Add a bug entry about mutex usage in process.

[loba.git] / process.cpp

diff --git a/process.cpp b/process.cpp
index 0afbf2950781c4ac33139c2cbfbb3f2bf0a70ebe..54fea8aeb2ed8534d4561715b9595b0950c967ef 100644 (file)
--- a/process.cpp
+++ b/process.cpp
@@ -11,6 +11,7 @@ XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
 
 #include "misc.h"
 #include "options.h"
 
 #include "misc.h"
 #include "options.h"

+#include "tracing.h"

 
 #include "process.h"
 
 
 #include "process.h"
 


@@ -18,9 +19,21 @@ 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;
 

+namespace {
+
+    void sleep_until_date(double& date, double duration)
+    {
+        double sleep_duration = date - MSG_get_clock();
+        if (sleep_duration > 0.0)
+            MSG_process_sleep(sleep_duration);
+        date = MSG_get_clock() + duration;
+    }
+
+}
+

 process::process(int argc, char* argv[])
 {
 process::process(int argc, char* argv[])
 {

-    if (argc < 2 || !(std::istringstream(argv[1]) >> load))
+    if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))

         throw std::invalid_argument("bad or missing initial load parameter");
 
     neigh.assign(argv + 2, argv + argc);
         throw std::invalid_argument("bad or missing initial load parameter");
 
     neigh.assign(argv + 2, argv + argc);


@@ -33,15 +46,21 @@ 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;
-    total_load_running += load;
-    total_load_init += 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;

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


@@ -54,120 +73,197 @@ process::process(int argc, char* argv[])
                        std::tr1::mem_fn(&neighbor::get_name));
         oss << neigh.back().get_name();
     }
                        std::tr1::mem_fn(&neighbor::get_name));
         oss << neigh.back().get_name();
     }

-    LOG1(logp, "Got %s.", oss.str().c_str());
+    XBT_LOG(logp, "Got %s.", oss.str().c_str());

     print_loads(false, logp);
 }
 
 process::~process()
 {
     print_loads(false, logp);
 }
 
 process::~process()
 {

-    total_load_exit += load;
+    delete lb_thread;
+    total_load_exit += real_load;
+    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()
 {

-    double next_iter_after_date = 0.0;
-    INFO1("Initial load: %g", load);
-    VERB0("Starting...");
-    comp_iter = lb_iter = 0;
-    while (true) {
-        if (opt::min_iter_duration) {
-            double now = MSG_get_clock();
-            if (now < next_iter_after_date){
-                double delay = next_iter_after_date - now;
-                DEBUG1("sleeping for %g s", delay);
-                MSG_process_sleep(next_iter_after_date - now);
-            }
-            next_iter_after_date = MSG_get_clock() + opt::min_iter_duration;
+    if (opt::log_rate >= 0) {
+        XBT_INFO("Initial load: %g", real_load);
+        XBT_VERB("Initial expected load: %g", expected_load);
+    }
+    XBT_VERB("Starting...");
+    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;
+}
+
+void process::load_balance_loop()
+{
+    using std::tr1::bind;
+    using std::tr1::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) {
+            mutex.acquire();
+            ++lb_iter;
+            cond.signal();
+            mutex.release();
+        } else {
+            ++lb_iter;

         }
 
         }
 

-        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);
-            }
+        mutex.acquire();
+        if (!opt::bookkeeping)
+            expected_load = real_load - get_sum_of_to_send();
+        // nothing to do with opt::bookkeeping

 
 

-            double ld = lb_load();
-            ld -= load_balance(ld);
-            lb_load() = ld;
+        if (opt::log_rate && lb_iter % opt::log_rate == 0) {
+            XBT_INFO("(%u:%u) current load: %g", lb_iter, comp_iter, real_load);
+            XBT_VERB("... expected load: %g", expected_load);
+        }

 
 

-            print_loads(true, xbt_log_priority_debug);
+        if (expected_load > 0.0)
+            load_balance();

 
 

-            send();
-            compute();
+        print_loads(true, xbt_log_priority_debug);

 
 

-        } else {
-            // send load information, and load when bookkeeping
-            send();
-        }
+        // send
+        comm.ctrl_flush(false);
+        std::for_each(neigh.begin(), neigh.end(),
+                      bind(&process::ctrl_send, this, _1));
+        prev_load_broadcast = expected_load;
+        mutex.release();

 
 

-        if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
-            VERB2("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
-            break;
-        }
-        if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
-            VERB2("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_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;
-        }
+        sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
+        ctrl_receive(0.0);
+    }

 
 

-        // block on receiving unless there is something to compute or
-        // to send
-        bool wait = (load == 0 && lb_load() == prev_load_broadcast);
-        receive(wait);
+    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));
+    while (ctrl_close_pending) {
+        comm.ctrl_flush(false);
+        XBT_DEBUG("waiting for %d CTRL CLOSE", ctrl_close_pending);
+        ctrl_receive(-1.0);
+    }
+    comm.ctrl_flush(true);
+}

 
 

-        // one of our neighbor is finalizing
-        if (opt::exit_on_close && close_received) {
-            VERB0("Close received");
-            break;
-        }
+void process::compute_loop()
+{
+    using std::tr1::bind;
+    using std::tr1::placeholders::_1;

 
 

-        // have no load and cannot receive anything
-        if (load == 0.0 && !may_receive()) {
-            VERB0("I'm a poor lonesome process, and I have no load...");
-            break;
-        }
+    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);
+        mutex.release();
+
+        // send
+        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;
+
+        // compute
+        ++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);
+        XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
+        MSG_task_execute(task);
+        add_comp_amount(flops);
+        MSG_task_destroy(task);

 
 

-        // 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.");
-            break;
-        }
+        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(),
+                  bind(&process::data_close, this, _1));
+    while (data_close_pending) {
+        comm.data_flush(false);
+        XBT_DEBUG("waiting for %d DATA CLOSE", data_close_pending);
+        data_receive(-1.0);

     }
     }

-    VERB0("Going to finalize...");
-    finalize();
+    comm.data_flush(true);
+}

 
 

-    /* Open Questions :
-     * - definition of load on heterogeneous hosts ?
-     * - how to detect convergence ?
-     * - how to manage link failures ?
-     */
+bool process::still_running()
+{
+    static bool last_status = true;

 
 

-    VERB0("Done.");
-    if (opt::bookkeeping) {
-        INFO4("Final load after %d:%d iterations: %g ; expected: %g",
-              comp_iter, lb_iter, load, expected_load);
-    } else {
-        INFO2("Final load after %d iterations: %g",
-              comp_iter, load);
-        if (comp_iter != lb_iter)
-            WARN2("comp_iter (%d) and lb_iter (%d) differ!",
-                  comp_iter, lb_iter);
+    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::lb_maxiter && lb_iter >= opt::lb_maxiter) {
+        XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
+        last_status = false;
+
+    } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
+        XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
+        last_status = false;
+
+    } else if (opt::exit_on_close && close_received) {
+        XBT_VERB("Close received");
+        last_status = false;
+
+    } else if (real_load == 0.0 && !data_close_pending) {
+        XBT_VERB("I'm a poor lonesome process, and I have no load...");
+        last_status = false;
+
+    } else if (100.0 * total_load_running / total_load_init <=
+               opt::load_ratio_threshold) {
+        // 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.");
+        last_status = false;

     }
     }

-    return 0;
+
+    return last_status;

 }
 
 }
 

-double process::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::bind;
     using std::tr1::placeholders::_1;


@@ -178,154 +274,146 @@ double process::sum_of_to_send() const
                                 _1, bind(&neighbor::get_to_send, _2)));
 }
 
                                 _1, bind(&neighbor::get_to_send, _2)));
 }
 

-double process::load_balance(double /*my_load*/)
+void process::load_balance()

 {
     if (lb_iter == 1)           // warn only once
 {
     if (lb_iter == 1)           // warn only once

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

 }
 
 }
 

-void process::compute()
+void process::send(neighbor& nb, double amount)

 {
 {

-    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));
-        MSG_task_execute(task);
-        MSG_task_destroy(task);
-    } else {
-        DEBUG0("nothing to compute !");
-    }
+    expected_load -= amount;
+    nb.set_to_send(nb.get_to_send() + amount);
+    nb.set_load(nb.get_load() + amount);

 }
 
 }
 

-void process::send1_no_bookkeeping(neighbor& nb)
+void process::ctrl_send(neighbor& nb)

 {
 {

-    if (load != prev_load_broadcast)
-        comm.send(nb.get_ctrl_mbox(), new message(message::INFO, load));
-    double load_to_send = nb.get_to_send();
-    if (load_to_send > 0.0) {
-        comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
-        nb.set_to_send(0.0);
+    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);
+            message* msg = new message(message::CREDIT, debt_to_send);
+            add_ctrl_send_mesg(msg->get_size());
+            comm.ctrl_send(nb.get_ctrl_mbox(), msg);
+        }

     }
 }
 
     }
 }
 

-void process::send1_bookkeeping(neighbor& nb)
+void process::data_send(neighbor& nb)

 {
 {

-    if (expected_load != prev_load_broadcast)
-        comm.send(nb.get_ctrl_mbox(),
-                  new message(message::INFO, expected_load));

     double load_to_send;
     double load_to_send;

-    double new_debt;
-    double debt_to_send = nb.get_to_send();
-    if (debt_to_send > 0.0) {
-        comm.send(nb.get_ctrl_mbox(),
-                  new message(message::CREDIT, debt_to_send));
-        nb.set_to_send(0.0);
-        new_debt = nb.get_debt() + debt_to_send;
+    if (opt::bookkeeping) {
+        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);
+            real_load -= load_to_send;
+        } else {
+            load_to_send = 0.0;
+        }

     } else {
     } else {

-        new_debt = nb.get_debt();
+        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;
+        }

     }
     }

-    if (load <= new_debt) {
-        load_to_send = load;
-        nb.set_debt(new_debt - load_to_send);
-        load = 0.0;
-    } else {
-        load_to_send = new_debt;
-        nb.set_debt(0.0);
-        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;
+        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.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));

 }
 
 }
 

-void process::send()
+void process::ctrl_close(neighbor& nb)

 {
 {

-    using std::tr1::bind;
-    using std::tr1::placeholders::_1;
+    comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
+}

 
 

-    if (opt::bookkeeping) {
-        std::for_each(neigh.begin(), neigh.end(),
-                      bind(&process::send1_bookkeeping, this, _1));
-        prev_load_broadcast = expected_load;
-    } else {
-        std::for_each(neigh.begin(), neigh.end(),
-                      bind(&process::send1_no_bookkeeping, this, _1));
-        prev_load_broadcast = load;
-    }
-    comm.flush(false);
+void process::data_close(neighbor& nb)
+{
+    comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));

 }
 
 }
 

-void process::receive(bool wait)
+void process::ctrl_receive(double timeout)

 {
     message* msg;
     m_host_t from;
 
 {
     message* msg;
     m_host_t from;
 

-    while (may_receive() && comm.recv(msg, from, wait)) {
-        switch (msg->get_type()) {
-        case message::INFO: {
-            neighbor* n = rev_neigh[from];
-            n->set_load(msg->get_amount());
-            break;
-        }
-        case message::CREDIT:
-            expected_load += msg->get_amount();
-            break;
-        case message::LOAD: {
-            double ld = msg->get_amount();
-            load += ld;
-            if (finalizing)
-                total_load_running -= ld;
-            break;
-        }
-        case message::CTRL_CLOSE:
-            ctrl_close_pending--;
-            close_received = true;
-            break;
-        case message::DATA_CLOSE:
-            data_close_pending--;
-            close_received = true;
-            break;
-        }
-        delete msg;
-        wait = false;           // only wait on first recv
+    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;

     }
     }

-    comm.flush(false);

 }
 
 }
 

-void process::finalize1(neighbor& nb)
+void process::data_receive(double timeout)

 {
 {

-    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));
+    message* msg;
+    m_host_t from;
+
+    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;
+    }

 }
 
 }
 

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

 {
 {

-    using std::tr1::bind;
-    using std::tr1::placeholders::_1;
-
-    finalizing = true;
-    total_load_running -= load;
-
-    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));
-
-    DEBUG2("wait for CLOSE from %lu neighbor%s",
-           (unsigned long )neigh.size(), ESSE(neigh.size()));
-    while (may_receive()) {
-        comm.flush(false);
-        receive(true);
+    switch (msg->get_type()) {
+    case message::INFO: {
+        neighbor* n = rev_neigh[from];
+        n->set_load(msg->get_amount() + n->get_to_send());
+        break;

     }
     }

-
-    comm.flush(true);
+    case message::CREDIT:
+        expected_load += msg->get_amount();
+        break;
+    case message::LOAD: {
+        double ld = msg->get_amount();
+        real_load += ld;
+        if (finalizing)
+            total_load_running -= ld;
+        break;
+    }
+    case message::CTRL_CLOSE:
+        ctrl_close_pending--;
+        close_received = true;
+        break;
+    case message::DATA_CLOSE:
+        data_close_pending--;
+        close_received = true;
+        break;
+    }
+    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;                               \
 }
 
 #define print_loads_generic(vec, verbose, logp, cat)                    \
     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
         using std::tr1::bind;                                           \
         using std::tr1::placeholders::_1;                               \

-        XCLOG0(cat, logp, "Neighbor loads:");                           \
+        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)
         std::for_each(vec.begin(), vec.end(),                           \
                       bind(&neighbor::print, _1, verbose, logp, cat));  \
     } else ((void)0)