+ 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()
+{
+ 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;
+ }
+
+ 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) {
+ XBT_INFO("(%u:%u:%u) current load: %g",
+ lb_iter, comp_iter, all_comp_iter, real_load);
+ XBT_VERB("... expected load: %g", expected_load);
+ }
+
+ if (expected_load > 0.0)
+ load_balance();
+
+ print_loads(true, xbt_log_priority_debug);
+
+ // send
+ comm.ctrl_flush(false);
+ for (neighbor& n : neigh)
+ ctrl_send(n);
+ prev_load_broadcast = expected_load;
+ mutex.release();
+
+ sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
+ }
+
+ XBT_VERB("Going to finalize for %s...", __func__);
+ XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
+ neigh.size(), ESSE(neigh.size()));
+ for (neighbor& n : neigh)
+ ctrl_close(n);
+ 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);
+}
+
+void process::compute_loop()
+{
+ double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
+ double idle_since_date = 0.0;
+ while (still_running()) {
+ // 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);
+
+ // send
+ comm.data_flush(false);
+ mutex.acquire();
+ real_load += received_load;
+ received_load = 0.0;
+ for (neighbor& n : neigh)
+ data_send(n);
+ mutex.release();
+
+ ++all_comp_iter;
+ if (real_load == 0.0)
+ continue;
+
+ convergence_check();
+
+ // compute
+ idle_duration += MSG_get_clock() - idle_since_date;
+ ++comp_iter;
+ double flops = opt::comp_cost(real_load);
+ msg_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
+ // MSG_task_set_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);
+
+ 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__);
+ // Note: idle duration is not counted during finalization
+ finalizing = true;
+ XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
+ neigh.size(), ESSE(neigh.size()));
+ for (neighbor& n : neigh)
+ data_close(n);
+ while (data_close_pending) {
+ comm.data_flush(false);
+ XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);
+ data_receive(-1.0);
+ }
+ real_load += received_load;
+ received_load = 0.0;
+ proc_mutex->acquire();
+ total_load_running -= real_load;
+ proc_mutex->release();
+ convergence_check();
+ comm.data_flush(true);
+}
+
+void process::convergence_check()
+{
+ double average = total_load_average; // fixme: get locked?
+ double load_diff = std::fabs(real_load - average);
+ bool converged = load_diff <= load_diff_threshold;
+
+ if (converged) {
+ if (convergence < 0) {
+ XBT_VERB("current load has converged: %g (%.4g%%)",
+ real_load, 100.0 * load_diff / average);
+ convergence = MSG_get_clock();
+ local_convergence_counter = opt::exit_on_convergence;
+ }
+ if (local_convergence_counter > 0 && --local_convergence_counter == 0)
+ --convergence_counter;
+ } else {
+ if (convergence >= 0.0) {
+ XBT_VERB("current load has diverged: %g (%.4g%%)",
+ real_load, 100.0 * load_diff / average);
+ convergence = -1.0;
+ if (local_convergence_counter == 0)
+ ++convergence_counter;
+ }
+ }
+}
+
+bool process::still_running()
+{
+ static bool last_status = true;
+
+ 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_convergence && convergence_counter == 0) {
+ XBT_VERB("Global convergence detected");
+ 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: get locked?
+ // 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 last_status;
+}
+
+double process::get_sum_of_to_send() const
+{
+ return std::accumulate(neigh.begin(), neigh.end(), 0.0,
+ [](double x, const neighbor& neigh) {
+ return x + neigh.get_to_send();
+ });
+}
+
+void process::load_balance()
+{
+ if (lb_iter == 1) // warn only once
+ XBT_WARN("process::load_balance() is a no-op!");
+}
+
+void process::send(neighbor& nb, double amount)
+{
+ expected_load -= amount;
+ nb.set_to_send(nb.get_to_send() + amount);
+ nb.set_load(nb.get_load() + amount);
+}
+
+void process::ctrl_send(neighbor& nb)
+{
+ double info_to_send = expected_load;
+ 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);
+ }
+ } 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;
+}
