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 mutex_t *process::proc_mutex;
  20
  21 double process::total_load_init = 0.0;
  22 double process::total_load_running = 0.0;
  23 double process::total_load_exit = 0.0;
  24
  25 int process::process_counter = 0;
  26 double process::total_load_average;
  27 double process::load_diff_threshold;
  28
  29 std::atomic<int> process::convergence_counter(0);
  30
  31 namespace {
  32
  33     void sleep_until_date(double& date, double duration)
  34     {
  35         double sleep_duration = date - MSG_get_clock();
  36         if (sleep_duration > 0.0)
  37             MSG_process_sleep(sleep_duration);
  38         date = MSG_get_clock() + duration;
  39     }
  40
  41 }
  42
  43 process::process(int argc, char* argv[])
  44 {
  45     if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))
  46         throw std::invalid_argument("bad or missing initial load parameter");
  47
  48     double iload = std::trunc(real_load);
  49     if (opt::integer_transfer && real_load != iload) {
  50         XBT_WARN("Initial load %g is not an integer.  Truncate it.",
  51                  real_load);
  52         real_load = iload;
  53     }
  54
  55     neigh.assign(argv + 2, argv + argc);
  56
  57     pneigh.reserve(neigh.size());
  58     for (unsigned i = 0 ; i < neigh.size() ; i++) {
  59         neighbor* ptr = &neigh[i];
  60         m_host_t host = MSG_get_host_by_name(ptr->get_name());
  61         pneigh.push_back(ptr);
  62         rev_neigh.insert(std::make_pair(host, ptr));
  63     }
  64
  65     prev_load_broadcast = -1;   // force sending of load on first send_all()
  66     expected_load = real_load;
  67     received_load = 0.0;
  68
  69     idle_duration = 0.0;
  70     convergence = -1.0;
  71
  72     proc_mutex->acquire();
  73     process_counter++;
  74     convergence_counter++;
  75     total_load_init += real_load;
  76     total_load_running += real_load;
  77     total_load_average = total_load_running / process_counter;
  78     load_diff_threshold = (opt::load_ratio_threshold +
  79                            opt::avg_load_ratio * total_load_average) / 100.0;
  80     proc_mutex->release();
  81
  82     ctrl_close_pending = data_close_pending = neigh.size();
  83     close_received = false;
  84     finalizing = false;
  85
  86     all_comp_iter = comp_iter = lb_iter = 0;
  87
  88     lb_thread = new_msg_thread("loba",
  89                                std::bind(&process::load_balance_loop, this));
  90
  91     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  92     if (!LOG_ISENABLED(logp))
  93         return;
  94     std::ostringstream oss;
  95     oss << neigh.size() << " neighbor";
  96     if (!neigh.empty()) {
  97         oss << ESSE(neigh.size()) << ": ";
  98         std::transform(neigh.begin(), neigh.end() - 1,
  99                        std::ostream_iterator<const char*>(oss, ", "),
 100                        std::mem_fn(&neighbor::get_name));
 101         oss << neigh.back().get_name();
 102     }
 103     XBT_LOG(logp, "Got %s.", oss.str().c_str());
 104     print_loads(false, logp);
 105 }
 106
 107 process::~process()
 108 {
 109     delete lb_thread;
 110     proc_mutex->acquire();
 111     total_load_exit += real_load;
 112     proc_mutex->release();
 113     xbt_assert(received_load == 0.0,
 114                "received_load is %g, but should be 0.0 !", received_load);
 115     if (opt::log_rate < 0)
 116         return;
 117     XBT_INFO("Final load after %d:%d:%d iterations: %g",
 118              lb_iter, comp_iter, all_comp_iter, real_load);
 119     if (convergence >= 0.0)
 120         XBT_INFO("Convergence within %g%% was achieved at time %g",
 121                  opt::avg_load_ratio, convergence);
 122     else
 123         XBT_INFO("Convergence within %g%% was not achieved",
 124                  opt::avg_load_ratio);
 125     XBT_VERB("Expected load was: %g", expected_load);
 126     XBT_VERB("Total computation for this process: %g", get_comp_amount());
 127     print_loads(true, xbt_log_priority_debug);
 128 }
 129
 130 double process::get_iter_deviation() const
 131 {
 132     double average_cost = opt::comp_cost(total_load_average); // fixme: get locked?
 133     // Do not count idle periods
 134     double comp_iter_opt = acc.comp_amount / average_cost;
 135 /*
 136     // Add iterations that could have been achieved while beeing idle
 137     // (kept for documentation)
 138     double self_speed = MSG_get_host_speed(MSG_host_self());
 139     double average_duration = average_cost / self_speed;
 140     comp_iter_opt += idle_duration / average_duration;
 141 */
 142     return comp_iter - comp_iter_opt;
 143 }
 144
 145 int process::run()
 146 {
 147     if (opt::log_rate >= 0) {
 148         XBT_INFO("Initial load: %g", real_load);
 149         XBT_VERB("Initial expected load: %g", expected_load);
 150     }
 151     XBT_VERB("Starting...");
 152     mutex.acquire();
 153     lb_thread->start();
 154     while (lb_iter <= opt::comp_iter_delay)
 155         cond.wait(mutex);
 156     mutex.release();
 157     double sleep_duration = opt::comp_time_delay - MSG_get_clock();
 158     if (sleep_duration > 0.0)
 159         MSG_process_sleep(sleep_duration);
 160     compute_loop();
 161     lb_thread->wait();
 162     XBT_VERB("Done.");
 163     return 0;
 164 }
 165
 166 void process::load_balance_loop()
 167 {
 168     using std::placeholders::_1;
 169
 170     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 171     while (still_running()) {
 172         if (lb_iter == opt::comp_iter_delay) {
 173             mutex.acquire();
 174             ++lb_iter;
 175             cond.signal();
 176             mutex.release();
 177         } else {
 178             ++lb_iter;
 179         }
 180
 181         ctrl_receive(0.0);
 182
 183         mutex.acquire();
 184         if (!opt::bookkeeping)
 185             expected_load = real_load - get_sum_of_to_send();
 186         // nothing to do with opt::bookkeeping
 187
 188         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 189             XBT_INFO("(%u:%u:%u) current load: %g",
 190                      lb_iter, comp_iter, all_comp_iter, real_load);
 191             XBT_VERB("... expected load: %g", expected_load);
 192         }
 193
 194         if (expected_load > 0.0)
 195             load_balance();
 196
 197         print_loads(true, xbt_log_priority_debug);
 198
 199         // send
 200         comm.ctrl_flush(false);
 201         std::for_each(neigh.begin(), neigh.end(),
 202                       std::bind(&process::ctrl_send, this, _1));
 203         prev_load_broadcast = expected_load;
 204         mutex.release();
 205
 206         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 207     }
 208
 209     XBT_VERB("Going to finalize for %s...", __func__);
 210     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 211               neigh.size(), ESSE(neigh.size()));
 212     std::for_each(neigh.begin(), neigh.end(),
 213                   std::bind(&process::ctrl_close, this, _1));
 214     while (ctrl_close_pending) {
 215         comm.ctrl_flush(false);
 216         XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);
 217         ctrl_receive(-1.0);
 218     }
 219     comm.ctrl_flush(true);
 220 }
 221
 222 void process::compute_loop()
 223 {
 224     using std::placeholders::_1;
 225
 226     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 227     double idle_since_date = 0.0;
 228     while (still_running()) {
 229         // receive
 230         // if there is something to compute, do not block
 231         // else, block the duration of an *lb* iteration
 232         data_receive(real_load > 0.0 ? 0.0 : opt::min_lb_iter_duration);
 233
 234         // send
 235         comm.data_flush(false);
 236         mutex.acquire();
 237         real_load += received_load;
 238         received_load = 0.0;
 239         std::for_each(neigh.begin(), neigh.end(),
 240                       std::bind(&process::data_send, this, _1));
 241         mutex.release();
 242
 243         ++all_comp_iter;
 244         if (real_load == 0.0)
 245             continue;
 246
 247         convergence_check();
 248
 249         // compute
 250         idle_duration += MSG_get_clock() - idle_since_date;
 251         ++comp_iter;
 252         double flops = opt::comp_cost(real_load);
 253         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 254         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 255         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 256         MSG_task_execute(task);
 257         add_comp_amount(flops);
 258         MSG_task_destroy(task);
 259
 260         idle_since_date = MSG_get_clock();
 261
 262         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 263     }
 264
 265     XBT_VERB("Going to finalize for %s...", __func__);
 266     // Note: idle duration is not counted during finalization
 267     finalizing = true;
 268     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 269               neigh.size(), ESSE(neigh.size()));
 270     std::for_each(neigh.begin(), neigh.end(),
 271                   std::bind(&process::data_close, this, _1));
 272     while (data_close_pending) {
 273         comm.data_flush(false);
 274         XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);
 275         data_receive(-1.0);
 276     }
 277     real_load += received_load;
 278     received_load = 0.0;
 279     proc_mutex->acquire();
 280     total_load_running -= real_load;
 281     proc_mutex->release();
 282     convergence_check();
 283     comm.data_flush(true);
 284 }
 285
 286 void process::convergence_check()
 287 {
 288     double average = total_load_average; // fixme: get locked?
 289     double load_diff = std::fabs(real_load - average);
 290     bool converged = load_diff <= load_diff_threshold;
 291
 292     if (convergence >= 0.0) {
 293         if (!converged) {
 294             XBT_VERB("current load has diverged: %g (%.4g%%)",
 295                      real_load, 100.0 * load_diff / average);
 296             convergence = -1.0;
 297             convergence_counter++;
 298         }
 299     } else {
 300         if (converged) {
 301             XBT_VERB("current load has converged: %g (%.4g%%)",
 302                      real_load,  100.0 * load_diff / average);
 303             convergence = MSG_get_clock();
 304             convergence_counter--;
 305         }
 306     }
 307 }
 308
 309 bool process::still_running()
 310 {
 311     static bool last_status = true;
 312
 313     if (!last_status) {
 314         /* nop */
 315
 316     } else if (opt::exit_request) {
 317         XBT_VERB("Global exit requested");
 318         last_status = false;
 319
 320     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 321         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 322         last_status = false;
 323
 324     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 325         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 326         last_status = false;
 327
 328     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 329         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 330         last_status = false;
 331
 332     } else if (opt::exit_on_convergence && convergence_counter == 0) {
 333         XBT_VERB("Global convergence detected");
 334         last_status = false;
 335
 336     } else if (opt::exit_on_close && close_received) {
 337         XBT_VERB("Close received");
 338         last_status = false;
 339
 340     } else if (real_load == 0.0 && !data_close_pending) {
 341         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 342         last_status = false;
 343
 344     } else if (100.0 * total_load_running / total_load_init <=
 345                opt::load_ratio_threshold) { // fixme: get locked?
 346         // fixme: this check should be implemented with a distributed
 347         // algorithm, and not a shared global variable!
 348         XBT_VERB("No more load to balance in system.");
 349         last_status = false;
 350     }
 351
 352     return last_status;
 353 }
 354
 355 double process::get_sum_of_to_send() const
 356 {
 357     using std::placeholders::_1;
 358     using std::placeholders::_2;
 359
 360     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 361                            std::bind(std::plus<double>(), _1,
 362                                      std::bind(&neighbor::get_to_send, _2)));
 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 = nb.get_debt() - nb.get_credit();
 416         load_to_send = std::min(excess_load,
 417                                 std::max(0.0, balance));
 418
 419         // adjust load to send (rounding, truncation, etc.)
 420         load_to_send = compute_load_to_send(load_to_send);
 421         if (load_to_send > 0.0)
 422             nb.set_debt(nb.get_debt() - load_to_send);
 423     } else {                    // !bookkeeping
 424         load_to_send = compute_load_to_send(nb.get_to_send());
 425         if (load_to_send > 0.0)
 426             nb.set_to_send(nb.get_to_send() - load_to_send);
 427     }
 428     real_load -= load_to_send;
 429     while (load_to_send > 0.0) {
 430         double amount;
 431         if (opt::max_transfer_amount)
 432             amount = std::min(load_to_send, opt::max_transfer_amount);
 433         else
 434             amount = load_to_send;
 435         message* msg = new message(message::DATA, amount);
 436         add_data_send_mesg(msg->get_size());
 437         comm.data_send(nb.get_data_mbox(), msg);
 438         load_to_send -= amount;
 439     }
 440 }
 441
 442 void process::ctrl_close(neighbor& nb)
 443 {
 444     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 445 }
 446
 447 void process::data_close(neighbor& nb)
 448 {
 449     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 450 }
 451
 452 void process::ctrl_receive(double timeout)
 453 {
 454     message* msg;
 455     m_host_t from;
 456
 457     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 458     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 459         if (msg->get_type() != message::CTRL_CLOSE)
 460             add_ctrl_recv_mesg(msg->get_size());
 461         handle_message(msg, from);
 462         timeout = 0.0;
 463     }
 464 }
 465
 466 void process::data_receive(double timeout)
 467 {
 468     message* msg;
 469     m_host_t from;
 470
 471     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 472     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 473         if (msg->get_type() != message::DATA_CLOSE)
 474             add_data_recv_mesg(msg->get_size());
 475         handle_message(msg, from);
 476         timeout = 0.0;
 477     }
 478 }
 479
 480 void process::handle_message(message* msg, m_host_t from)
 481 {
 482     switch (msg->get_type()) {
 483     case message::CTRL: {
 484         neighbor* n = rev_neigh[from];
 485         n->set_load(msg->get_amount() + n->get_to_send());
 486         if (opt::bookkeeping) {
 487             double credit = msg->get_credit();
 488             expected_load += credit;
 489             n->set_credit(n->get_credit() + credit);
 490         }
 491         break;
 492     }
 493     case message::DATA: {
 494         neighbor* n = rev_neigh[from];
 495         double ld = msg->get_amount();
 496         received_load += ld;
 497         n->set_credit(n->get_credit() - ld);
 498         break;
 499     }
 500     case message::CTRL_CLOSE:
 501         ctrl_close_pending--;
 502         close_received = true;
 503         break;
 504     case message::DATA_CLOSE:
 505         data_close_pending--;
 506         close_received = true;
 507         break;
 508     }
 509     delete msg;
 510 }
 511
 512 #define print_loads_generic(vec, verbose, logp, cat)                    \
 513     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 514         using std::placeholders::_1;                                    \
 515         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 516                   "Neighbor loads:", real_load, expected_load);         \
 517         std::for_each(vec.begin(), vec.end(),                           \
 518                       std::bind(&neighbor::print, _1, verbose, logp, cat)); \
 519     } else ((void)0)
 520
 521 void process::print_loads(bool verbose,
 522                           e_xbt_log_priority_t logp,
 523                           xbt_log_category_t cat) const
 524 {
 525     print_loads_generic(neigh, 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     print_loads_generic(pneigh, verbose, logp, cat);
 533 }
 534
 535 #undef print_loads_generic
 536
 537 // Local variables:
 538 // mode: c++
 539 // End: