process.cpp

   1 #include <algorithm>
   2 #include <cmath>
   3 #include <iterator>
   4 #include <numeric>
   5 #include <stdexcept>
   6 #include <sstream>
   7 #include <xbt/log.h>
   8
   9 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
  10
  11 #include "misc.h"
  12 #include "options.h"
  13 #include "tracing.h"
  14
  15 #include "process.h"
  16
  17 mutex_t *process::proc_mutex;
  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 int process::process_counter = 0;
  24 double process::total_load_average;
  25 double process::average_load_ratio;
  26 double process::load_diff_threshold;
  27
  28 std::atomic<int> process::convergence_counter(0);
  29
  30 namespace {
  31
  32     void sleep_until_date(double& date, double duration)
  33     {
  34         double sleep_duration = date - MSG_get_clock();
  35         if (sleep_duration > 0.0)
  36             MSG_process_sleep(sleep_duration);
  37         date = MSG_get_clock() + duration;
  38     }
  39
  40 }
  41
  42 process::process(int argc, char* argv[])
  43 {
  44     if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))
  45         throw std::invalid_argument("bad or missing initial load parameter");
  46
  47     double iload = std::trunc(real_load);
  48     if (opt::integer_transfer && real_load != iload) {
  49         XBT_WARN("Initial load %g is not an integer.  Truncate it.",
  50                  real_load);
  51         real_load = iload;
  52     }
  53
  54     neigh.assign(argv + 2, argv + argc);
  55
  56     pneigh.reserve(neigh.size());
  57     for (unsigned i = 0 ; i < neigh.size() ; i++) {
  58         neighbor* ptr = &neigh[i];
  59         msg_host_t host = MSG_get_host_by_name(ptr->get_name());
  60         pneigh.push_back(ptr);
  61         rev_neigh.insert(std::make_pair(host, ptr));
  62     }
  63
  64     prev_load_broadcast = -1;   // force sending of load on first send_all()
  65     expected_load = real_load;
  66     received_load = 0.0;
  67
  68     idle_duration = 0.0;
  69     convergence = -1.0;
  70
  71     proc_mutex->acquire();
  72     process_counter++;
  73     convergence_counter++;
  74     total_load_init += real_load;
  75     total_load_running += real_load;
  76     total_load_average = total_load_running / process_counter;
  77     if (opt::avg_load_ratio >= 0.0)
  78         average_load_ratio = opt::avg_load_ratio;
  79     else
  80         average_load_ratio = 100.0 *
  81             (process_counter / -opt::avg_load_ratio) / total_load_average;
  82     load_diff_threshold = (opt::load_ratio_threshold +
  83                            average_load_ratio * total_load_average) / 100.0;
  84     proc_mutex->release();
  85
  86     ctrl_close_pending = data_close_pending = neigh.size();
  87     close_received = false;
  88     finalizing = false;
  89
  90     all_comp_iter = comp_iter = lb_iter = 0;
  91
  92     lb_thread = new_msg_thread("loba", [this]() { this->load_balance_loop(); });
  93
  94     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  95     if (!LOG_ISENABLED(logp))
  96         return;
  97     std::ostringstream oss;
  98     oss << neigh.size() << " neighbor";
  99     if (!neigh.empty()) {
 100         oss << ESSE(neigh.size()) << ": ";
 101         std::transform(neigh.begin(), neigh.end() - 1,
 102                        std::ostream_iterator<const char*>(oss, ", "),
 103                        [](const neighbor& neigh) { return neigh.get_name(); });
 104         oss << neigh.back().get_name();
 105     }
 106     XBT_LOG(logp, "Got %s.", oss.str().c_str());
 107     print_loads(false, logp);
 108 }
 109
 110 process::~process()
 111 {
 112     delete lb_thread;
 113     proc_mutex->acquire();
 114     total_load_exit += real_load;
 115     proc_mutex->release();
 116     xbt_assert(received_load == 0.0,
 117                "received_load is %g, but should be 0.0 !", received_load);
 118     if (opt::log_rate < 0)
 119         return;
 120     XBT_INFO("Final load after %d:%d:%d iterations: %g",
 121              lb_iter, comp_iter, all_comp_iter, real_load);
 122     if (convergence >= 0.0)
 123         XBT_INFO("Convergence within %g%% was achieved at time %g",
 124                  average_load_ratio, convergence);
 125     else
 126         XBT_INFO("Convergence within %g%% was not achieved",
 127                  average_load_ratio);
 128     XBT_VERB("Expected load was: %g", expected_load);
 129     XBT_VERB("Total computation for this process: %g", get_comp_amount());
 130     print_loads(true, xbt_log_priority_debug);
 131 }
 132
 133 double process::get_iter_deviation() const
 134 {
 135     double average_cost = opt::comp_cost(total_load_average); // fixme: get locked?
 136     // Do not count idle periods
 137     double comp_iter_opt = acc.comp_amount / average_cost;
 138 /*
 139     // Add iterations that could have been achieved while beeing idle
 140     // (kept for documentation)
 141     double self_speed = MSG_get_host_speed(MSG_host_self());
 142     double average_duration = average_cost / self_speed;
 143     comp_iter_opt += idle_duration / average_duration;
 144 */
 145     return comp_iter - comp_iter_opt;
 146 }
 147
 148 int process::run()
 149 {
 150     if (opt::log_rate >= 0) {
 151         XBT_INFO("Initial load: %g", real_load);
 152         XBT_VERB("Initial expected load: %g", expected_load);
 153     }
 154     XBT_VERB("Starting...");
 155     mutex.acquire();
 156     lb_thread->start();
 157     while (lb_iter <= opt::comp_iter_delay)
 158         cond.wait(mutex);
 159     mutex.release();
 160     double sleep_duration = opt::comp_time_delay - MSG_get_clock();
 161     if (sleep_duration > 0.0)
 162         MSG_process_sleep(sleep_duration);
 163     compute_loop();
 164     lb_thread->wait();
 165     XBT_VERB("Done.");
 166     return 0;
 167 }
 168
 169 void process::load_balance_loop()
 170 {
 171     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 172     while (still_running()) {
 173         if (lb_iter == opt::comp_iter_delay) {
 174             mutex.acquire();
 175             ++lb_iter;
 176             cond.signal();
 177             mutex.release();
 178         } else {
 179             ++lb_iter;
 180         }
 181
 182         ctrl_receive(0.0);
 183
 184         mutex.acquire();
 185         if (!opt::bookkeeping)
 186             expected_load = real_load - get_sum_of_to_send();
 187         // nothing to do with opt::bookkeeping
 188
 189         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 190             XBT_INFO("(%u:%u:%u) current load: %g",
 191                      lb_iter, comp_iter, all_comp_iter, real_load);
 192             XBT_VERB("... expected load: %g", expected_load);
 193         }
 194
 195         if (expected_load > 0.0)
 196             load_balance();
 197
 198         print_loads(true, xbt_log_priority_debug);
 199
 200         // send
 201         comm.ctrl_flush(false);
 202         for (neighbor& n : neigh)
 203             ctrl_send(n);
 204         prev_load_broadcast = expected_load;
 205         mutex.release();
 206
 207         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 208     }
 209
 210     XBT_VERB("Going to finalize for %s...", __func__);
 211     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 212               neigh.size(), ESSE(neigh.size()));
 213     for (neighbor& n : neigh)
 214         ctrl_close(n);
 215     while (ctrl_close_pending) {
 216         comm.ctrl_flush(false);
 217         XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);
 218         ctrl_receive(-1.0);
 219     }
 220     comm.ctrl_flush(true);
 221 }
 222
 223 void process::compute_loop()
 224 {
 225     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 226     double idle_since_date = 0.0;
 227     while (still_running()) {
 228         // receive
 229         // if there is something to compute, do not block
 230         // else, block the duration of an *lb* iteration
 231         data_receive(real_load > 0.0 ? 0.0 : opt::min_lb_iter_duration);
 232
 233         // send
 234         comm.data_flush(false);
 235         mutex.acquire();
 236         real_load += received_load;
 237         received_load = 0.0;
 238         for (neighbor& n : neigh)
 239             data_send(n);
 240         mutex.release();
 241
 242         ++all_comp_iter;
 243         if (real_load == 0.0)
 244             continue;
 245
 246         convergence_check();
 247
 248         // compute
 249         idle_duration += MSG_get_clock() - idle_since_date;
 250         ++comp_iter;
 251         double flops = opt::comp_cost(real_load);
 252         msg_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 253         // MSG_task_set_category(task, TRACE_CAT_COMP);
 254         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 255         MSG_task_execute(task);
 256         add_comp_amount(flops);
 257         MSG_task_destroy(task);
 258
 259         idle_since_date = MSG_get_clock();
 260
 261         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 262     }
 263
 264     XBT_VERB("Going to finalize for %s...", __func__);
 265     // Note: idle duration is not counted during finalization
 266     finalizing = true;
 267     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 268               neigh.size(), ESSE(neigh.size()));
 269     for (neighbor& n : neigh)
 270         data_close(n);
 271     while (data_close_pending) {
 272         comm.data_flush(false);
 273         XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);
 274         data_receive(-1.0);
 275     }
 276     real_load += received_load;
 277     received_load = 0.0;
 278     proc_mutex->acquire();
 279     total_load_running -= real_load;
 280     proc_mutex->release();
 281     convergence_check();
 282     comm.data_flush(true);
 283 }
 284
 285 void process::convergence_check()
 286 {
 287     double average = total_load_average; // fixme: get locked?
 288     double load_diff = std::fabs(real_load - average);
 289     bool converged = load_diff <= load_diff_threshold;
 290
 291     if (converged) {
 292         if (convergence < 0) {
 293             XBT_VERB("current load has converged: %g (%.4g%%)",
 294                      real_load,  100.0 * load_diff / average);
 295             convergence = MSG_get_clock();
 296             local_convergence_counter = opt::exit_on_convergence;
 297         }
 298         if (local_convergence_counter > 0 && --local_convergence_counter == 0)
 299                 --convergence_counter;
 300     } else {
 301         if (convergence >= 0.0) {
 302             XBT_VERB("current load has diverged: %g (%.4g%%)",
 303                      real_load, 100.0 * load_diff / average);
 304             convergence = -1.0;
 305             if (local_convergence_counter == 0)
 306                 ++convergence_counter;
 307         }
 308     }
 309 }
 310
 311 bool process::still_running()
 312 {
 313     static bool last_status = true;
 314
 315     if (!last_status) {
 316         /* nop */
 317
 318     } else if (opt::exit_request) {
 319         XBT_VERB("Global exit requested");
 320         last_status = false;
 321
 322     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 323         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 324         last_status = false;
 325
 326     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 327         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 328         last_status = false;
 329
 330     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 331         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 332         last_status = false;
 333
 334     } else if (opt::exit_on_convergence && convergence_counter == 0) {
 335         XBT_VERB("Global convergence detected");
 336         last_status = false;
 337
 338     } else if (opt::exit_on_close && close_received) {
 339         XBT_VERB("Close received");
 340         last_status = false;
 341
 342     } else if (real_load == 0.0 && !data_close_pending) {
 343         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 344         last_status = false;
 345
 346     } else if (100.0 * total_load_running / total_load_init <=
 347                opt::load_ratio_threshold) { // fixme: get locked?
 348         // fixme: this check should be implemented with a distributed
 349         // algorithm, and not a shared global variable!
 350         XBT_VERB("No more load to balance in system.");
 351         last_status = false;
 352     }
 353
 354     return last_status;
 355 }
 356
 357 double process::get_sum_of_to_send() const
 358 {
 359     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 360                            [](double x, const neighbor& neigh) {
 361                                return x + neigh.get_to_send();
 362                            });
 363 }
 364
 365 void process::load_balance()
 366 {
 367     if (lb_iter == 1)           // warn only once
 368         XBT_WARN("process::load_balance() is a no-op!");
 369 }
 370
 371 void process::send(neighbor& nb, double amount)
 372 {
 373     expected_load -= amount;
 374     nb.set_to_send(nb.get_to_send() + amount);
 375     nb.set_load(nb.get_load() + amount);
 376 }
 377
 378 void process::ctrl_send(neighbor& nb)
 379 {
 380     double info_to_send = expected_load;
 381     double debt_to_send;
 382     if (opt::bookkeeping) {     // bookkeeping
 383         debt_to_send = nb.get_to_send();
 384         if (debt_to_send > 0.0) {
 385             nb.set_to_send(0.0);
 386             nb.set_debt(nb.get_debt() + debt_to_send);
 387         }
 388     } else {                    // !bookkeeping
 389         debt_to_send = 0.0;
 390     }
 391     if (info_to_send != prev_load_broadcast || debt_to_send > 0.0) {
 392         message* msg = new message(message::CTRL, info_to_send, debt_to_send);
 393         add_ctrl_send_mesg(msg->get_size());
 394         comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 395     }
 396 }
 397
 398 double process::compute_load_to_send(double desired)
 399 {
 400     if (opt::integer_transfer)
 401         desired = std::floor(desired);
 402     return desired >= opt::min_transfer_amount ? desired : 0.0;
 403 }
 404
 405 void process::data_send(neighbor& nb)
 406 {
 407     double load_to_send;
 408     if (opt::bookkeeping) {     // bookkeeping
 409         double excess_load;     // load amount we are able to send
 410         if (opt::egocentric)
 411             excess_load = std::max(0.0, real_load - expected_load);
 412         else
 413             excess_load = real_load;
 414
 415         double balance;
 416         if (nb.get_credit() > 0.0)
 417             balance = nb.get_debt() - nb.get_credit();
 418         else
 419             balance = nb.get_debt();
 420         load_to_send = std::min(excess_load,
 421                                 std::max(0.0, balance));
 422
 423         // adjust load to send (rounding, truncation, etc.)
 424         load_to_send = compute_load_to_send(load_to_send);
 425         if (load_to_send > 0.0)
 426             nb.set_debt(nb.get_debt() - load_to_send);
 427     } else {                    // !bookkeeping
 428         load_to_send = compute_load_to_send(nb.get_to_send());
 429         if (load_to_send > 0.0)
 430             nb.set_to_send(nb.get_to_send() - load_to_send);
 431     }
 432     real_load -= load_to_send;
 433     while (load_to_send > 0.0) {
 434         double amount;
 435         if (opt::max_transfer_amount)
 436             amount = std::min(load_to_send, opt::max_transfer_amount);
 437         else
 438             amount = load_to_send;
 439         message* msg = new message(message::DATA, amount);
 440         add_data_send_mesg(msg->get_size());
 441         comm.data_send(nb.get_data_mbox(), msg);
 442         load_to_send -= amount;
 443     }
 444 }
 445
 446 void process::ctrl_close(neighbor& nb)
 447 {
 448     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 449 }
 450
 451 void process::data_close(neighbor& nb)
 452 {
 453     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 454 }
 455
 456 void process::ctrl_receive(double timeout)
 457 {
 458     message* msg;
 459     msg_host_t from;
 460
 461     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 462     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 463         if (msg->get_type() != message::CTRL_CLOSE)
 464             add_ctrl_recv_mesg(msg->get_size());
 465         handle_message(msg, from);
 466         timeout = 0.0;
 467     }
 468 }
 469
 470 void process::data_receive(double timeout)
 471 {
 472     message* msg;
 473     msg_host_t from;
 474
 475     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 476     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 477         if (msg->get_type() != message::DATA_CLOSE)
 478             add_data_recv_mesg(msg->get_size());
 479         handle_message(msg, from);
 480         timeout = 0.0;
 481     }
 482 }
 483
 484 void process::handle_message(message* msg, msg_host_t from)
 485 {
 486     switch (msg->get_type()) {
 487     case message::CTRL: {
 488         neighbor* n = rev_neigh[from];
 489         n->set_load(msg->get_amount() + n->get_to_send());
 490         if (opt::bookkeeping) {
 491             double credit = msg->get_credit();
 492             expected_load += credit;
 493             n->set_credit(n->get_credit() + credit);
 494         }
 495         break;
 496     }
 497     case message::DATA: {
 498         neighbor* n = rev_neigh[from];
 499         double ld = msg->get_amount();
 500         received_load += ld;
 501         n->set_credit(n->get_credit() - ld);
 502         break;
 503     }
 504     case message::CTRL_CLOSE:
 505         ctrl_close_pending--;
 506         close_received = true;
 507         break;
 508     case message::DATA_CLOSE:
 509         data_close_pending--;
 510         close_received = true;
 511         break;
 512     }
 513     delete msg;
 514 }
 515
 516 void process::print_loads(bool verbose,
 517                           e_xbt_log_priority_t logp,
 518                           xbt_log_category_t cat) const
 519 {
 520     if (!_XBT_LOG_ISENABLEDV((*cat), logp))
 521         return;
 522     XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  Neighbor loads:",
 523               real_load, expected_load);
 524     for (const neighbor& n : neigh)
 525         n.print(verbose, logp, cat);
 526 }
 527
 528 void process::print_loads_p(bool verbose,
 529                             e_xbt_log_priority_t logp,
 530                             xbt_log_category_t cat) const
 531 {
 532     if (!_XBT_LOG_ISENABLEDV((*cat), logp))
 533         return;
 534     XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  Neighbor loads:",
 535               real_load, expected_load);
 536     for (const neighbor* n : pneigh)
 537         n->print(verbose, logp, cat);
 538 }
 539
 540 #undef print_loads_generic
 541
 542 // Local variables:
 543 // mode: c++
 544 // End: