Add statistics about idle duration.

[loba.git] / process.cpp

diff --git a/process.cpp b/process.cpp
index d0bfc21120b8cbc8ea0230b4eb9bb97e4eeb26f3..eac87d1d3de4cff4bbf9206499ab33fd9f92cc30 100644 (file)
--- a/process.cpp
+++ b/process.cpp
@@ -1,5 +1,6 @@
 #include <algorithm>
 #include <algorithm>

-#include <tr1/functional>
+#include <cmath>
+#include <functional>

 #include <iterator>
 #include <numeric>
 #include <stdexcept>
 #include <iterator>
 #include <numeric>
 #include <stdexcept>


@@ -19,9 +20,13 @@ double process::total_load_init = 0.0;
 double process::total_load_running = 0.0;
 double process::total_load_exit = 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 {
 
 namespace {
 

-    void sleep_until_date(double& date, double duration = 0.0)
+    void sleep_until_date(double& date, double duration)

     {
         double sleep_duration = date - MSG_get_clock();
         if (sleep_duration > 0.0)
     {
         double sleep_duration = date - MSG_get_clock();
         if (sleep_duration > 0.0)


@@ -36,6 +41,13 @@ process::process(int argc, char* argv[])
     if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))
         throw std::invalid_argument("bad or missing initial load parameter");
 
     if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))
         throw std::invalid_argument("bad or missing initial load parameter");
 

+    double iload = std::trunc(real_load);
+    if (opt::integer_transfer && real_load != iload) {
+        XBT_WARN("Initial load %g is not an integer.  Truncate it.",
+                 real_load);
+        real_load = iload;
+    }
+

     neigh.assign(argv + 2, argv + argc);
 
     pneigh.reserve(neigh.size());
     neigh.assign(argv + 2, argv + argc);
 
     pneigh.reserve(neigh.size());


@@ -46,20 +58,31 @@ process::process(int argc, char* argv[])
         rev_neigh.insert(std::make_pair(host, 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;
     total_load_running += real_load;
     total_load_init += 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;
+
+    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;
 
 
     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",
 
     lb_thread = new_msg_thread("loba",

-                               std::tr1::bind(&process::load_balance_loop,
-                                              this));
+                               std::bind(&process::load_balance_loop, this));

 
     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))


@@ -70,7 +93,7 @@ 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::tr1::mem_fn(&neighbor::get_name));
+                       std::mem_fn(&neighbor::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());


@@ -81,12 +104,21 @@ process::~process()
 {
     delete lb_thread;
     total_load_exit += real_load;
 {
     delete lb_thread;
     total_load_exit += real_load;

+    xbt_assert(received_load == 0.0,
+               "received_load is %g, but should be 0.0 !", received_load);

     if (opt::log_rate < 0)
         return;
     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());
     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);

 }
 
 int process::run()
 }
 
 int process::run()


@@ -112,8 +144,7 @@ int process::run()
 
 void process::load_balance_loop()
 {
 
 void process::load_balance_loop()
 {

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

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


@@ -126,12 +157,16 @@ void process::load_balance_loop()
             ++lb_iter;
         }
 
             ++lb_iter;
         }
 

+        ctrl_receive(0.0);
+
+        mutex.acquire();

         if (!opt::bookkeeping)
             expected_load = real_load - get_sum_of_to_send();
         // nothing to do with opt::bookkeeping
 
         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
         if (!opt::bookkeeping)
             expected_load = real_load - get_sum_of_to_send();
         // 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);
         }
 
             XBT_VERB("... expected load: %g", expected_load);
         }
 


@@ -141,24 +176,23 @@ void process::load_balance_loop()
         print_loads(true, xbt_log_priority_debug);
 
         // send
         print_loads(true, xbt_log_priority_debug);
 
         // send

+        comm.ctrl_flush(false);

         std::for_each(neigh.begin(), neigh.end(),
         std::for_each(neigh.begin(), neigh.end(),

-                      bind(&process::ctrl_send, this, _1));
+                      std::bind(&process::ctrl_send, this, _1));

         prev_load_broadcast = expected_load;
         prev_load_broadcast = expected_load;

+        mutex.release();

 
         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 
         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);

-        ctrl_receive(0.0);
-
-        comm.ctrl_flush(false);

     }
 
     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(),
     }
 
     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(),

-                  bind(&process::ctrl_close, this, _1));
+                  std::bind(&process::ctrl_close, this, _1));

     while (ctrl_close_pending) {
         comm.ctrl_flush(false);
     while (ctrl_close_pending) {
         comm.ctrl_flush(false);

-        XBT_DEBUG("waiting for %d CTRL CLOSE", ctrl_close_pending);
+        XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);

         ctrl_receive(-1.0);
     }
     comm.ctrl_flush(true);
         ctrl_receive(-1.0);
     }
     comm.ctrl_flush(true);


@@ -166,27 +200,35 @@ void process::load_balance_loop()
 
 void process::compute_loop()
 {
 
 void process::compute_loop()
 {

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

 
     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 
     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;

+    double idle_since_date = 0.0;

     while (still_running()) {
     while (still_running()) {

-        // receive
-        if (real_load > 0.0)
-            data_receive(0.0);
-        else
-            data_receive(opt::min_comp_iter_duration);

 
 

-        comm.data_flush(false);
+        do {
+            // receive
+            // if there is something to compute, do not block
+            // else, block the duration of an *lb* iteration
+            data_receive(real_load > 0.0 ? 0.0 : opt::min_lb_iter_duration);

 
 

-        if (real_load == 0.0)
-            continue;
+            // send
+            comm.data_flush(false);
+            mutex.acquire();
+            real_load += received_load;
+            received_load = 0.0;
+            std::for_each(neigh.begin(), neigh.end(),
+                          std::bind(&process::data_send, this, _1));
+            mutex.release();

 
 

-        // send
-        std::for_each(neigh.begin(), neigh.end(),
-                      bind(&process::data_send, this, _1));
+            ++all_comp_iter;
+
+        } while (real_load == 0.0);
+
+        convergence_check();

 
         // compute
 
         // compute

+        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);
         ++comp_iter;
         double flops = opt::comp_cost(real_load);
         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);


@@ -196,27 +238,49 @@ void process::compute_loop()
         add_comp_amount(flops);
         MSG_task_destroy(task);
 
         add_comp_amount(flops);
         MSG_task_destroy(task);
 

+        idle_since_date = MSG_get_clock();
+

         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
     }
 
     XBT_VERB("Going to finalize for %s...", __func__);
         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
     }
 
     XBT_VERB("Going to finalize for %s...", __func__);

-    // last send, for not losing load scheduled to be sent
-    std::for_each(neigh.begin(), neigh.end(),
-                  bind(&process::data_send, this, _1));

     finalizing = true;
     finalizing = true;

-    total_load_running -= real_load;

     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
               neigh.size(), ESSE(neigh.size()));
     std::for_each(neigh.begin(), neigh.end(),
     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
               neigh.size(), ESSE(neigh.size()));
     std::for_each(neigh.begin(), neigh.end(),

-                  bind(&process::data_close, this, _1));
+                  std::bind(&process::data_close, this, _1));

     while (data_close_pending) {
         comm.data_flush(false);
     while (data_close_pending) {
         comm.data_flush(false);

-        XBT_DEBUG("waiting for %d DATA CLOSE", data_close_pending);
+        XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);

         data_receive(-1.0);
     }
         data_receive(-1.0);
     }

+    real_load += received_load;
+    received_load = 0.0;
+    total_load_running -= real_load;
+    convergence_check();

     comm.data_flush(true);
 }
 
     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;
 bool process::still_running()
 {
     static bool last_status = true;


@@ -224,6 +288,10 @@ bool process::still_running()
     if (!last_status) {
         /* nop */
 
     if (!last_status) {
         /* nop */
 

+    } else if (opt::exit_request) {
+        XBT_VERB("Global exit requested");
+        last_status = false;
+

     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
         last_status = false;
     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
         last_status = false;


@@ -257,13 +325,12 @@ bool process::still_running()
 
 double process::get_sum_of_to_send() const
 {
 
 double process::get_sum_of_to_send() const
 {

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

-                           bind(std::plus<double>(),
-                                _1, bind(&neighbor::get_to_send, _2)));
+                           std::bind(std::plus<double>(), _1,
+                                     std::bind(&neighbor::get_to_send, _2)));

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


@@ -282,50 +349,61 @@ void process::send(neighbor& nb, double amount)
 void process::ctrl_send(neighbor& nb)
 {
     double info_to_send = expected_load;
 void process::ctrl_send(neighbor& nb)
 {
     double info_to_send = expected_load;

-    if (info_to_send != prev_load_broadcast) {
-        message* msg = new message(message::INFO, info_to_send);
-        add_ctrl_send_mesg(msg->get_size());
-        comm.ctrl_send(nb.get_ctrl_mbox(), msg);
-    }
-    if (opt::bookkeeping) {
-        double debt_to_send = nb.get_to_send();
+    double debt_to_send;
+    if (opt::bookkeeping) {     // bookkeeping
+        debt_to_send = nb.get_to_send();

         if (debt_to_send > 0.0) {
             nb.set_to_send(0.0);
             nb.set_debt(nb.get_debt() + debt_to_send);
         if (debt_to_send > 0.0) {
             nb.set_to_send(0.0);
             nb.set_debt(nb.get_debt() + debt_to_send);

-            message* msg = new message(message::CREDIT, debt_to_send);
-            add_ctrl_send_mesg(msg->get_size());
-            comm.ctrl_send(nb.get_ctrl_mbox(), msg);

         }
         }

+    } else {                    // !bookkeeping
+        debt_to_send = 0.0;
+    }
+    if (info_to_send != prev_load_broadcast || debt_to_send > 0.0) {
+        message* msg = new message(message::CTRL, info_to_send, debt_to_send);
+        add_ctrl_send_mesg(msg->get_size());
+        comm.ctrl_send(nb.get_ctrl_mbox(), msg);

     }
 }
 
     }
 }
 

+double process::compute_load_to_send(double desired)
+{
+    if (opt::integer_transfer)
+        desired = std::floor(desired);
+    return desired >= opt::min_transfer_amount ? desired : 0.0;
+}
+

 void process::data_send(neighbor& nb)
 {
     double load_to_send;
 void process::data_send(neighbor& nb)
 {
     double load_to_send;

-    if (opt::bookkeeping) {
-        load_to_send = std::min(real_load, nb.get_debt());
-        if (load_to_send >= opt::min_transfer_amount) {
+    if (opt::bookkeeping) {     // bookkeeping
+        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;
+
+        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);
             nb.set_debt(nb.get_debt() - load_to_send);

-            real_load -= load_to_send;
-        } else {
-            load_to_send = 0.0;
-        }
-    } else {
-        load_to_send = nb.get_to_send();
-        if (load_to_send >= opt::min_transfer_amount) {
-            nb.set_to_send(0.0);
-            real_load -= load_to_send;
-        } else {
-            load_to_send = 0.0;
-        }
+    } else {                    // !bookkeeping
+        load_to_send = compute_load_to_send(nb.get_to_send());
+        if (load_to_send > 0.0)
+            nb.set_to_send(nb.get_to_send() - load_to_send);

     }
     }

+    real_load -= load_to_send;

     while (load_to_send > 0.0) {
         double amount;
         if (opt::max_transfer_amount)
             amount = std::min(load_to_send, opt::max_transfer_amount);
         else
             amount = load_to_send;
     while (load_to_send > 0.0) {
         double amount;
         if (opt::max_transfer_amount)
             amount = std::min(load_to_send, opt::max_transfer_amount);
         else
             amount = load_to_send;

-        message* msg = new message(message::LOAD, amount);
+        message* msg = new message(message::DATA, amount);

         add_data_send_mesg(msg->get_size());
         comm.data_send(nb.get_data_mbox(), msg);
         load_to_send -= amount;
         add_data_send_mesg(msg->get_size());
         comm.data_send(nb.get_data_mbox(), msg);
         load_to_send -= amount;


@@ -373,19 +451,21 @@ void process::data_receive(double timeout)
 void process::handle_message(message* msg, m_host_t from)
 {
     switch (msg->get_type()) {
 void process::handle_message(message* msg, m_host_t from)
 {
     switch (msg->get_type()) {

-    case message::INFO: {
+    case message::CTRL: {

         neighbor* n = rev_neigh[from];
         n->set_load(msg->get_amount() + n->get_to_send());
         neighbor* n = rev_neigh[from];
         n->set_load(msg->get_amount() + n->get_to_send());

+        if (opt::bookkeeping) {
+            double credit = msg->get_credit();
+            expected_load += credit;
+            n->set_credit(n->get_credit() + credit);
+        }

         break;
     }
         break;
     }

-    case message::CREDIT:
-        expected_load += msg->get_amount();
-        break;
-    case message::LOAD: {
+    case message::DATA: {
+        neighbor* n = rev_neigh[from];

         double ld = msg->get_amount();
         double ld = msg->get_amount();

-        real_load += ld;
-        if (finalizing)
-            total_load_running -= ld;
+        received_load += ld;
+        n->set_credit(n->get_credit() - ld);

         break;
     }
     case message::CTRL_CLOSE:
         break;
     }
     case message::CTRL_CLOSE:


@@ -402,12 +482,11 @@ void process::handle_message(message* msg, m_host_t from)
 
 #define print_loads_generic(vec, verbose, logp, cat)                    \
     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 
 #define print_loads_generic(vec, verbose, logp, cat)                    \
     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \

-        using std::tr1::bind;                                           \
-        using std::tr1::placeholders::_1;                               \
+        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(),                           \
         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
                   "Neighbor loads:", real_load, expected_load);         \
         std::for_each(vec.begin(), vec.end(),                           \

-                      bind(&neighbor::print, _1, verbose, logp, cat));  \
+                      std::bind(&neighbor::print, _1, verbose, logp, cat)); \

     } else ((void)0)
 
 void process::print_loads(bool verbose,
     } else ((void)0)
 
 void process::print_loads(bool verbose,