 +67,64 @@ 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;
-    bool one_more = true;
-    do {
-        ++iter;
-
-        if (opt::log_rate && iter % opt::log_rate == 0) {
+    comp_iter = lb_iter = 0;
+    while (true) {
+        if (load > 0.0) {
+            ++comp_iter;
+            if (opt::log_rate && comp_iter % opt::log_rate == 0) {
+                if (opt::bookkeeping)
+                    INFO4("(%u:%u) current load: %g ; expected: %g",
+                          comp_iter, lb_iter, load, expected_load);
+                else
+                    INFO2("(%u) current load: %g",
+                          comp_iter, load);
+            }
+

             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.
+        } else {
+            // send load information, and load when bookkeeping
+            send();
+        }

 
 

-            // fixme: review this chunk, and remove this NDS macro!
+        if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter)
+            break;
+        if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter)
+            break;

 
 

-            bool recv_wait = (load == 0 && !NDS);
-            bool close_received = !receive(recv_wait? WAIT: NO_WAIT);
+        // block on receiving unless there is something to compute or
+        // to send
+        bool wait = (load == 0 &&
+                     ((opt::bookkeeping ? expected_load : load)
+                      == prev_load_broadcast));
+        receive(wait);

 
 

-            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);
+        // one of our neighbor is finalizing
+        if (opt::exit_on_close && close_received)
+            break;

 
 

-        } while (one_more && can_recv && load == 0 && !NDS);
-#undef NDS
+        // have no load and cannot receive anything
+        if (load == 0.0 && !may_receive())
+            break;

 
 

-    } while (one_more);
+        // 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() ?
+        if (100.0 * total_load_running / total_load_init <=
+            opt::load_ratio_threshold) {
+            VERB0("No more load to balance in system, stopping.");
+            break;
+        }
+
+    }

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


@@ -126,7 +135,8 @@ int process::run()
      */
 
     VERB0("Done.");
      */
 
     VERB0("Done.");

-    INFO3("Final load after %d iteration%s: %g", iter, ESSE(iter), load);
+    INFO3("Final load after %d iteration%s: %g",
+          comp_iter, ESSE(comp_iter), load);

     if (opt::bookkeeping)
         INFO1("Expected load: %g", expected_load);
     return 0;
     if (opt::bookkeeping)
         INFO1("Expected load: %g", expected_load);
     return 0;


@@ -134,8 +144,9 @@ int process::run()
 
 double process::sum_of_to_send() const
 {
 
 double process::sum_of_to_send() const
 {

-    using namespace std::tr1;
-    using namespace std::tr1::placeholders;
+    using std::tr1::bind;
+    using std::tr1::placeholders::_1;
+    using std::tr1::placeholders::_2;

 
     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
                            bind(std::plus<double>(),
 
     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
                            bind(std::plus<double>(),


@@ -144,7 +155,7 @@ double process::sum_of_to_send() const
 
 double process::load_balance(double /*my_load*/)
 {
 
 double process::load_balance(double /*my_load*/)
 {

-    if (iter == 1)
+    if (lb_iter == 1)           // warn only once

         WARN0("process::load_balance is a no-op!");
     return 0.0;
 }
         WARN0("process::load_balance is a no-op!");
     return 0.0;
 }


@@ -154,7 +165,7 @@ void process::compute()
     if (load > 0.0) {
         double duration = opt::comp_cost(load);
         m_task_t task = MSG_task_create("computation", duration, 0.0, NULL);
     if (load > 0.0) {
         double duration = opt::comp_cost(load);
         m_task_t task = MSG_task_create("computation", duration, 0.0, NULL);

-        DEBUG2("compute %g flop%s.", duration, ESSE(duration));
+        DEBUG2("compute %g flop%s", duration, ESSE(duration));

         MSG_task_execute(task);
         MSG_task_destroy(task);
     } else {
         MSG_task_execute(task);
         MSG_task_destroy(task);
     } else {


@@ -204,8 +215,8 @@ void process::send1_bookkeeping(neighbor& nb)
 
 void process::send()
 {
 
 void process::send()
 {

-    using namespace std::tr1;
-    using namespace std::tr1::placeholders;
+    using std::tr1::bind;
+    using std::tr1::placeholders::_1;

 
     if (opt::bookkeeping) {
         std::for_each(neigh.begin(), neigh.end(),
 
     if (opt::bookkeeping) {
         std::for_each(neigh.begin(), neigh.end(),


@@ -218,15 +229,12 @@ void process::send()
     }
 }
 
     }
 }
 

-bool process::receive(recv_wait_mode wait)
+void process::receive(bool 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)) {
+
+    while (may_receive() && comm.recv(msg, from, 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];


@@ -236,56 +244,58 @@ 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:
         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;
+            ctrl_close_pending--;
+            close_received = true;

             break;
         case message::DATA_CLOSE:
             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;
+            data_close_pending--;
+            close_received = true;

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

-        do_recv = (wait == WAIT_FOR_CLOSE) &&
-            (ctrl_close_pending || data_close_pending);
+        wait = false;           // only wait on first recv

     }
     }

-    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()
 {

-    using namespace std::tr1;
-    using namespace std::tr1::placeholders;
+    using std::tr1::bind;
+    using std::tr1::placeholders::_1;
+
+    finalizing = true;
+    total_load_running -= load;

 
 

-    DEBUG2("send CLOSE to %d neighbor%s.",
-           (int )neigh.size(), ESSE(neigh.size()));
+    DEBUG2("send CLOSE to %lu neighbor%s",
+           (unsigned long )neigh.size(), ESSE(neigh.size()));

     std::for_each(neigh.begin(), neigh.end(),
                   bind(&process::finalize1, this, _1));
 
     std::for_each(neigh.begin(), neigh.end(),
                   bind(&process::finalize1, this, _1));
 

-    DEBUG2("wait for CLOSE from %d neighbor%s.",
-           (int )neigh.size(), ESSE(neigh.size()));
-    receive(WAIT_FOR_CLOSE);
+    DEBUG2("wait for CLOSE from %lu neighbor%s",
+           (unsigned long )neigh.size(), ESSE(neigh.size()));
+    while (may_receive())
+        receive(true);

 
     comm.flush(true);
 }
 
 #define print_loads_generic(vec, verbose, logp, cat)                    \
     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 
     comm.flush(true);
 }
 
 #define print_loads_generic(vec, verbose, logp, cat)                    \
     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \

-        using namespace std::tr1;                                       \
-        using namespace std::tr1::placeholders;                         \
+        using std::tr1::bind;                                           \
+        using std::tr1::placeholders::_1;                               \

         XCLOG0(cat, logp, "Neighbor loads:");                           \
         std::for_each(vec.begin(), vec.end(),                           \
                       bind(&neighbor::print, _1, verbose, logp, cat));  \
         XCLOG0(cat, logp, "Neighbor loads:");                           \
         std::for_each(vec.begin(), vec.end(),                           \
                       bind(&neighbor::print, _1, verbose, logp, cat));  \