There is a bug...

[loba.git] / process.cpp

diff --git a/process.cpp b/process.cpp
index 76058180c01d7a4e08cd008b07eaa043e0575fbf..0544b21d9e5b62b08a6a09a25618386913c0d377 100644 (file)
--- a/process.cpp
+++ b/process.cpp
@@ -1,54 +1,156 @@
 #include <algorithm>
 #include <algorithm>

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

 #include <iterator>
 #include <iterator>

+#include <numeric>

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

+#include <xbt/time.h>
+
+XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
+

 #include "misc.h"
 #include "options.h"
 #include "misc.h"
 #include "options.h"

-#include "process.h"

 
 

-XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simu);
+#include "process.h"

 
 

-struct message {
-    double measure;
-    double transfer;
-};
+double process::total_load_init = 0.0;
+double process::total_load_running = 0.0;
+double process::total_load_exit = 0.0;

 
 

-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]) >> load))

-        throw std::invalid_argument("bad or missing initial load");
+        throw std::invalid_argument("bad or missing initial load parameter");
+

     neigh.assign(argv + 2, argv + argc);
     neigh.assign(argv + 2, argv + argc);

+
+    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());
+        pneigh.push_back(ptr);
+        rev_neigh.insert(std::make_pair(host, ptr));
+    }
+
+    comp = 0.0;
+
+    prev_load_broadcast = -1;   // force sending of load on first send()
+    expected_load = load;
+    total_load_running += load;
+    total_load_init += load;
+
+    ctrl_close_pending = data_close_pending = neigh.size();
+    close_received = false;
+    finalizing = false;
+

     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;

-    LOG1(logp, "My initial load is: %g", load);

     std::ostringstream oss;
     oss << neigh.size() << " neighbor";
     if (!neigh.empty()) {
     std::ostringstream oss;
     oss << neigh.size() << " neighbor";
     if (!neigh.empty()) {

-        oss << (neigh.size() > 1 ? "s: " : ": ");
+        oss << ESSE(neigh.size()) << ": ";

         std::transform(neigh.begin(), neigh.end() - 1,
         std::transform(neigh.begin(), neigh.end() - 1,

-                       std::ostream_iterator<std::string>(oss, ", "),
-                       std::mem_fun_ref(&neighbor::getName));
-        oss << neigh.back().getName();
+                       std::ostream_iterator<const char*>(oss, ", "),
+                       std::tr1::mem_fn(&neighbor::get_name));
+        oss << neigh.back().get_name();

     }
     LOG1(logp, "Got %s.", oss.str().c_str());
     }
     LOG1(logp, "Got %s.", oss.str().c_str());

-    print_loads(logp);
+    print_loads(false, logp);
+}
+
+process::~process()
+{
+    total_load_exit += load;

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

-    INFO0("Coucou !");
-    MSG_process_sleep(100.0);   // xxx
-    /* xxx:
-     * while (there is something to do) {
-     *    compute some task;
-     *    get received tasks;
-     *    compute load balancing;
-     *    send tasks to neighbors;
-     * }
-     */
+    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(delay);
+            }
+            next_iter_after_date = MSG_get_clock() + opt::min_iter_duration;
+        }
+
+        double ld = lb_load();
+        if (ld > 0.0) {
+            ++lb_iter;
+
+            if (opt::log_rate && lb_iter % opt::log_rate == 0) {
+                if (opt::bookkeeping)
+                    INFO4("(%u:%u) current load: %g ; expected: %g",
+                          lb_iter, comp_iter, load, expected_load);
+                else
+                    INFO2("(%u) current load: %g",
+                          lb_iter, load);
+            }
+
+            ld -= load_balance(ld);
+
+            print_loads(true, xbt_log_priority_debug);
+        }
+        lb_load() = ld;
+
+        // send load information, and load (data) if any
+        send();
+        if (load > 0.0) {
+            ++comp_iter;
+            compute();
+        }
+
+        if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
+            VERB2("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
+            break;
+        }
+        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;
+        }
+
+        // block on receiving unless there is something to compute or
+        // to send
+        bool wait = (load == 0 && lb_load() == prev_load_broadcast);
+        receive(wait);
+
+        // one of our neighbor is finalizing
+        if (opt::exit_on_close && close_received) {
+            VERB0("Close received");
+            break;
+        }
+
+        // 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;
+        }
+
+        // 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;
+        } else {
+            DEBUG1("still %g load to balance, continuing...", total_load_running);
+        }
+    }
+    VERB0("Going to finalize...");
+    finalize();

 
     /* Open Questions :
      * - definition of load on heterogeneous hosts ?
 
     /* Open Questions :
      * - definition of load on heterogeneous hosts ?


@@ -56,35 +158,201 @@ int process::run()
      * - how to manage link failures ?
      */
 
      * - how to manage link failures ?
      */
 

-    // xxx: shall we retrieve pending tasks?
-
+    VERB0("Done.");
+    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);
+    }

     return 0;
 }
 
     return 0;
 }
 

+double process::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)));
+}
+
+double process::load_balance(double /*my_load*/)
+{
+    if (lb_iter == 1)           // warn only once
+        WARN0("process::load_balance() is a no-op!");
+    return 0.0;
+}
+

 void process::compute()
 {
 void process::compute()
 {

-    double duration = opt::comp_cost(load);
-    m_task_t task = MSG_task_create("computation", duration, 0.0, NULL);
-    MSG_task_execute(task);
-    MSG_task_destroy(task);
+    if (load > 0.0) {
+        double flops = opt::comp_cost(load);
+        m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
+        DEBUG2("compute %g flop%s", flops, ESSE(flops));
+        MSG_task_execute(task);
+        comp += flops;
+        MSG_task_destroy(task);
+    } else {
+        DEBUG0("nothing to compute !");
+    }

 }
 
 }
 

-void process::print_loads(e_xbt_log_priority_t logp)
+void process::send1_no_bookkeeping(neighbor& nb)

 {
 {

-    if (!LOG_ISENABLED(logp))
-        return;
-    std::ostringstream oss;
-    if (neigh.empty()) {
-        oss << "no neighbor!";
+    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);
+    }
+}
+
+void process::send1_bookkeeping(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 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;

     } else {
     } else {

-        std::transform(neigh.begin(), neigh.end() - 1,
-                       std::ostream_iterator<double>(oss, ", "),
-                       std::mem_fun_ref(&neighbor::getLoad));
-        oss << neigh.back().getLoad();
+        new_debt = nb.get_debt();
+    }
+    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;
+    }
+    if (load_to_send > 0.0)
+        comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
+}
+
+void process::send()
+{
+    using std::tr1::bind;
+    using std::tr1::placeholders::_1;
+
+    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::receive(bool wait)
+{
+    message* msg;
+    m_host_t from;
+
+    DEBUG1("%sblocking receive", "\0non-" + !wait);
+    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

     }
     }

-    LOG1(logp, "Neighbor loads: %s", oss.str().c_str());
+    comm.flush(false);

 }
 
 }
 

+void process::finalize1(neighbor& nb)
+{
+    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));
+}
+
+void process::finalize()
+{
+    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);
+    }
+
+    comm.flush(true);
+}
+
+#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:");                           \
+        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: