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::average_load_ratio;
  28 double process::load_diff_threshold;
  29
  30 std::atomic<int> process::convergence_counter(0);
  31
  32 namespace {
  33
  34     void sleep_until_date(double& date, double duration)
  35     {
  36         double sleep_duration = date - MSG_get_clock();
  37         if (sleep_duration > 0.0)
  38             MSG_process_sleep(sleep_duration);
  39         date = MSG_get_clock() + duration;
  40     }
  41
  42 }
  43
  44 process::process(int argc, char* argv[])
  45 {
  46     if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))
  47         throw std::invalid_argument("bad or missing initial load parameter");
  48
  49     double iload = std::trunc(real_load);
  50     if (opt::integer_transfer && real_load != iload) {
  51         XBT_WARN("Initial load %g is not an integer.  Truncate it.",
  52                  real_load);
  53         real_load = iload;
  54     }
  55
  56     neigh.assign(argv + 2, argv + argc);
  57
  58     pneigh.reserve(neigh.size());
  59     for (unsigned i = 0 ; i < neigh.size() ; i++) {
  60         neighbor* ptr = &neigh[i];
  61         msg_host_t host = MSG_get_host_by_name(ptr->get_name());
  62         pneigh.push_back(ptr);
  63         rev_neigh.insert(std::make_pair(host, ptr));
  64     }
  65
  66     prev_load_broadcast = -1;   // force sending of load on first send_all()
  67     expected_load = real_load;
  68     received_load = 0.0;
  69
  70     idle_duration = 0.0;
  71     convergence = -1.0;
  72
  73     proc_mutex->acquire();
  74     process_counter++;
  75     convergence_counter++;
  76     total_load_init += real_load;
  77     total_load_running += real_load;
  78     total_load_average = total_load_running / process_counter;
  79     if (opt::avg_load_ratio >= 0.0)
  80         average_load_ratio = opt::avg_load_ratio;
  81     else
  82         average_load_ratio = 100.0 *
  83             (process_counter / -opt::avg_load_ratio) / total_load_average;
  84     load_diff_threshold = (opt::load_ratio_threshold +
  85                            average_load_ratio * total_load_average) / 100.0;
  86     proc_mutex->release();
  87
  88     ctrl_close_pending = data_close_pending = neigh.size();
  89     close_received = false;
  90     finalizing = false;
  91
  92     all_comp_iter = comp_iter = lb_iter = 0;
  93
  94     lb_thread = new_msg_thread("loba",
  95                                std::bind(&process::load_balance_loop, this));
  96
  97     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  98     if (!LOG_ISENABLED(logp))
  99         return;
 100     std::ostringstream oss;
 101     oss << neigh.size() << " neighbor";
 102     if (!neigh.empty()) {
 103         oss << ESSE(neigh.size()) << ": ";
 104         std::transform(neigh.begin(), neigh.end() - 1,
 105                        std::ostream_iterator<const char*>(oss, ", "),
 106                        std::mem_fn(&neighbor::get_name));
 107         oss << neigh.back().get_name();
 108     }
 109     XBT_LOG(logp, "Got %s.", oss.str().c_str());
 110     print_loads(false, logp);
 111 }
 112
 113 process::~process()
 114 {
 115     delete lb_thread;
 116     proc_mutex->acquire();
 117     total_load_exit += real_load;
 118     proc_mutex->release();
 119     xbt_assert(received_load == 0.0,
 120                "received_load is %g, but should be 0.0 !", received_load);
 121     if (opt::log_rate < 0)
 122         return;
 123     XBT_INFO("Final load after %d:%d:%d iterations: %g",
 124              lb_iter, comp_iter, all_comp_iter, real_load);
 125     if (convergence >= 0.0)
 126         XBT_INFO("Convergence within %g%% was achieved at time %g",
 127                  average_load_ratio, convergence);
 128     else
 129         XBT_INFO("Convergence within %g%% was not achieved",
 130                  average_load_ratio);
 131     XBT_VERB("Expected load was: %g", expected_load);
 132     XBT_VERB("Total computation for this process: %g", get_comp_amount());
 133     print_loads(true, xbt_log_priority_debug);
 134 }
 135
 136 double process::get_iter_deviation() const
 137 {
 138     double average_cost = opt::comp_cost(total_load_average); // fixme: get locked?
 139     // Do not count idle periods
 140     double comp_iter_opt = acc.comp_amount / average_cost;
 141 /*
 142     // Add iterations that could have been achieved while beeing idle
 143     // (kept for documentation)
 144     double self_speed = MSG_get_host_speed(MSG_host_self());
 145     double average_duration = average_cost / self_speed;
 146     comp_iter_opt += idle_duration / average_duration;
 147 */
 148     return comp_iter - comp_iter_opt;
 149 }
 150
 151 int process::run()
 152 {
 153     if (opt::log_rate >= 0) {
 154         XBT_INFO("Initial load: %g", real_load);
 155         XBT_VERB("Initial expected load: %g", expected_load);
 156     }
 157     XBT_VERB("Starting...");
 158     mutex.acquire();
 159     lb_thread->start();
 160     while (lb_iter <= opt::comp_iter_delay)
 161         cond.wait(mutex);
 162     mutex.release();
 163     double sleep_duration = opt::comp_time_delay - MSG_get_clock();
 164     if (sleep_duration > 0.0)
 165         MSG_process_sleep(sleep_duration);
 166     compute_loop();
 167     lb_thread->wait();
 168     XBT_VERB("Done.");
 169     return 0;
 170 }
 171
 172 void process::load_balance_loop()
 173 {
 174     using std::placeholders::_1;
 175
 176     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 177     while (still_running()) {
 178         if (lb_iter == opt::comp_iter_delay) {
 179             mutex.acquire();
 180             ++lb_iter;
 181             cond.signal();
 182             mutex.release();
 183         } else {
 184             ++lb_iter;
 185         }
 186
 187         ctrl_receive(0.0);
 188
 189         mutex.acquire();
 190         if (!opt::bookkeeping)
 191             expected_load = real_load - get_sum_of_to_send();
 192         // nothing to do with opt::bookkeeping
 193
 194         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 195             XBT_INFO("(%u:%u:%u) current load: %g",
 196                      lb_iter, comp_iter, all_comp_iter, real_load);
 197             XBT_VERB("... expected load: %g", expected_load);
 198         }
 199
 200         if (expected_load > 0.0)
 201             load_balance();
 202
 203         print_loads(true, xbt_log_priority_debug);
 204
 205         // send
 206         comm.ctrl_flush(false);
 207         std::for_each(neigh.begin(), neigh.end(),
 208                       std::bind(&process::ctrl_send, this, _1));
 209         prev_load_broadcast = expected_load;
 210         mutex.release();
 211
 212         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 213     }
 214
 215     XBT_VERB("Going to finalize for %s...", __func__);
 216     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 217               neigh.size(), ESSE(neigh.size()));
 218     std::for_each(neigh.begin(), neigh.end(),
 219                   std::bind(&process::ctrl_close, this, _1));
 220     while (ctrl_close_pending) {
 221         comm.ctrl_flush(false);
 222         XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);
 223         ctrl_receive(-1.0);
 224     }
 225     comm.ctrl_flush(true);
 226 }
 227
 228 void process::compute_loop()
 229 {
 230     using std::placeholders::_1;
 231
 232     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 233     double idle_since_date = 0.0;
 234     while (still_running()) {
 235         // receive
 236         // if there is something to compute, do not block
 237         // else, block the duration of an *lb* iteration
 238         data_receive(real_load > 0.0 ? 0.0 : opt::min_lb_iter_duration);
 239
 240         // send
 241         comm.data_flush(false);
 242         mutex.acquire();
 243         real_load += received_load;
 244         received_load = 0.0;
 245         std::for_each(neigh.begin(), neigh.end(),
 246                       std::bind(&process::data_send, this, _1));
 247         mutex.release();
 248
 249         ++all_comp_iter;
 250         if (real_load == 0.0)
 251             continue;
 252
 253         convergence_check();
 254
 255         // compute
 256         idle_duration += MSG_get_clock() - idle_since_date;
 257         ++comp_iter;
 258         double flops = opt::comp_cost(real_load);
 259         msg_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 260         // MSG_task_set_category(task, TRACE_CAT_COMP);
 261         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 262         MSG_task_execute(task);
 263         add_comp_amount(flops);
 264         MSG_task_destroy(task);
 265
 266         idle_since_date = MSG_get_clock();
 267
 268         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 269     }
 270
 271     XBT_VERB("Going to finalize for %s...", __func__);
 272     // Note: idle duration is not counted during finalization
 273     finalizing = true;
 274     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 275               neigh.size(), ESSE(neigh.size()));
 276     std::for_each(neigh.begin(), neigh.end(),
 277                   std::bind(&process::data_close, this, _1));
 278     while (data_close_pending) {
 279         comm.data_flush(false);
 280         XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);
 281         data_receive(-1.0);
 282     }
 283     real_load += received_load;
 284     received_load = 0.0;
 285     proc_mutex->acquire();
 286     total_load_running -= real_load;
 287     proc_mutex->release();
 288     convergence_check();
 289     comm.data_flush(true);
 290 }
 291
 292 void process::convergence_check()
 293 {
 294     double average = total_load_average; // fixme: get locked?
 295     double load_diff = std::fabs(real_load - average);
 296     bool converged = load_diff <= load_diff_threshold;
 297
 298     if (converged) {
 299         if (convergence < 0) {
 300             XBT_VERB("current load has converged: %g (%.4g%%)",
 301                      real_load,  100.0 * load_diff / average);
 302             convergence = MSG_get_clock();
 303             local_convergence_counter = opt::exit_on_convergence;
 304         }
 305         if (local_convergence_counter > 0 && --local_convergence_counter == 0)
 306                 --convergence_counter;
 307     } else {
 308         if (convergence >= 0.0) {
 309             XBT_VERB("current load has diverged: %g (%.4g%%)",
 310                      real_load, 100.0 * load_diff / average);
 311             convergence = -1.0;
 312             if (local_convergence_counter == 0)
 313                 ++convergence_counter;
 314         }
 315     }
 316 }
 317
 318 bool process::still_running()
 319 {
 320     static bool last_status = true;
 321
 322     if (!last_status) {
 323         /* nop */
 324
 325     } else if (opt::exit_request) {
 326         XBT_VERB("Global exit requested");
 327         last_status = false;
 328
 329     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 330         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 331         last_status = false;
 332
 333     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 334         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 335         last_status = false;
 336
 337     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 338         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 339         last_status = false;
 340
 341     } else if (opt::exit_on_convergence && convergence_counter == 0) {
 342         XBT_VERB("Global convergence detected");
 343         last_status = false;
 344
 345     } else if (opt::exit_on_close && close_received) {
 346         XBT_VERB("Close received");
 347         last_status = false;
 348
 349     } else if (real_load == 0.0 && !data_close_pending) {
 350         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 351         last_status = false;
 352
 353     } else if (100.0 * total_load_running / total_load_init <=
 354                opt::load_ratio_threshold) { // fixme: get locked?
 355         // fixme: this check should be implemented with a distributed
 356         // algorithm, and not a shared global variable!
 357         XBT_VERB("No more load to balance in system.");
 358         last_status = false;
 359     }
 360
 361     return last_status;
 362 }
 363
 364 double process::get_sum_of_to_send() const
 365 {
 366     using std::placeholders::_1;
 367     using std::placeholders::_2;
 368
 369     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 370                            std::bind(std::plus<double>(), _1,
 371                                      std::bind(&neighbor::get_to_send, _2)));
 372 }
 373
 374 void process::load_balance()
 375 {
 376     if (lb_iter == 1)           // warn only once
 377         XBT_WARN("process::load_balance() is a no-op!");
 378 }
 379
 380 void process::send(neighbor& nb, double amount)
 381 {
 382     expected_load -= amount;
 383     nb.set_to_send(nb.get_to_send() + amount);
 384     nb.set_load(nb.get_load() + amount);
 385 }
 386
 387 void process::ctrl_send(neighbor& nb)
 388 {
 389     double info_to_send = expected_load;
 390     double debt_to_send;
 391     if (opt::bookkeeping) {     // bookkeeping
 392         debt_to_send = nb.get_to_send();
 393         if (debt_to_send > 0.0) {
 394             nb.set_to_send(0.0);
 395             nb.set_debt(nb.get_debt() + debt_to_send);
 396         }
 397     } else {                    // !bookkeeping
 398         debt_to_send = 0.0;
 399     }
 400     if (info_to_send != prev_load_broadcast || debt_to_send > 0.0) {
 401         message* msg = new message(message::CTRL, info_to_send, debt_to_send);
 402         add_ctrl_send_mesg(msg->get_size());
 403         comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 404     }
 405 }
 406
 407 double process::compute_load_to_send(double desired)
 408 {
 409     if (opt::integer_transfer)
 410         desired = std::floor(desired);
 411     return desired >= opt::min_transfer_amount ? desired : 0.0;
 412 }
 413
 414 void process::data_send(neighbor& nb)
 415 {
 416     double load_to_send;
 417     if (opt::bookkeeping) {     // bookkeeping
 418         double excess_load;     // load amount we are able to send
 419         if (opt::egocentric)
 420             excess_load = std::max(0.0, real_load - expected_load);
 421         else
 422             excess_load = real_load;
 423
 424         double balance;
 425         if (nb.get_credit() > 0.0)
 426             balance = nb.get_debt() - nb.get_credit();
 427         else
 428             balance = nb.get_debt();
 429         load_to_send = std::min(excess_load,
 430                                 std::max(0.0, balance));
 431
 432         // adjust load to send (rounding, truncation, etc.)
 433         load_to_send = compute_load_to_send(load_to_send);
 434         if (load_to_send > 0.0)
 435             nb.set_debt(nb.get_debt() - load_to_send);
 436     } else {                    // !bookkeeping
 437         load_to_send = compute_load_to_send(nb.get_to_send());
 438         if (load_to_send > 0.0)
 439             nb.set_to_send(nb.get_to_send() - load_to_send);
 440     }
 441     real_load -= load_to_send;
 442     while (load_to_send > 0.0) {
 443         double amount;
 444         if (opt::max_transfer_amount)
 445             amount = std::min(load_to_send, opt::max_transfer_amount);
 446         else
 447             amount = load_to_send;
 448         message* msg = new message(message::DATA, amount);
 449         add_data_send_mesg(msg->get_size());
 450         comm.data_send(nb.get_data_mbox(), msg);
 451         load_to_send -= amount;
 452     }
 453 }
 454
 455 void process::ctrl_close(neighbor& nb)
 456 {
 457     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 458 }
 459
 460 void process::data_close(neighbor& nb)
 461 {
 462     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 463 }
 464
 465 void process::ctrl_receive(double timeout)
 466 {
 467     message* msg;
 468     msg_host_t from;
 469
 470     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 471     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 472         if (msg->get_type() != message::CTRL_CLOSE)
 473             add_ctrl_recv_mesg(msg->get_size());
 474         handle_message(msg, from);
 475         timeout = 0.0;
 476     }
 477 }
 478
 479 void process::data_receive(double timeout)
 480 {
 481     message* msg;
 482     msg_host_t from;
 483
 484     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 485     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 486         if (msg->get_type() != message::DATA_CLOSE)
 487             add_data_recv_mesg(msg->get_size());
 488         handle_message(msg, from);
 489         timeout = 0.0;
 490     }
 491 }
 492
 493 void process::handle_message(message* msg, msg_host_t from)
 494 {
 495     switch (msg->get_type()) {
 496     case message::CTRL: {
 497         neighbor* n = rev_neigh[from];
 498         n->set_load(msg->get_amount() + n->get_to_send());
 499         if (opt::bookkeeping) {
 500             double credit = msg->get_credit();
 501             expected_load += credit;
 502             n->set_credit(n->get_credit() + credit);
 503         }
 504         break;
 505     }
 506     case message::DATA: {
 507         neighbor* n = rev_neigh[from];
 508         double ld = msg->get_amount();
 509         received_load += ld;
 510         n->set_credit(n->get_credit() - ld);
 511         break;
 512     }
 513     case message::CTRL_CLOSE:
 514         ctrl_close_pending--;
 515         close_received = true;
 516         break;
 517     case message::DATA_CLOSE:
 518         data_close_pending--;
 519         close_received = true;
 520         break;
 521     }
 522     delete msg;
 523 }
 524
 525 #define print_loads_generic(vec, verbose, logp, cat)                    \
 526     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 527         using std::placeholders::_1;                                    \
 528         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 529                   "Neighbor loads:", real_load, expected_load);         \
 530         std::for_each(vec.begin(), vec.end(),                           \
 531                       std::bind(&neighbor::print, _1, verbose, logp, cat)); \
 532     } else ((void)0)
 533
 534 void process::print_loads(bool verbose,
 535                           e_xbt_log_priority_t logp,
 536                           xbt_log_category_t cat) const
 537 {
 538     print_loads_generic(neigh, verbose, logp, cat);
 539 }
 540
 541 void process::print_loads_p(bool verbose,
 542                             e_xbt_log_priority_t logp,
 543                             xbt_log_category_t cat) const
 544 {
 545     print_loads_generic(pneigh, verbose, logp, cat);
 546 }
 547
 548 #undef print_loads_generic
 549
 550 // Local variables:
 551 // mode: c++
 552 // End: