X-Git-Url: https://bilbo.iut-bm.univ-fcomte.fr/and/gitweb/loba.git/blobdiff_plain/8bdfda4ed37966fcac00b37ddca740afaba48b28..ea9e469b934dd81706ec3f727156b9057e4cf2bb:/process.cpp

diff --git a/process.cpp b/process.cpp
index 99bd039..a0447de 100644
--- a/process.cpp
+++ b/process.cpp
@@ -11,6 +11,7 @@ XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
 
 #include "misc.h"
 #include "options.h"
+#include "tracing.h"
 
 #include "process.h"
 
@@ -33,7 +34,9 @@ process::process(int argc, char* argv[])
         rev_neigh.insert(std::make_pair(host, ptr));
     }
 
-    prev_load_broadcast = -1;   // force sending of load on first send()
+    comp = 0.0;
+
+    prev_load_broadcast = -1;   // force sending of load on first send_all()
     expected_load = load;
     total_load_running += load;
     total_load_init += load;
@@ -61,6 +64,17 @@ process::process(int argc, char* argv[])
 process::~process()
 {
     total_load_exit += load;
+    if (opt::bookkeeping) {
+        INFO4("Final load after %d:%d iterations: %g ; expected: %g",
+              lb_iter, comp_iter, load, expected_load);
+    } else {
+        INFO2("Final load after %d iterations: %g",
+              lb_iter, load);
+        if (lb_iter != comp_iter)
+            WARN2("lb_iter (%d) and comp_iter (%d) differ!",
+                  lb_iter, comp_iter);
+    }
+    VERB1("Total computation for this process: %g", comp);
 }
 
 int process::run()
@@ -70,46 +84,45 @@ int process::run()
     VERB0("Starting...");
     comp_iter = lb_iter = 0;
     while (true) {
-        if (opt::min_iter_duration) {
+        double ld = lb_load();
+        if (ld > 0.0) {
             double now = MSG_get_clock();
-            if (now < next_iter_after_date){
-                double delay = next_iter_after_date - now;
-                DEBUG1("sleeping for %g s", delay);
+            if (now < next_iter_after_date)
                 MSG_process_sleep(next_iter_after_date - now);
-            }
             next_iter_after_date = MSG_get_clock() + opt::min_iter_duration;
-        }
 
-        if (load > 0.0) {
-            ++comp_iter;
-            if (opt::log_rate && comp_iter % opt::log_rate == 0) {
+            ++lb_iter;
+
+            if (opt::log_rate && lb_iter % opt::log_rate == 0) {
                 if (opt::bookkeeping)
                     INFO4("(%u:%u) current load: %g ; expected: %g",
-                          comp_iter, lb_iter, load, expected_load);
+                          lb_iter, comp_iter, load, expected_load);
                 else
                     INFO2("(%u) current load: %g",
-                          comp_iter, load);
+                          lb_iter, load);
             }
 
-            if (opt::bookkeeping)
-                expected_load -= load_balance(expected_load);
-            else
-                load -= load_balance(load);
+            ld -= load_balance(ld);
 
             print_loads(true, xbt_log_priority_debug);
+        }
+        lb_load() = ld;
 
-            send();
+        // send load information, and load (data) if any
+        send_all();
+        if (load > 0.0) {
+            ++comp_iter;
             compute();
-
-        } else {
-            // send load information, and load when bookkeeping
-            send();
         }
 
-        if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter)
+        if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
+            VERB2("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
             break;
-        if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter)
+        }
+        if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
+            VERB2("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
             break;
+        }
         if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
             VERB2("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
             break;
@@ -117,28 +130,37 @@ int process::run()
 
         // 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);
+        double timeout;
+        if (load != 0 || lb_load() != prev_load_broadcast)
+            timeout = 0.0;
+        else if (opt::min_iter_duration)
+            timeout = opt::min_iter_duration;
+        else
+            timeout = 1.0;
+        receive(timeout);
 
         // one of our neighbor is finalizing
-        if (opt::exit_on_close && close_received)
+        if (opt::exit_on_close && close_received) {
+            VERB0("Close received");
             break;
+        }
 
         // have no load and cannot receive anything
-        if (load == 0.0 && !may_receive())
+        if (load == 0.0 && !may_receive()) {
+            VERB0("I'm a poor lonesome process, and I have no load...");
             break;
+        }
 
         // 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.");
+            VERB0("No more load to balance in system.");
             break;
+        } else {
+            DEBUG1("still %g load to balance, continuing...", total_load_running);
         }
-
     }
     VERB0("Going to finalize...");
     finalize();
@@ -150,10 +172,6 @@ int process::run()
      */
 
     VERB0("Done.");
-    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;
 }
 
@@ -178,10 +196,12 @@ double process::load_balance(double /*my_load*/)
 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);
-        DEBUG2("compute %g flop%s", duration, ESSE(duration));
+        double flops = opt::comp_cost(load);
+        m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
+        TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
+        DEBUG2("compute %g flop%s", flops, ESSE(flops));
         MSG_task_execute(task);
+        comp += flops;
         MSG_task_destroy(task);
     } else {
         DEBUG0("nothing to compute !");
@@ -228,7 +248,7 @@ void process::send1_bookkeeping(neighbor& nb)
         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 }
 
-void process::send()
+void process::send_all()
 {
     using std::tr1::bind;
     using std::tr1::placeholders::_1;
@@ -245,12 +265,13 @@ void process::send()
     comm.flush(false);
 }
 
-void process::receive(bool wait)
+void process::receive(double timeout)
 {
     message* msg;
     m_host_t from;
 
-    while (may_receive() && comm.recv(msg, from, wait)) {
+    DEBUG2("%sblocking receive (%g)", "\0non-" + !timeout, timeout);
+    while (may_receive() && comm.recv(msg, from, timeout)) {
         switch (msg->get_type()) {
         case message::INFO: {
             neighbor* n = rev_neigh[from];
@@ -277,7 +298,7 @@ void process::receive(bool wait)
             break;
         }
         delete msg;
-        wait = false;           // only wait on first recv
+        timeout = 0.0;          // only wait on first recv
     }
     comm.flush(false);
 }
@@ -305,7 +326,7 @@ void process::finalize()
            (unsigned long )neigh.size(), ESSE(neigh.size()));
     while (may_receive()) {
         comm.flush(false);
-        receive(true);
+        receive(-1.0);
     }
 
     comm.flush(true);