process.cpp

   1 #include <algorithm>
   2 #include <cmath>
   3 #include <functional>
   4 #include <iterator>
   5 #include <numeric>
   6 #include <stdexcept>
   7 #include <sstream>
   8 #include <xbt/log.h>
   9 #include <xbt/time.h>
  10
  11 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
  12
  13 #include "misc.h"
  14 #include "options.h"
  15 #include "tracing.h"
  16
  17 #include "process.h"
  18
  19 double process::total_load_init = 0.0;
  20 double process::total_load_running = 0.0;
  21 double process::total_load_exit = 0.0;
  22
  23 namespace {
  24
  25     void sleep_until_date(double& date, double duration)
  26     {
  27         double sleep_duration = date - MSG_get_clock();
  28         if (sleep_duration > 0.0)
  29             MSG_process_sleep(sleep_duration);
  30         date = MSG_get_clock() + duration;
  31     }
  32
  33 }
  34
  35 process::process(int argc, char* argv[])
  36 {
  37     if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))
  38         throw std::invalid_argument("bad or missing initial load parameter");
  39
  40     neigh.assign(argv + 2, argv + argc);
  41
  42     pneigh.reserve(neigh.size());
  43     for (unsigned i = 0 ; i < neigh.size() ; i++) {
  44         neighbor* ptr = &neigh[i];
  45         m_host_t host = MSG_get_host_by_name(ptr->get_name());
  46         pneigh.push_back(ptr);
  47         rev_neigh.insert(std::make_pair(host, ptr));
  48     }
  49
  50     prev_load_broadcast = -1;   // force sending of load on first send_all()
  51     expected_load = real_load;
  52     total_load_running += real_load;
  53     total_load_init += real_load;
  54     received_load = 0.0;
  55
  56     ctrl_close_pending = data_close_pending = neigh.size();
  57     close_received = false;
  58     finalizing = false;
  59
  60     comp_iter = lb_iter = 0;
  61
  62     lb_thread = new_msg_thread("loba",
  63                                std::bind(&process::load_balance_loop, this));
  64
  65     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  66     if (!LOG_ISENABLED(logp))
  67         return;
  68     std::ostringstream oss;
  69     oss << neigh.size() << " neighbor";
  70     if (!neigh.empty()) {
  71         oss << ESSE(neigh.size()) << ": ";
  72         std::transform(neigh.begin(), neigh.end() - 1,
  73                        std::ostream_iterator<const char*>(oss, ", "),
  74                        std::mem_fn(&neighbor::get_name));
  75         oss << neigh.back().get_name();
  76     }
  77     XBT_LOG(logp, "Got %s.", oss.str().c_str());
  78     print_loads(false, logp);
  79 }
  80
  81 process::~process()
  82 {
  83     delete lb_thread;
  84     total_load_exit += real_load;
  85     xbt_assert(received_load == 0.0,
  86                "received_load is %g, but should be 0.0 !", received_load);
  87     if (opt::log_rate < 0)
  88         return;
  89     XBT_INFO("Final load after %d:%d iterations: %g",
  90              lb_iter, comp_iter, real_load);
  91     XBT_VERB("Expected load was: %g", expected_load);
  92     XBT_VERB("Total computation for this process: %g", get_comp_amount());
  93 }
  94
  95 int process::run()
  96 {
  97     if (opt::log_rate >= 0) {
  98         XBT_INFO("Initial load: %g", real_load);
  99         XBT_VERB("Initial expected load: %g", expected_load);
 100     }
 101     XBT_VERB("Starting...");
 102     mutex.acquire();
 103     lb_thread->start();
 104     while (lb_iter <= opt::comp_iter_delay)
 105         cond.wait(mutex);
 106     mutex.release();
 107     double sleep_duration = opt::comp_time_delay - MSG_get_clock();
 108     if (sleep_duration > 0.0)
 109         MSG_process_sleep(sleep_duration);
 110     compute_loop();
 111     lb_thread->wait();
 112     XBT_VERB("Done.");
 113     return 0;
 114 }
 115
 116 void process::load_balance_loop()
 117 {
 118     using std::placeholders::_1;
 119
 120     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 121     while (still_running()) {
 122         if (lb_iter == opt::comp_iter_delay) {
 123             mutex.acquire();
 124             ++lb_iter;
 125             cond.signal();
 126             mutex.release();
 127         } else {
 128             ++lb_iter;
 129         }
 130
 131         ctrl_receive(0.0);
 132
 133         mutex.acquire();
 134         if (!opt::bookkeeping)
 135             expected_load = real_load - get_sum_of_to_send();
 136         // nothing to do with opt::bookkeeping
 137
 138         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 139             XBT_INFO("(%u:%u) current load: %g", lb_iter, comp_iter, real_load);
 140             XBT_VERB("... expected load: %g", expected_load);
 141         }
 142
 143         if (expected_load > 0.0)
 144             load_balance();
 145
 146         print_loads(true, xbt_log_priority_debug);
 147
 148         // send
 149         comm.ctrl_flush(false);
 150         std::for_each(neigh.begin(), neigh.end(),
 151                       std::bind(&process::ctrl_send, this, _1));
 152         prev_load_broadcast = expected_load;
 153         mutex.release();
 154
 155         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 156     }
 157
 158     XBT_VERB("Going to finalize for %s...", __func__);
 159     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 160               neigh.size(), ESSE(neigh.size()));
 161     std::for_each(neigh.begin(), neigh.end(),
 162                   std::bind(&process::ctrl_close, this, _1));
 163     while (ctrl_close_pending) {
 164         comm.ctrl_flush(false);
 165         XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);
 166         ctrl_receive(-1.0);
 167     }
 168     comm.ctrl_flush(true);
 169 }
 170
 171 void process::compute_loop()
 172 {
 173     using std::placeholders::_1;
 174
 175     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 176     while (still_running()) {
 177         // receive (do not block if there is something to compute)
 178         data_receive(real_load > 0.0 ? 0.0 : opt::min_comp_iter_duration);
 179
 180         // send
 181         comm.data_flush(false);
 182         mutex.acquire();
 183         real_load += received_load;
 184         received_load = 0.0;
 185         std::for_each(neigh.begin(), neigh.end(),
 186                       std::bind(&process::data_send, this, _1));
 187         mutex.release();
 188
 189         if (real_load == 0.0)
 190             continue;
 191
 192         // compute
 193         ++comp_iter;
 194         double flops = opt::comp_cost(real_load);
 195         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 196         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 197         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 198         MSG_task_execute(task);
 199         add_comp_amount(flops);
 200         MSG_task_destroy(task);
 201
 202         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 203     }
 204
 205     XBT_VERB("Going to finalize for %s...", __func__);
 206     finalizing = true;
 207     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 208               neigh.size(), ESSE(neigh.size()));
 209     std::for_each(neigh.begin(), neigh.end(),
 210                   std::bind(&process::data_close, this, _1));
 211     while (data_close_pending) {
 212         comm.data_flush(false);
 213         XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);
 214         data_receive(-1.0);
 215     }
 216     real_load += received_load;
 217     received_load = 0.0;
 218     total_load_running -= real_load;
 219     comm.data_flush(true);
 220 }
 221
 222 bool process::still_running()
 223 {
 224     static bool last_status = true;
 225
 226     if (!last_status) {
 227         /* nop */
 228
 229     } else if (opt::exit_request) {
 230         XBT_VERB("Global exit requested");
 231         last_status = false;
 232
 233     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 234         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 235         last_status = false;
 236
 237     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 238         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 239         last_status = false;
 240
 241     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 242         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 243         last_status = false;
 244
 245     } else if (opt::exit_on_close && close_received) {
 246         XBT_VERB("Close received");
 247         last_status = false;
 248
 249     } else if (real_load == 0.0 && !data_close_pending) {
 250         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 251         last_status = false;
 252
 253     } else if (100.0 * total_load_running / total_load_init <=
 254                opt::load_ratio_threshold) {
 255         // fixme: this check should be implemented with a distributed
 256         // algorithm, and not a shared global variable!
 257         XBT_VERB("No more load to balance in system.");
 258         last_status = false;
 259     }
 260
 261     return last_status;
 262 }
 263
 264 double process::get_sum_of_to_send() const
 265 {
 266     using std::placeholders::_1;
 267     using std::placeholders::_2;
 268
 269     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 270                            std::bind(std::plus<double>(), _1,
 271                                      std::bind(&neighbor::get_to_send, _2)));
 272 }
 273
 274 void process::load_balance()
 275 {
 276     if (lb_iter == 1)           // warn only once
 277         XBT_WARN("process::load_balance() is a no-op!");
 278 }
 279
 280 void process::send(neighbor& nb, double amount)
 281 {
 282     expected_load -= amount;
 283     nb.set_to_send(nb.get_to_send() + amount);
 284     nb.set_load(nb.get_load() + amount);
 285 }
 286
 287 void process::ctrl_send(neighbor& nb)
 288 {
 289     double info_to_send = expected_load;
 290     double debt_to_send;
 291     if (opt::bookkeeping) {     // bookkeeping
 292         debt_to_send = nb.get_to_send();
 293         if (debt_to_send > 0.0) {
 294             nb.set_to_send(0.0);
 295             nb.set_debt(nb.get_debt() + debt_to_send);
 296         }
 297     } else {                    // !bookkeeping
 298         debt_to_send = 0.0;
 299     }
 300     if (info_to_send != prev_load_broadcast || debt_to_send > 0.0) {
 301         message* msg = new message(message::CTRL, info_to_send, debt_to_send);
 302         add_ctrl_send_mesg(msg->get_size());
 303         comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 304     }
 305 }
 306
 307 double process::compute_load_to_send(double desired)
 308 {
 309     if (opt::integer_transfer)
 310         desired = std::floor(desired);
 311     return desired >= opt::min_transfer_amount ? desired : 0.0;
 312 }
 313
 314 void process::data_send(neighbor& nb)
 315 {
 316     double load_to_send;
 317     if (opt::bookkeeping) {
 318         load_to_send = compute_load_to_send(std::min(real_load, nb.get_debt()));
 319         if (load_to_send > 0.0)
 320             nb.set_debt(nb.get_debt() - load_to_send);
 321     } else {
 322         load_to_send = compute_load_to_send(nb.get_to_send());
 323         if (load_to_send > 0.0)
 324             nb.set_to_send(nb.get_to_send() - load_to_send);
 325     }
 326     real_load -= load_to_send;
 327     while (load_to_send > 0.0) {
 328         double amount;
 329         if (opt::max_transfer_amount)
 330             amount = std::min(load_to_send, opt::max_transfer_amount);
 331         else
 332             amount = load_to_send;
 333         message* msg = new message(message::DATA, amount);
 334         add_data_send_mesg(msg->get_size());
 335         comm.data_send(nb.get_data_mbox(), msg);
 336         load_to_send -= amount;
 337     }
 338 }
 339
 340 void process::ctrl_close(neighbor& nb)
 341 {
 342     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 343 }
 344
 345 void process::data_close(neighbor& nb)
 346 {
 347     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 348 }
 349
 350 void process::ctrl_receive(double timeout)
 351 {
 352     message* msg;
 353     m_host_t from;
 354
 355     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 356     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 357         if (msg->get_type() != message::CTRL_CLOSE)
 358             add_ctrl_recv_mesg(msg->get_size());
 359         handle_message(msg, from);
 360         timeout = 0.0;
 361     }
 362 }
 363
 364 void process::data_receive(double timeout)
 365 {
 366     message* msg;
 367     m_host_t from;
 368
 369     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 370     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 371         if (msg->get_type() != message::DATA_CLOSE)
 372             add_data_recv_mesg(msg->get_size());
 373         handle_message(msg, from);
 374         timeout = 0.0;
 375     }
 376 }
 377
 378 void process::handle_message(message* msg, m_host_t from)
 379 {
 380     switch (msg->get_type()) {
 381     case message::CTRL: {
 382         neighbor* n = rev_neigh[from];
 383         n->set_load(msg->get_amount() + n->get_to_send());
 384         expected_load += msg->get_credit(); // may be 0.0 if !opt::bookkeeping
 385         break;
 386     }
 387     case message::DATA: {
 388         double ld = msg->get_amount();
 389         received_load += ld;
 390         break;
 391     }
 392     case message::CTRL_CLOSE:
 393         ctrl_close_pending--;
 394         close_received = true;
 395         break;
 396     case message::DATA_CLOSE:
 397         data_close_pending--;
 398         close_received = true;
 399         break;
 400     }
 401     delete msg;
 402 }
 403
 404 #define print_loads_generic(vec, verbose, logp, cat)                    \
 405     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 406         using std::placeholders::_1;                                    \
 407         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 408                   "Neighbor loads:", real_load, expected_load);         \
 409         std::for_each(vec.begin(), vec.end(),                           \
 410                       std::bind(&neighbor::print, _1, verbose, logp, cat)); \
 411     } else ((void)0)
 412
 413 void process::print_loads(bool verbose,
 414                           e_xbt_log_priority_t logp,
 415                           xbt_log_category_t cat) const
 416 {
 417     print_loads_generic(neigh, verbose, logp, cat);
 418 }
 419
 420 void process::print_loads_p(bool verbose,
 421                             e_xbt_log_priority_t logp,
 422                             xbt_log_category_t cat) const
 423 {
 424     print_loads_generic(pneigh, verbose, logp, cat);
 425 }
 426
 427 #undef print_loads_generic
 428
 429 // Local variables:
 430 // mode: c++
 431 // End: