double process::total_load_running = 0.0;
double process::total_load_exit = 0.0;
+int process::process_counter = 0;
+double process::total_load_average;
+double process::load_diff_threshold;
+
namespace {
void sleep_until_date(double& date, double duration)
rev_neigh.insert(std::make_pair(host, ptr));
}
+ // Note: there should not be race condition with the current
+ // version of Simgrid, when updating the global variables.
+
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;
received_load = 0.0;
+ idle_duration = 0.0;
+ convergence = -1.0;
+
+ process_counter++;
+ total_load_average = total_load_running / process_counter;
+ load_diff_threshold = (opt::load_ratio_threshold +
+ opt::avg_load_ratio * total_load_average) / 100.0;
+
ctrl_close_pending = data_close_pending = neigh.size();
close_received = false;
finalizing = false;
- comp_iter = lb_iter = 0;
+ all_comp_iter = comp_iter = lb_iter = 0;
lb_thread = new_msg_thread("loba",
std::bind(&process::load_balance_loop, this));
"received_load is %g, but should be 0.0 !", received_load);
if (opt::log_rate < 0)
return;
- XBT_INFO("Final load after %d:%d iterations: %g",
- lb_iter, comp_iter, real_load);
+ XBT_INFO("Final load after %d:%d:%d iterations: %g",
+ lb_iter, comp_iter, all_comp_iter, real_load);
+ if (convergence >= 0.0)
+ XBT_INFO("Convergence within %g%% was achieved at time %g",
+ opt::avg_load_ratio, convergence);
+ else
+ XBT_INFO("Convergence within %g%% was not achieved",
+ opt::avg_load_ratio);
XBT_VERB("Expected load was: %g", expected_load);
XBT_VERB("Total computation for this process: %g", get_comp_amount());
print_loads(true, xbt_log_priority_debug);
// nothing to do with opt::bookkeeping
if (opt::log_rate && lb_iter % opt::log_rate == 0) {
- XBT_INFO("(%u:%u) current load: %g", lb_iter, comp_iter, real_load);
+ XBT_INFO("(%u:%u:%u) current load: %g",
+ lb_iter, comp_iter, all_comp_iter, real_load);
XBT_VERB("... expected load: %g", expected_load);
}
using std::placeholders::_1;
double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
+ double idle_since_date = 0.0;
while (still_running()) {
- // receive (do not block if there is something to compute)
- data_receive(real_load > 0.0 ? 0.0 : opt::min_comp_iter_duration);
- // send
- comm.data_flush(false);
- mutex.acquire();
- real_load += received_load;
- received_load = 0.0;
- std::for_each(neigh.begin(), neigh.end(),
- std::bind(&process::data_send, this, _1));
- mutex.release();
+ do {
+ // 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;
+ std::for_each(neigh.begin(), neigh.end(),
+ std::bind(&process::data_send, this, _1));
+ mutex.release();
- if (real_load == 0.0)
- continue;
+ ++all_comp_iter;
+
+ } while (real_load == 0.0);
+
+ convergence_check();
// compute
+ idle_duration += MSG_get_clock() - idle_since_date;
++comp_iter;
double flops = opt::comp_cost(real_load);
m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
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);
}
real_load += received_load;
received_load = 0.0;
total_load_running -= real_load;
+ convergence_check();
comm.data_flush(true);
}
+void process::convergence_check()
+{
+ double load_diff = std::fabs(real_load - total_load_average);
+ bool converged = load_diff <= load_diff_threshold;
+
+ if (convergence >= 0.0) {
+ if (!converged) {
+ XBT_VERB("current load has diverged: %g (%.4g%%)",
+ real_load, 100.0 * load_diff / total_load_average);
+ convergence = -1.0;
+ }
+ } else {
+ if (converged) {
+ XBT_VERB("current load has converged: %g (%.4g%%)",
+ real_load, 100.0 * load_diff / total_load_average);
+ convergence = MSG_get_clock();
+ }
+ }
+}
+
bool process::still_running()
{
static bool last_status = true;
{
double load_to_send;
if (opt::bookkeeping) { // bookkeeping
- double excess_load;
+ double excess_load; // load amount we are able to send
if (opt::egocentric)
excess_load = std::max(0.0, real_load - expected_load);
else
excess_load = real_load;
- load_to_send = compute_load_to_send(std::min(excess_load,
- nb.get_debt()));
+
+ double balance = nb.get_debt() - nb.get_credit();
+ load_to_send = std::min(excess_load,
+ std::max(0.0, balance));
+
+ // adjust load to send (rounding, truncation, etc.)
+ load_to_send = compute_load_to_send(load_to_send);
if (load_to_send > 0.0)
nb.set_debt(nb.get_debt() - load_to_send);
} else { // !bookkeeping
