Bug fix: only send to a lesser loaded neighbor!

[loba.git] / process.cpp

diff --git a/process.cpp b/process.cpp
index b30fab4b0a9d8a43838e1d346bade1e06baa64f8..a74b1149fcdc406358c7ad320aafd85a337d05af 100644 (file)
--- a/process.cpp
+++ b/process.cpp
@@ -1,62 +1,168 @@
 #include <algorithm>
 #include <algorithm>

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

 #include <iterator>
 #include <stdexcept>
 #include <sstream>
 #include <xbt/log.h>
 #include <xbt/time.h>
 #include <iterator>
 #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 "tracing.h"
+

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

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

-    expected_load = load;
+
+    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_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;
+

     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::ostream_iterator<const char*>(oss, ", "),
         std::transform(neigh.begin(), neigh.end() - 1,
                        std::ostream_iterator<const char*>(oss, ", "),

-                       std::mem_fun_ref(&neighbor::get_name));
+                       std::tr1::mem_fn(&neighbor::get_name));

         oss << neigh.back().get_name();
     }
     LOG1(logp, "Got %s.", oss.str().c_str());
         oss << neigh.back().get_name();
     }
     LOG1(logp, "Got %s.", oss.str().c_str());

-    print_loads(logp);
+    print_loads(false, logp);
+}
+
+process::~process()
+{
+    total_load_exit += real_load;
+    if (opt::bookkeeping) {
+        INFO4("Final load after %d:%d iterations: %g ; expected: %g",
+              lb_iter, comp_iter, real_load, expected_load);
+    } else {
+        INFO2("Final load after %d iterations: %g",
+              lb_iter, real_load);
+        if (lb_iter != comp_iter)
+            WARN2("lb_iter (%d) and comp_iter (%d) differ!",
+                  lb_iter, comp_iter);
+    }
+    VERB1("Total computation for this process: %g", comp);

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

-    INFO0("Coucou !");
+    double next_iter_after_date = 0.0;
+    INFO1("Initial load: %g", real_load);
+    VERB0("Starting...");
+    comp_iter = lb_iter = 0;
+    while (true) {
+        double ld = get_load();
+        if (ld > 0.0) {
+            double now = MSG_get_clock();
+            if (now < next_iter_after_date)
+                MSG_process_sleep(next_iter_after_date - now);
+            next_iter_after_date = MSG_get_clock() + opt::min_iter_duration;
+
+            ++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, real_load, expected_load);
+                else
+                    INFO2("(%u) current load: %g",
+                          lb_iter, real_load);
+            }

 
 

-    int n = 100;
-    while (n--) {
-        if (load > 0)
+            ld -= load_balance(ld);
+
+            print_loads(true, xbt_log_priority_debug);
+        }
+        set_load(ld);
+
+        // send load information, and load (data) if any
+        send_all();
+        if (real_load > 0.0) {
+            ++comp_iter;

             compute();
             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
+        double timeout;
+        if (real_load != 0 || get_load() != prev_load_broadcast)
+            timeout = 0.0;
+        else if (opt::min_iter_duration)
+            timeout = opt::min_iter_duration;

         else
         else

-            xbt_sleep(0.5);     // fixme
-        receive();
+            timeout = 1.0;
+        receive(timeout);
+
+        // 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 (real_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);
+        }

     }
     }

-    //    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;
-     * }
-     * finalize;
-     * wait for pending messages;
-     */
+    VERB0("Going to finalize...");
+    finalize();

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


@@ -64,61 +170,181 @@ int process::run()
      * - how to manage link failures ?
      */
 
      * - how to manage link failures ?
      */
 

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

     return 0;
 }
 
     return 0;
 }
 

-void process::receive()
+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()
+{
+    if (real_load > 0.0) {
+        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);
+        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::send1_no_bookkeeping(neighbor& nb)
+{
+    if (real_load != prev_load_broadcast)
+        comm.send(nb.get_ctrl_mbox(), new message(message::INFO, real_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 {
+        new_debt = nb.get_debt();
+    }
+    if (real_load <= new_debt) {
+        load_to_send = real_load;
+        nb.set_debt(new_debt - load_to_send);
+        real_load = 0.0;
+    } else {
+        load_to_send = new_debt;
+        nb.set_debt(0.0);
+        real_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_all()
+{
+    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 = real_load;
+    }
+    comm.flush(false);
+}
+
+void process::receive(double timeout)

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

-    while (comm.recv(msg, from)) {
+
+    DEBUG2("%sblocking receive (%g)", "\0non-" + !timeout, timeout);
+    while (may_receive() && comm.recv(msg, from, timeout)) {

         switch (msg->get_type()) {
         switch (msg->get_type()) {

-        case message::INFO:
-            // fixme: update neighbor
+        case message::INFO: {
+            neighbor* n = rev_neigh[from];
+            n->set_load(msg->get_amount());

             break;
             break;

+        }

         case message::CREDIT:
             expected_load += msg->get_amount();
             break;
         case message::CREDIT:
             expected_load += msg->get_amount();
             break;

-        case message::LOAD:
-            load += msg->get_amount();
+        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;
             break;
         }
         delete msg;

+        timeout = 0.0;          // only wait on first recv

     }
     }

+    comm.flush(false);

 }
 
 }
 

-void process::compute()
+void process::finalize1(neighbor& nb)

 {
 {

-    double duration = opt::comp_cost(load);
-    m_task_t task = MSG_task_create("computation", duration, 0.0, NULL);
-    DEBUG2("Compute %g flop%s.", duration, duration > 1 ? "s" : "");
-    MSG_task_execute(task);
-    MSG_task_destroy(task);
+    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()
 {
 }
 
 void process::finalize()
 {

-    // fixme
+    using std::tr1::bind;
+    using std::tr1::placeholders::_1;
+
+    finalizing = true;
+    total_load_running -= real_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(-1.0);
+    }
+
+    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(e_xbt_log_priority_t logp)
+void process::print_loads_p(bool verbose,
+                            e_xbt_log_priority_t logp,
+                            xbt_log_category_t cat) const

 {
 {

-    if (!LOG_ISENABLED(logp))
-        return;
-    std::ostringstream oss;
-    if (neigh.empty()) {
-        oss << "no neighbor!";
-    } else {
-        std::transform(neigh.begin(), neigh.end() - 1,
-                       std::ostream_iterator<double>(oss, ", "),
-                       std::mem_fun_ref(&neighbor::get_load));
-        oss << neigh.back().get_load();
-    }
-    LOG1(logp, "Neighbor loads: %s", oss.str().c_str());
+    print_loads_generic(pneigh, verbose, logp, cat);

 }
 
 }
 

+#undef print_loads_generic
+

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