#include <algorithm>
-#include <tr1/functional>
+#include <cmath>
#include <iterator>
+#include <numeric>
#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 "tracing.h"
+
#include "process.h"
-XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simu);
+mutex_t *process::proc_mutex;
+
+double process::total_load_init = 0.0;
+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::average_load_ratio;
+double process::load_diff_threshold;
+
+std::atomic<int> process::convergence_counter(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[])
{
- 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");
+
+ double iload = std::trunc(real_load);
+ if (opt::integer_transfer && real_load != iload) {
+ XBT_WARN("Initial load %g is not an integer. Truncate it.",
+ real_load);
+ real_load = iload;
+ }
neigh.assign(argv + 2, argv + argc);
- expected_load = load;
- ctrl_close_pending = data_close_pending = neigh.size();
- if (neigh.size() == 1) {
- comm.next_close_on_ctrl_is_last();
- comm.next_close_on_data_is_last();
+ pneigh.reserve(neigh.size());
+ for (unsigned i = 0 ; i < neigh.size() ; i++) {
+ neighbor* ptr = &neigh[i];
+ msg_host_t host = MSG_get_host_by_name(ptr->get_name());
+ pneigh.push_back(ptr);
+ rev_neigh.insert(std::make_pair(host, ptr));
}
- if (neigh.size() > 0)
- comm.listen();
+
+ prev_load_broadcast = -1; // force sending of load on first send_all()
+ expected_load = real_load;
+ received_load = 0.0;
+
+ idle_duration = 0.0;
+ convergence = -1.0;
+
+ proc_mutex->acquire();
+ process_counter++;
+ convergence_counter++;
+ total_load_init += real_load;
+ total_load_running += real_load;
+ total_load_average = total_load_running / process_counter;
+ if (opt::avg_load_ratio >= 0.0)
+ average_load_ratio = opt::avg_load_ratio;
+ else
+ average_load_ratio = 100.0 *
+ (process_counter / -opt::avg_load_ratio) / total_load_average;
+ load_diff_threshold = (opt::load_ratio_threshold +
+ average_load_ratio * total_load_average) / 100.0;
+ proc_mutex->release();
+
+ ctrl_close_pending = data_close_pending = neigh.size();
+ close_received = false;
+ finalizing = false;
+
+ all_comp_iter = comp_iter = lb_iter = 0;
+
+ lb_thread = new_msg_thread("loba", [this]() { this->load_balance_loop(); });
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()) {
oss << ESSE(neigh.size()) << ": ";
std::transform(neigh.begin(), neigh.end() - 1,
std::ostream_iterator<const char*>(oss, ", "),
- std::tr1::mem_fn(&neighbor::get_name));
+ [](const neighbor& neigh) { return neigh.get_name(); });
oss << neigh.back().get_name();
}
- LOG1(logp, "Got %s.", oss.str().c_str());
- print_loads(logp);
+ XBT_LOG(logp, "Got %s.", oss.str().c_str());
+ print_loads(false, logp);
+}
+
+process::~process()
+{
+ delete lb_thread;
+ proc_mutex->acquire();
+ total_load_exit += real_load;
+ proc_mutex->release();
+ xbt_assert(received_load == 0.0,
+ "received_load is %g, but should be 0.0 !", received_load);
+ if (opt::log_rate < 0)
+ return;
+ 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",
+ average_load_ratio, convergence);
+ else
+ XBT_INFO("Convergence within %g%% was not achieved",
+ average_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);
+}
+
+double process::get_iter_deviation() const
+{
+ double average_cost = opt::comp_cost(total_load_average); // fixme: get locked?
+ // Do not count idle periods
+ double comp_iter_opt = acc.comp_amount / average_cost;
+/*
+ // Add iterations that could have been achieved while beeing idle
+ // (kept for documentation)
+ double self_speed = MSG_get_host_speed(MSG_host_self());
+ double average_duration = average_cost / self_speed;
+ comp_iter_opt += idle_duration / average_duration;
+*/
+ return comp_iter - comp_iter_opt;
}
int process::run()
{
- bool one_more = true;
- unsigned iter = 0;
- VERB0("Starting...");
- while (one_more) {
- bool close_received;
+ 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;
+}
- if (opt::bookkeeping)
- INFO3("(%u) current load: %g ; expected: %g",
- iter, load, expected_load);
- else
- INFO2("(%u) current load: %g",
- iter, load);
+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);
- compute();
- close_received = !receive(false);
+ mutex.acquire();
+ if (!opt::bookkeeping)
+ expected_load = real_load - get_sum_of_to_send();
+ // nothing to do with opt::bookkeeping
- /*
- * compute load balancing;
- * send tasks to neighbors;
- */
+ 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();
- comm.flush(false);
- ++iter;
+ print_loads(true, xbt_log_priority_debug);
- if (opt::exit_on_close && close_received)
- one_more = false;
- if (opt::maxiter && iter >= opt::maxiter)
- one_more = false;
+ // 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);
}
- VERB0("Going to finalize...");
- finalize();
- /* Open Questions :
- * - definition of load on heterogeneous hosts ?
- * - how to detect convergence ?
- * - how to manage link failures ?
- */
+ 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);
+}
- VERB0("Done.");
- return 0;
+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::compute()
+void process::convergence_check()
{
- // fixme: shall we do something special when duration is 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);
+ 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;
-// Returns false if a CLOSE message was received.
-bool process::receive(bool wait_for_close)
+ } 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
{
- bool result = true;
- message* msg;
- m_host_t from;
- while ((ctrl_close_pending ||
- data_close_pending) && comm.recv(msg, from, wait_for_close)) {
- DEBUG2("received %s from %s",
- msg->to_string().c_str(), MSG_host_get_name(from));
- switch (msg->get_type()) {
- case message::INFO:
- // fixme: update neighbor
- // need a map m_host_t -> neighbor&
- break;
- case message::CREDIT:
- expected_load += msg->get_amount();
- break;
- case message::LOAD:
- load += msg->get_amount();
- break;
- case message::CTRL_CLOSE:
- if (--ctrl_close_pending == 1)
- comm.next_close_on_ctrl_is_last();
- DEBUG1("ctrl_close_pending = %d", ctrl_close_pending);
- result = false;
- break;
- case message::DATA_CLOSE:
- if (--data_close_pending == 1)
- comm.next_close_on_data_is_last();
- DEBUG1("data_close_pending = %d", data_close_pending);
- result = false;
- break;
+ 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);
}
- delete msg;
+ } 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;
+}
+
+void process::data_send(neighbor& nb)
+{
+ double load_to_send;
+ if (opt::bookkeeping) { // bookkeeping
+ 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;
+
+ double balance;
+ if (nb.get_credit() > 0.0)
+ balance = nb.get_debt() - nb.get_credit();
+ else
+ balance = nb.get_debt();
+ 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
+ load_to_send = compute_load_to_send(nb.get_to_send());
+ if (load_to_send > 0.0)
+ nb.set_to_send(nb.get_to_send() - load_to_send);
}
- return result;
+ real_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::DATA, amount);
+ add_data_send_mesg(msg->get_size());
+ comm.data_send(nb.get_data_mbox(), msg);
+ load_to_send -= amount;
+ }
+}
+
+void process::ctrl_close(neighbor& nb)
+{
+ comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
+}
+
+void process::data_close(neighbor& nb)
+{
+ comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
}
-void process::finalize()
+void process::ctrl_receive(double timeout)
{
- DEBUG2("send CLOSE to %d neighbor%s.",
- (int )neigh.size(), ESSE(neigh.size()));
- std::vector<neighbor>::iterator n;
- for (n = neigh.begin() ; n != neigh.end() ; ++n) {
- comm.send(n->get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
- comm.send(n->get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
+ message* msg;
+ msg_host_t from;
+
+ 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;
}
+}
- DEBUG2("wait for CLOSE from %d neighbor%s.",
- (int )neigh.size(), ESSE(neigh.size()));
- receive(true);
+void process::data_receive(double timeout)
+{
+ message* msg;
+ msg_host_t from;
- comm.flush(true);
+ 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::print_loads(e_xbt_log_priority_t logp)
+void process::handle_message(message* msg, msg_host_t from)
{
- if (!LOG_ISENABLED(logp))
+ switch (msg->get_type()) {
+ case message::CTRL: {
+ neighbor* n = rev_neigh[from];
+ n->set_load(msg->get_amount() + n->get_to_send());
+ if (opt::bookkeeping) {
+ double credit = msg->get_credit();
+ expected_load += credit;
+ n->set_credit(n->get_credit() + credit);
+ }
+ break;
+ }
+ case message::DATA: {
+ neighbor* n = rev_neigh[from];
+ double ld = msg->get_amount();
+ received_load += ld;
+ n->set_credit(n->get_credit() - 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;
+}
+
+void process::print_loads(bool verbose,
+ e_xbt_log_priority_t logp,
+ xbt_log_category_t cat) const
+{
+ if (!_XBT_LOG_ISENABLEDV((*cat), logp))
return;
+ XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected). Neighbor loads:",
+ real_load, expected_load);
+ for (const neighbor& n : neigh)
+ n.print(verbose, logp, cat);
+}
- std::ostringstream oss;
- if (neigh.empty()) {
- oss << "no neighbor!";
- } else {
- std::transform(neigh.begin(), neigh.end() - 1,
- std::ostream_iterator<double>(oss, ", "),
- std::tr1::mem_fn(&neighbor::get_load));
- oss << neigh.back().get_load();
- }
- LOG1(logp, "Neighbor loads: %s", oss.str().c_str());
+void process::print_loads_p(bool verbose,
+ e_xbt_log_priority_t logp,
+ xbt_log_category_t cat) const
+{
+ if (!_XBT_LOG_ISENABLEDV((*cat), logp))
+ return;
+ XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected). Neighbor loads:",
+ real_load, expected_load);
+ for (const neighbor* n : pneigh)
+ n->print(verbose, logp, cat);
}
+#undef print_loads_generic
+
// Local variables:
// mode: c++
// End:
