#include <algorithm>
-#include <functional>
+#include <tr1/functional>
#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 "process.h"
+#include "tracing.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))
- 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);
+
+ 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;
+
+ comp_iter = lb_iter = 0;
+
+ compute_thread = new_msg_thread("compute",
+ std::tr1::bind(&process::compute_loop,
+ this));
+
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 << (neigh.size() > 1 ? "s: " : ": ");
+ oss << ESSE(neigh.size()) << ": ";
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();
+ }
+ XBT_LOG(logp, "Got %s.", oss.str().c_str());
+ print_loads(false, logp);
+}
+
+process::~process()
+{
+ delete compute_thread;
+ total_load_exit += real_load;
+ if (opt::bookkeeping) {
+ XBT_INFO("Final load after %d:%d iterations: %g ; expected: %g",
+ lb_iter, comp_iter, real_load, expected_load);
+ } else {
+ XBT_INFO("Final load after %d iterations: %g",
+ lb_iter, real_load);
+ if (lb_iter != comp_iter)
+ XBT_WARN("lb_iter (%d) and comp_iter (%d) differ!",
+ lb_iter, comp_iter);
}
- LOG1(logp, "Got %s.", oss.str().c_str());
- print_loads(logp);
+ XBT_VERB("Total computation for this process: %g", comp);
}
int process::run()
{
- INFO0("Coucou !");
+ XBT_INFO("Initial load: %g", real_load);
+ XBT_VERB("Starting...");
+ compute_thread->start();
+ load_balance_loop();
+ compute_thread->wait();
+ XBT_VERB("Done.");
+ return 0;
+}
+
+void process::load_balance_loop()
+{
+ using std::tr1::bind;
+ using std::tr1::placeholders::_1;
+
+ while (still_running()) {
+ ++lb_iter;
- int n = 100;
- while (n--) {
- compute();
- receive();
+ if (opt::log_rate && lb_iter % opt::log_rate == 0) {
+ if (opt::bookkeeping)
+ XBT_INFO("(%u:%u) current load: %g ; expected: %g",
+ lb_iter, comp_iter, real_load, expected_load);
+ else
+ XBT_INFO("(%u:%u) current load: %g",
+ lb_iter, comp_iter, real_load);
+ }
+
+ if (get_load() > 0.0)
+ load_balance();
+
+ print_loads(true, xbt_log_priority_debug);
+
+ // send
+ std::for_each(neigh.begin(), neigh.end(),
+ bind(&process::ctrl_send, this, _1));
+
+ // block on receiving unless there is something to compute or
+ // to send
+ ctrl_receive(opt::min_lb_iter_duration);
+
+ comm.ctrl_flush(false);
}
- // 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;
- * }
- */
-
- /* Open Questions :
- * - definition of load on heterogeneous hosts ?
- * - how to detect convergence ?
- * - how to manage link failures ?
- */
-
- // xxx: shall we retrieve pending tasks?
- return 0;
+ XBT_VERB("Going to finalize for %s...", __func__);
+ XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
+ neigh.size(), ESSE(neigh.size()));
+ std::for_each(neigh.begin(), neigh.end(),
+ bind(&process::ctrl_close, this, _1));
+ 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::receive()
+void process::compute_loop()
{
- m_task_t task;
- while ((task = comm.recv())) {
- message *msg = (message *)MSG_task_get_data(task);
- DEBUG3("Received load: %g, info: %g from %s",
- msg->transfer, msg->measure,
- MSG_host_get_name(MSG_task_get_source(task)));
- load += msg->transfer;
- // fixme: what is xxx ???
- // neigh[xxx].setLoad(msg->measure);
+ using std::tr1::bind;
+ using std::tr1::placeholders::_1;
+
+ double next_iter_after_date = 0.0;
+ while (still_running()) {
+ // receive
+ double sleep_duration = real_load
+ ? std::max(MSG_get_clock() - next_iter_after_date, 0.0)
+ : opt::min_comp_iter_duration;
+ data_receive(sleep_duration);
+
+ comm.data_flush(false);
+
+ if (real_load == 0.0)
+ continue;
+
+ // send
+ std::for_each(neigh.begin(), neigh.end(),
+ bind(&process::data_send, this, _1));
+
+ // compute
+ ++comp_iter;
+ 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);
+ XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
+ MSG_task_execute(task);
+ comp += flops;
+ MSG_task_destroy(task);
+
+ next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
+ }
+
+ XBT_VERB("Going to finalize for %s...", __func__);
+ XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
+ neigh.size(), ESSE(neigh.size()));
+ std::for_each(neigh.begin(), neigh.end(),
+ bind(&process::data_close, this, _1));
+ while (data_close_pending) {
+ comm.data_flush(false);
+ XBT_DEBUG("waiting for %d DATA CLOSE", data_close_pending);
+ data_receive(-1.0);
}
+ comm.data_flush(true);
}
-void process::compute()
+bool process::still_running()
{
- 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);
+ static bool last_status = true;
+
+ if (!last_status) {
+ /* nop */
+
+ } 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_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: 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() ?
+ XBT_VERB("No more load to balance in system.");
+ last_status = false;
+ }
+
+ return last_status;
}
-void process::print_loads(e_xbt_log_priority_t logp)
+void process::load_balance()
{
- if (!LOG_ISENABLED(logp))
- return;
- std::ostringstream oss;
- if (neigh.empty()) {
- oss << "no neighbor!";
+ if (lb_iter == 1) // warn only once
+ XBT_WARN("process::load_balance() is a no-op!");
+}
+
+void process::send(neighbor& nb, double amount)
+{
+ set_load(get_load() - amount);
+ nb.set_to_send(nb.get_to_send() + amount);
+ nb.set_load(nb.get_load() + amount);
+}
+
+#define print_loads_generic(vec, verbose, logp, cat) \
+ if (_XBT_LOG_ISENABLEDV((*cat), logp)) { \
+ using std::tr1::bind; \
+ using std::tr1::placeholders::_1; \
+ XBT_XCLOG(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
+
+void process::ctrl_send(neighbor& nb)
+{
+ double info_to_send = get_load();
+ if (info_to_send != prev_load_broadcast)
+ comm.ctrl_send(nb.get_ctrl_mbox(),
+ new message(message::INFO, info_to_send));
+ if (opt::bookkeeping) {
+ double 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);
+ comm.ctrl_send(nb.get_ctrl_mbox(),
+ new message(message::CREDIT, debt_to_send));
+ }
+ }
+}
+
+void process::data_send(neighbor& nb)
+{
+ double load_to_send;
+ if (opt::bookkeeping) {
+ if (real_load <= nb.get_debt()) {
+ load_to_send = real_load;
+ nb.set_debt(nb.get_debt() - load_to_send);
+ real_load = 0.0;
+ } else {
+ load_to_send = nb.get_debt();
+ nb.set_debt(0.0);
+ real_load -= load_to_send;
+ }
} else {
- std::transform(neigh.begin(), neigh.end() - 1,
- std::ostream_iterator<double>(oss, ", "),
- std::mem_fun_ref(&neighbor::getLoad));
- oss << neigh.back().getLoad();
+ load_to_send = nb.get_to_send();
+ nb.set_to_send(0.0);
+ }
+ if (load_to_send > 0.0)
+ comm.data_send(nb.get_data_mbox(),
+ new message(message::LOAD, load_to_send));
+}
+
+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::ctrl_receive(double timeout)
+{
+ message* msg;
+ m_host_t from;
+
+ XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
+ while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
+ handle_message(msg, from);
+ timeout = 0.0;
+ }
+}
+
+void process::data_receive(double timeout)
+{
+ message* msg;
+ m_host_t from;
+
+ XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
+ while (data_close_pending && comm.data_recv(msg, from, timeout)) {
+ handle_message(msg, from);
+ timeout = 0.0;
+ }
+}
+
+void process::handle_message(message* msg, m_host_t from)
+{
+ 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();
+ 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;
}
- LOG1(logp, "Neighbor loads: %s", oss.str().c_str());
+ delete msg;
}
// Local variables:
