Happy new year.

[loba.git] / process.cpp

diff --git a/process.cpp b/process.cpp
index c85b6051a6a4224452f22ddce086740f310ed61e..811804fa81558253c282b3975751673e4d98f1da 100644 (file)
--- a/process.cpp
+++ b/process.cpp
@@ -1,12 +1,10 @@
 #include <algorithm>
 #include <cmath>
 #include <algorithm>
 #include <cmath>

-#include <functional>

 #include <iterator>
 #include <numeric>
 #include <stdexcept>
 #include <sstream>
 #include <xbt/log.h>
 #include <iterator>
 #include <numeric>
 #include <stdexcept>
 #include <sstream>
 #include <xbt/log.h>

-#include <xbt/time.h>

 
 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
 
 
 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
 


@@ -16,14 +14,19 @@ XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
 
 #include "process.h"
 
 
 #include "process.h"
 

+mutex_t *process::proc_mutex;
+

 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::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::average_load_ratio;

 double process::load_diff_threshold;
 
 double process::load_diff_threshold;
 

+std::atomic<int> process::convergence_counter(0);
+

 namespace {
 
     void sleep_until_date(double& date, double duration)
 namespace {
 
     void sleep_until_date(double& date, double duration)


@@ -53,27 +56,32 @@ process::process(int argc, char* argv[])
     pneigh.reserve(neigh.size());
     for (unsigned i = 0 ; i < neigh.size() ; i++) {
         neighbor* ptr = &neigh[i];
     pneigh.reserve(neigh.size());
     for (unsigned i = 0 ; i < neigh.size() ; i++) {
         neighbor* ptr = &neigh[i];

-        m_host_t host = MSG_get_host_by_name(ptr->get_name());
+        msg_host_t host = MSG_get_host_by_name(ptr->get_name());

         pneigh.push_back(ptr);
         rev_neigh.insert(std::make_pair(host, ptr));
     }
 
         pneigh.push_back(ptr);
         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;
     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;
 
     idle_duration = 0.0;
     convergence = -1.0;
 
     received_load = 0.0;
 
     idle_duration = 0.0;
     convergence = -1.0;
 

+    proc_mutex->acquire();

     process_counter++;
     process_counter++;

+    convergence_counter++;
+    total_load_init += real_load;
+    total_load_running += real_load;

     total_load_average = total_load_running / process_counter;
     total_load_average = total_load_running / process_counter;

+    if (opt::avg_load_ratio >= 0.0)
+        average_load_ratio = opt::avg_load_ratio;
+    else
+        average_load_ratio = 100.0 *
+            (process_counter / -opt::avg_load_ratio) / total_load_average;

     load_diff_threshold = (opt::load_ratio_threshold +
     load_diff_threshold = (opt::load_ratio_threshold +

-                           opt::avg_load_ratio * total_load_average) / 100.0;
+                           average_load_ratio * total_load_average) / 100.0;
+    proc_mutex->release();

 
     ctrl_close_pending = data_close_pending = neigh.size();
     close_received = false;
 
     ctrl_close_pending = data_close_pending = neigh.size();
     close_received = false;


@@ -81,8 +89,7 @@ process::process(int argc, char* argv[])
 
     all_comp_iter = 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));
+    lb_thread = new_msg_thread("loba", [this]() { this->load_balance_loop(); });

 
     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))


@@ -93,7 +100,7 @@ process::process(int argc, char* argv[])
         oss << ESSE(neigh.size()) << ": ";
         std::transform(neigh.begin(), neigh.end() - 1,
                        std::ostream_iterator<const char*>(oss, ", "),
         oss << ESSE(neigh.size()) << ": ";
         std::transform(neigh.begin(), neigh.end() - 1,
                        std::ostream_iterator<const char*>(oss, ", "),

-                       std::mem_fn(&neighbor::get_name));
+                       [](const neighbor& neigh) { return neigh.get_name(); });

         oss << neigh.back().get_name();
     }
     XBT_LOG(logp, "Got %s.", oss.str().c_str());
         oss << neigh.back().get_name();
     }
     XBT_LOG(logp, "Got %s.", oss.str().c_str());


@@ -103,7 +110,9 @@ process::process(int argc, char* argv[])
 process::~process()
 {
     delete lb_thread;
 process::~process()
 {
     delete lb_thread;

+    proc_mutex->acquire();

     total_load_exit += real_load;
     total_load_exit += real_load;

+    proc_mutex->release();

     xbt_assert(received_load == 0.0,
                "received_load is %g, but should be 0.0 !", received_load);
     if (opt::log_rate < 0)
     xbt_assert(received_load == 0.0,
                "received_load is %g, but should be 0.0 !", received_load);
     if (opt::log_rate < 0)


@@ -112,10 +121,10 @@ process::~process()
              lb_iter, comp_iter, all_comp_iter, real_load);
     if (convergence >= 0.0)
         XBT_INFO("Convergence within %g%% was achieved at time %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);
+                 average_load_ratio, convergence);

     else
         XBT_INFO("Convergence within %g%% was not achieved",
     else
         XBT_INFO("Convergence within %g%% was not achieved",

-                 opt::avg_load_ratio);
+                 average_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);
     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);


@@ -123,7 +132,7 @@ process::~process()
 
 double process::get_iter_deviation() const
 {
 
 double process::get_iter_deviation() const
 {

-    double average_cost = opt::comp_cost(total_load_average);
+    double average_cost = opt::comp_cost(total_load_average); // fixme: get locked?

     // Do not count idle periods
     double comp_iter_opt = acc.comp_amount / average_cost;
 /*
     // Do not count idle periods
     double comp_iter_opt = acc.comp_amount / average_cost;
 /*


@@ -159,8 +168,6 @@ int process::run()
 
 void process::load_balance_loop()
 {
 
 void process::load_balance_loop()
 {

-    using std::placeholders::_1;
-

     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
     while (still_running()) {
         if (lb_iter == opt::comp_iter_delay) {
     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
     while (still_running()) {
         if (lb_iter == opt::comp_iter_delay) {


@@ -192,8 +199,8 @@ void process::load_balance_loop()
 
         // send
         comm.ctrl_flush(false);
 
         // send
         comm.ctrl_flush(false);

-        std::for_each(neigh.begin(), neigh.end(),
-                      std::bind(&process::ctrl_send, this, _1));
+        for (neighbor& n : neigh)
+            ctrl_send(n);

         prev_load_broadcast = expected_load;
         mutex.release();
 
         prev_load_broadcast = expected_load;
         mutex.release();
 


@@ -203,8 +210,8 @@ void process::load_balance_loop()
     XBT_VERB("Going to finalize for %s...", __func__);
     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
               neigh.size(), ESSE(neigh.size()));
     XBT_VERB("Going to finalize for %s...", __func__);
     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
               neigh.size(), ESSE(neigh.size()));

-    std::for_each(neigh.begin(), neigh.end(),
-                  std::bind(&process::ctrl_close, this, _1));
+    for (neighbor& n : neigh)
+        ctrl_close(n);

     while (ctrl_close_pending) {
         comm.ctrl_flush(false);
         XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);
     while (ctrl_close_pending) {
         comm.ctrl_flush(false);
         XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);


@@ -215,8 +222,6 @@ void process::load_balance_loop()
 
 void process::compute_loop()
 {
 
 void process::compute_loop()
 {

-    using std::placeholders::_1;
-

     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
     double idle_since_date = 0.0;
     while (still_running()) {
     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
     double idle_since_date = 0.0;
     while (still_running()) {


@@ -230,8 +235,8 @@ void process::compute_loop()
         mutex.acquire();
         real_load += received_load;
         received_load = 0.0;
         mutex.acquire();
         real_load += received_load;
         received_load = 0.0;

-        std::for_each(neigh.begin(), neigh.end(),
-                      std::bind(&process::data_send, this, _1));
+        for (neighbor& n : neigh)
+            data_send(n);

         mutex.release();
 
         ++all_comp_iter;
         mutex.release();
 
         ++all_comp_iter;


@@ -244,8 +249,8 @@ void process::compute_loop()
         idle_duration += MSG_get_clock() - idle_since_date;
         ++comp_iter;
         double flops = opt::comp_cost(real_load);
         idle_duration += MSG_get_clock() - idle_since_date;
         ++comp_iter;
         double flops = opt::comp_cost(real_load);

-        m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
-        TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
+        msg_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
+        // MSG_task_set_category(task, TRACE_CAT_COMP);

         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
         MSG_task_execute(task);
         add_comp_amount(flops);
         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
         MSG_task_execute(task);
         add_comp_amount(flops);


@@ -261,8 +266,8 @@ void process::compute_loop()
     finalizing = true;
     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
               neigh.size(), ESSE(neigh.size()));
     finalizing = true;
     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
               neigh.size(), ESSE(neigh.size()));

-    std::for_each(neigh.begin(), neigh.end(),
-                  std::bind(&process::data_close, this, _1));
+    for (neighbor& n : neigh)
+        data_close(n);

     while (data_close_pending) {
         comm.data_flush(false);
         XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);
     while (data_close_pending) {
         comm.data_flush(false);
         XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);


@@ -270,27 +275,35 @@ void process::compute_loop()
     }
     real_load += received_load;
     received_load = 0.0;
     }
     real_load += received_load;
     received_load = 0.0;

+    proc_mutex->acquire();

     total_load_running -= real_load;
     total_load_running -= real_load;

+    proc_mutex->release();

     convergence_check();
     comm.data_flush(true);
 }
 
 void process::convergence_check()
 {
     convergence_check();
     comm.data_flush(true);
 }
 
 void process::convergence_check()
 {

-    double load_diff = std::fabs(real_load - total_load_average);
+    double average = total_load_average; // fixme: get locked?
+    double load_diff = std::fabs(real_load - average);

     bool converged = load_diff <= load_diff_threshold;
 
     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) {
+    if (converged) {
+        if (convergence < 0) {

             XBT_VERB("current load has converged: %g (%.4g%%)",
             XBT_VERB("current load has converged: %g (%.4g%%)",

-                     real_load,  100.0 * load_diff / total_load_average);
+                     real_load,  100.0 * load_diff / average);

             convergence = MSG_get_clock();
             convergence = MSG_get_clock();

+            local_convergence_counter = opt::exit_on_convergence;
+        }
+        if (local_convergence_counter > 0 && --local_convergence_counter == 0)
+                --convergence_counter;
+    } else {
+        if (convergence >= 0.0) {
+            XBT_VERB("current load has diverged: %g (%.4g%%)",
+                     real_load, 100.0 * load_diff / average);
+            convergence = -1.0;
+            if (local_convergence_counter == 0)
+                ++convergence_counter;

         }
     }
 }
         }
     }
 }


@@ -318,6 +331,10 @@ bool process::still_running()
         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
         last_status = false;
 
         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
         last_status = false;
 

+    } else if (opt::exit_on_convergence && convergence_counter == 0) {
+        XBT_VERB("Global convergence detected");
+        last_status = false;
+

     } else if (opt::exit_on_close && close_received) {
         XBT_VERB("Close received");
         last_status = false;
     } else if (opt::exit_on_close && close_received) {
         XBT_VERB("Close received");
         last_status = false;


@@ -327,7 +344,7 @@ bool process::still_running()
         last_status = false;
 
     } else if (100.0 * total_load_running / total_load_init <=
         last_status = false;
 
     } else if (100.0 * total_load_running / total_load_init <=

-               opt::load_ratio_threshold) {
+               opt::load_ratio_threshold) { // fixme: get locked?

         // fixme: this check should be implemented with a distributed
         // algorithm, and not a shared global variable!
         XBT_VERB("No more load to balance in system.");
         // fixme: this check should be implemented with a distributed
         // algorithm, and not a shared global variable!
         XBT_VERB("No more load to balance in system.");


@@ -339,12 +356,10 @@ bool process::still_running()
 
 double process::get_sum_of_to_send() const
 {
 
 double process::get_sum_of_to_send() const
 {

-    using std::placeholders::_1;
-    using std::placeholders::_2;
-

     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
     return std::accumulate(neigh.begin(), neigh.end(), 0.0,

-                           std::bind(std::plus<double>(), _1,
-                                     std::bind(&neighbor::get_to_send, _2)));
+                           [](double x, const neighbor& neigh) {
+                               return x + neigh.get_to_send();
+                           });

 }
 
 void process::load_balance()
 }
 
 void process::load_balance()


@@ -397,7 +412,11 @@ void process::data_send(neighbor& nb)
         else
             excess_load = real_load;
 
         else
             excess_load = real_load;
 

-        double balance = nb.get_debt() - nb.get_credit();
+        double balance;
+        if (nb.get_credit() > 0.0)
+            balance = nb.get_debt() - nb.get_credit();
+        else
+            balance = nb.get_debt();

         load_to_send = std::min(excess_load,
                                 std::max(0.0, balance));
 
         load_to_send = std::min(excess_load,
                                 std::max(0.0, balance));
 


@@ -437,7 +456,7 @@ void process::data_close(neighbor& nb)
 void process::ctrl_receive(double timeout)
 {
     message* msg;
 void process::ctrl_receive(double timeout)
 {
     message* msg;

-    m_host_t from;
+    msg_host_t from;

 
     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 
     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {


@@ -451,7 +470,7 @@ void process::ctrl_receive(double timeout)
 void process::data_receive(double timeout)
 {
     message* msg;
 void process::data_receive(double timeout)
 {
     message* msg;

-    m_host_t from;
+    msg_host_t from;

 
     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 
     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
     while (data_close_pending && comm.data_recv(msg, from, timeout)) {


@@ -462,7 +481,7 @@ void process::data_receive(double timeout)
     }
 }
 
     }
 }
 

-void process::handle_message(message* msg, m_host_t from)
+void process::handle_message(message* msg, msg_host_t from)

 {
     switch (msg->get_type()) {
     case message::CTRL: {
 {
     switch (msg->get_type()) {
     case message::CTRL: {


@@ -494,27 +513,28 @@ void process::handle_message(message* msg, m_host_t from)
     delete msg;
 }
 
     delete msg;
 }
 

-#define print_loads_generic(vec, verbose, logp, cat)                    \
-    if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
-        using std::placeholders::_1;                                    \
-        XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
-                  "Neighbor loads:", real_load, expected_load);         \
-        std::for_each(vec.begin(), vec.end(),                           \
-                      std::bind(&neighbor::print, _1, verbose, logp, cat)); \
-    } else ((void)0)
-

 void process::print_loads(bool verbose,
                           e_xbt_log_priority_t logp,
                           xbt_log_category_t cat) const
 {
 void process::print_loads(bool verbose,
                           e_xbt_log_priority_t logp,
                           xbt_log_category_t cat) const
 {

-    print_loads_generic(neigh, verbose, logp, cat);
+    if (!_XBT_LOG_ISENABLEDV((*cat), logp))
+        return;
+    XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  Neighbor loads:",
+              real_load, expected_load);
+    for (const neighbor& n : neigh)
+        n.print(verbose, logp, cat);

 }
 
 void process::print_loads_p(bool verbose,
                             e_xbt_log_priority_t logp,
                             xbt_log_category_t cat) const
 {
 }
 
 void process::print_loads_p(bool verbose,
                             e_xbt_log_priority_t logp,
                             xbt_log_category_t cat) const
 {

-    print_loads_generic(pneigh, verbose, logp, cat);
+    if (!_XBT_LOG_ISENABLEDV((*cat), logp))
+        return;
+    XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  Neighbor loads:",
+              real_load, expected_load);
+    for (const neighbor* n : pneigh)
+        n->print(verbose, logp, cat);

 }
 
 #undef print_loads_generic
 }
 
 #undef print_loads_generic