Added Least Load Neighbors and another bulk algorithm

[loba.git] / process.cpp

diff --git a/process.cpp b/process.cpp
index eeb5d8c0052ba166f56a20512053e6d4c62c88db..54fea8aeb2ed8534d4561715b9595b0950c967ef 100644 (file)
--- a/process.cpp
+++ b/process.cpp
@@ -1,6 +1,7 @@
 #include <algorithm>
 #include <tr1/functional>
 #include <iterator>
 #include <algorithm>
 #include <tr1/functional>
 #include <iterator>

+#include <numeric>

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


@@ -20,7 +21,7 @@ double process::total_load_exit = 0.0;
 
 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)


@@ -45,8 +46,6 @@ process::process(int argc, char* argv[])
         rev_neigh.insert(std::make_pair(host, ptr));
     }
 
         rev_neigh.insert(std::make_pair(host, ptr));
     }
 

-    comp = 0.0;
-

     prev_load_broadcast = -1;   // force sending of load on first send_all()
     expected_load = real_load;
     total_load_running += real_load;
     prev_load_broadcast = -1;   // force sending of load on first send_all()
     expected_load = real_load;
     total_load_running += real_load;


@@ -58,9 +57,9 @@ process::process(int argc, char* argv[])
 
     comp_iter = lb_iter = 0;
 
 
     comp_iter = lb_iter = 0;
 

-    compute_thread = new_msg_thread("compute",
-                                    std::tr1::bind(&process::compute_loop,
-                                                   this));
+    lb_thread = new_msg_thread("loba",
+                               std::tr1::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))


@@ -80,28 +79,33 @@ process::process(int argc, char* argv[])
 
 process::~process()
 {
 
 process::~process()
 {

-    delete compute_thread;
+    delete lb_thread;

     total_load_exit += real_load;
     total_load_exit += real_load;

-    if (opt::bookkeeping) {
-        XBT_INFO("Final load after %d:%d iterations: %g ; expected: %g",
-                 lb_iter, comp_iter, real_load, expected_load);
-    } else {
-        XBT_INFO("Final load after %d iterations: %g",
-                 lb_iter, real_load);
-        if (lb_iter != comp_iter)
-            XBT_WARN("lb_iter (%d) and comp_iter (%d) differ!",
-                     lb_iter, comp_iter);
-    }
-    XBT_VERB("Total computation for this process: %g", comp);
+    if (opt::log_rate < 0)
+        return;
+    XBT_INFO("Final load after %d:%d iterations: %g",
+             lb_iter, comp_iter, real_load);
+    XBT_VERB("Expected load was: %g", expected_load);
+    XBT_VERB("Total computation for this process: %g", get_comp_amount());

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

-    XBT_INFO("Initial load: %g", real_load);
+    if (opt::log_rate >= 0) {
+        XBT_INFO("Initial load: %g", real_load);
+        XBT_VERB("Initial expected load: %g", expected_load);
+    }

     XBT_VERB("Starting...");
     XBT_VERB("Starting...");

-    compute_thread->start();
-    load_balance_loop();
-    compute_thread->wait();
+    mutex.acquire();
+    lb_thread->start();
+    while (lb_iter <= opt::comp_iter_delay)
+        cond.wait(mutex);
+    mutex.release();
+    double sleep_duration = opt::comp_time_delay - MSG_get_clock();
+    if (sleep_duration > 0.0)
+        MSG_process_sleep(sleep_duration);
+    compute_loop();
+    lb_thread->wait();

     XBT_VERB("Done.");
     return 0;
 }
     XBT_VERB("Done.");
     return 0;
 }


@@ -113,30 +117,39 @@ void process::load_balance_loop()
 
     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()) {

-        ++lb_iter;
+        if (lb_iter == opt::comp_iter_delay) {
+            mutex.acquire();
+            ++lb_iter;
+            cond.signal();
+            mutex.release();
+        } else {
+            ++lb_iter;
+        }
+
+        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::log_rate && lb_iter % opt::log_rate == 0) {

-            if (opt::bookkeeping)
-                XBT_INFO("(%u:%u) current load: %g ; expected: %g",
-                         lb_iter, comp_iter, real_load, expected_load);
-            else
-                XBT_INFO("(%u:%u) current load: %g",
-                         lb_iter, comp_iter, real_load);
+            XBT_INFO("(%u:%u) current load: %g", lb_iter, comp_iter, real_load);
+            XBT_VERB("... expected load: %g", expected_load);

         }
 
         }
 

-        if (get_load() > 0.0)
+        if (expected_load > 0.0)

             load_balance();
 
         print_loads(true, xbt_log_priority_debug);
 
         // send
             load_balance();
 
         print_loads(true, xbt_log_priority_debug);
 
         // send

+        comm.ctrl_flush(false);

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

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

 
         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
         ctrl_receive(0.0);
 
         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_VERB("Going to finalize for %s...", __func__);


@@ -160,20 +173,23 @@ void process::compute_loop()
     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
     while (still_running()) {
         // receive
     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
     while (still_running()) {
         // receive

+        mutex.acquire();

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

+        mutex.release();

 
 

+        // send

         comm.data_flush(false);
         comm.data_flush(false);

+        mutex.acquire();
+        std::for_each(neigh.begin(), neigh.end(),
+                      bind(&process::data_send, this, _1));
+        mutex.release();

 
         if (real_load == 0.0)
             continue;
 
 
         if (real_load == 0.0)
             continue;
 

-        // send
-        std::for_each(neigh.begin(), neigh.end(),
-                      bind(&process::data_send, this, _1));
-

         // compute
         ++comp_iter;
         double flops = opt::comp_cost(real_load);
         // compute
         ++comp_iter;
         double flops = opt::comp_cost(real_load);


@@ -181,13 +197,18 @@ void process::compute_loop()
         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
         MSG_task_execute(task);
         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
         MSG_task_execute(task);

-        comp += flops;
+        add_comp_amount(flops);

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


@@ -207,6 +228,10 @@ 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;


@@ -231,7 +256,6 @@ bool process::still_running()
                opt::load_ratio_threshold) {
         // fixme: this check should be implemented with a distributed
         // algorithm, and not a shared global variable!
                opt::load_ratio_threshold) {
         // 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() ?

         XBT_VERB("No more load to balance in system.");
         last_status = false;
     }
         XBT_VERB("No more load to balance in system.");
         last_status = false;
     }


@@ -239,6 +263,17 @@ bool process::still_running()
     return last_status;
 }
 
     return last_status;
 }
 

+double process::get_sum_of_to_send() const
+{
+    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>(),
+                                _1, bind(&neighbor::get_to_send, _2)));
+}
+

 void process::load_balance()
 {
     if (lb_iter == 1)           // warn only once
 void process::load_balance()
 {
     if (lb_iter == 1)           // warn only once


@@ -247,49 +282,27 @@ void process::load_balance()
 
 void process::send(neighbor& nb, double amount)
 {
 
 void process::send(neighbor& nb, double amount)
 {

-    set_load(get_load() - amount);
+    expected_load -= amount;

     nb.set_to_send(nb.get_to_send() + amount);
     nb.set_load(nb.get_load() + amount);
 }
 
     nb.set_to_send(nb.get_to_send() + amount);
     nb.set_load(nb.get_load() + amount);
 }
 

-#define print_loads_generic(vec, verbose, logp, cat)                    \
-    if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
-        using std::tr1::bind;                                           \
-        using std::tr1::placeholders::_1;                               \
-        XBT_XCLOG(cat, logp, "Neighbor loads:");                        \
-        std::for_each(vec.begin(), vec.end(),                           \
-                      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
-{
-    print_loads_generic(neigh, verbose, logp, cat);
-}
-
-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);
-}
-
-#undef print_loads_generic
-

 void process::ctrl_send(neighbor& nb)
 {
 void process::ctrl_send(neighbor& nb)
 {

-    double info_to_send = get_load();
-    if (info_to_send != prev_load_broadcast)
-        comm.ctrl_send(nb.get_ctrl_mbox(),
-                       new message(message::INFO, info_to_send));
+    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();
         if (debt_to_send > 0.0) {
             nb.set_to_send(0.0);
             nb.set_debt(nb.get_debt() + debt_to_send);
     if (opt::bookkeeping) {
         double 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);

-            comm.ctrl_send(nb.get_ctrl_mbox(),
-                           new message(message::CREDIT, 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);

         }
     }
 }
         }
     }
 }


@@ -298,22 +311,33 @@ void process::data_send(neighbor& nb)
 {
     double load_to_send;
     if (opt::bookkeeping) {
 {
     double load_to_send;
     if (opt::bookkeeping) {

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

             nb.set_debt(nb.get_debt() - load_to_send);
             nb.set_debt(nb.get_debt() - load_to_send);

-            real_load = 0.0;
-        } else {
-            load_to_send = nb.get_debt();
-            nb.set_debt(0.0);

             real_load -= load_to_send;
             real_load -= load_to_send;

+        } else {
+            load_to_send = 0.0;

         }
     } else {
         load_to_send = nb.get_to_send();
         }
     } else {
         load_to_send = nb.get_to_send();

-        nb.set_to_send(0.0);
+        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;
+        }
+    }
+    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);
+        add_data_send_mesg(msg->get_size());
+        comm.data_send(nb.get_data_mbox(), msg);
+        load_to_send -= amount;

     }
     }

-    if (load_to_send > 0.0)
-        comm.data_send(nb.get_data_mbox(),
-                       new message(message::LOAD, load_to_send));

 }
 
 void process::ctrl_close(neighbor& nb)
 }
 
 void process::ctrl_close(neighbor& nb)


@@ -333,6 +357,8 @@ void process::ctrl_receive(double timeout)
 
     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)) {

+        if (msg->get_type() != message::CTRL_CLOSE)
+            add_ctrl_recv_mesg(msg->get_size());

         handle_message(msg, from);
         timeout = 0.0;
     }
         handle_message(msg, from);
         timeout = 0.0;
     }


@@ -345,6 +371,8 @@ void process::data_receive(double timeout)
 
     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)) {

+        if (msg->get_type() != message::DATA_CLOSE)
+            add_data_recv_mesg(msg->get_size());

         handle_message(msg, from);
         timeout = 0.0;
     }
         handle_message(msg, from);
         timeout = 0.0;
     }


@@ -355,7 +383,7 @@ void process::handle_message(message* msg, m_host_t from)
     switch (msg->get_type()) {
     case message::INFO: {
         neighbor* n = rev_neigh[from];
     switch (msg->get_type()) {
     case message::INFO: {
         neighbor* n = rev_neigh[from];

-        n->set_load(msg->get_amount());
+        n->set_load(msg->get_amount() + n->get_to_send());

         break;
     }
     case message::CREDIT:
         break;
     }
     case message::CREDIT:


@@ -380,6 +408,32 @@ 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::tr1::bind;                                           \
+        using std::tr1::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(),                           \
+                      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
+{
+    print_loads_generic(neigh, verbose, logp, cat);
+}
+
+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);
+}
+
+#undef print_loads_generic
+

 // Local variables:
 // mode: c++
 // End:
 // Local variables:
 // mode: c++
 // End: