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
  10 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
  11
  12 #include "misc.h"
  13 #include "options.h"
  14 #include "tracing.h"
  15
  16 #include "process.h"
  17
  18 mutex_t *process::proc_mutex;
  19
  20 double process::total_load_init = 0.0;
  21 double process::total_load_running = 0.0;
  22 double process::total_load_exit = 0.0;
  23
  24 int process::process_counter = 0;
  25 double process::total_load_average;
  26 double process::average_load_ratio;
  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         msg_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     if (opt::avg_load_ratio >= 0.0)
  79         average_load_ratio = opt::avg_load_ratio;
  80     else
  81         average_load_ratio = 100.0 *
  82             (process_counter / -opt::avg_load_ratio) / total_load_average;
  83     load_diff_threshold = (opt::load_ratio_threshold +
  84                            average_load_ratio * total_load_average) / 100.0;
  85     proc_mutex->release();
  86
  87     ctrl_close_pending = data_close_pending = neigh.size();
  88     close_received = false;
  89     finalizing = false;
  90
  91     all_comp_iter = comp_iter = lb_iter = 0;
  92
  93     lb_thread = new_msg_thread("loba",
  94                                std::bind(&process::load_balance_loop, this));
  95
  96     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  97     if (!LOG_ISENABLED(logp))
  98         return;
  99     std::ostringstream oss;
 100     oss << neigh.size() << " neighbor";
 101     if (!neigh.empty()) {
 102         oss << ESSE(neigh.size()) << ": ";
 103         std::transform(neigh.begin(), neigh.end() - 1,
 104                        std::ostream_iterator<const char*>(oss, ", "),
 105                        std::mem_fn(&neighbor::get_name));
 106         oss << neigh.back().get_name();
 107     }
 108     XBT_LOG(logp, "Got %s.", oss.str().c_str());
 109     print_loads(false, logp);
 110 }
 111
 112 process::~process()
 113 {
 114     delete lb_thread;
 115     proc_mutex->acquire();
 116     total_load_exit += real_load;
 117     proc_mutex->release();
 118     xbt_assert(received_load == 0.0,
 119                "received_load is %g, but should be 0.0 !", received_load);
 120     if (opt::log_rate < 0)
 121         return;
 122     XBT_INFO("Final load after %d:%d:%d iterations: %g",
 123              lb_iter, comp_iter, all_comp_iter, real_load);
 124     if (convergence >= 0.0)
 125         XBT_INFO("Convergence within %g%% was achieved at time %g",
 126                  average_load_ratio, convergence);
 127     else
 128         XBT_INFO("Convergence within %g%% was not achieved",
 129                  average_load_ratio);
 130     XBT_VERB("Expected load was: %g", expected_load);
 131     XBT_VERB("Total computation for this process: %g", get_comp_amount());
 132     print_loads(true, xbt_log_priority_debug);
 133 }
 134
 135 double process::get_iter_deviation() const
 136 {
 137     double average_cost = opt::comp_cost(total_load_average); // fixme: get locked?
 138     // Do not count idle periods
 139     double comp_iter_opt = acc.comp_amount / average_cost;
 140 /*
 141     // Add iterations that could have been achieved while beeing idle
 142     // (kept for documentation)
 143     double self_speed = MSG_get_host_speed(MSG_host_self());
 144     double average_duration = average_cost / self_speed;
 145     comp_iter_opt += idle_duration / average_duration;
 146 */
 147     return comp_iter - comp_iter_opt;
 148 }
 149
 150 int process::run()
 151 {
 152     if (opt::log_rate >= 0) {
 153         XBT_INFO("Initial load: %g", real_load);
 154         XBT_VERB("Initial expected load: %g", expected_load);
 155     }
 156     XBT_VERB("Starting...");
 157     mutex.acquire();
 158     lb_thread->start();
 159     while (lb_iter <= opt::comp_iter_delay)
 160         cond.wait(mutex);
 161     mutex.release();
 162     double sleep_duration = opt::comp_time_delay - MSG_get_clock();
 163     if (sleep_duration > 0.0)
 164         MSG_process_sleep(sleep_duration);
 165     compute_loop();
 166     lb_thread->wait();
 167     XBT_VERB("Done.");
 168     return 0;
 169 }
 170
 171 void process::load_balance_loop()
 172 {
 173     using std::placeholders::_1;
 174
 175     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 176     while (still_running()) {
 177         if (lb_iter == opt::comp_iter_delay) {
 178             mutex.acquire();
 179             ++lb_iter;
 180             cond.signal();
 181             mutex.release();
 182         } else {
 183             ++lb_iter;
 184         }
 185
 186         ctrl_receive(0.0);
 187
 188         mutex.acquire();
 189         if (!opt::bookkeeping)
 190             expected_load = real_load - get_sum_of_to_send();
 191         // nothing to do with opt::bookkeeping
 192
 193         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 194             XBT_INFO("(%u:%u:%u) current load: %g",
 195                      lb_iter, comp_iter, all_comp_iter, real_load);
 196             XBT_VERB("... expected load: %g", expected_load);
 197         }
 198
 199         if (expected_load > 0.0)
 200             load_balance();
 201
 202         print_loads(true, xbt_log_priority_debug);
 203
 204         // send
 205         comm.ctrl_flush(false);
 206         std::for_each(neigh.begin(), neigh.end(),
 207                       std::bind(&process::ctrl_send, this, _1));
 208         prev_load_broadcast = expected_load;
 209         mutex.release();
 210
 211         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 212     }
 213
 214     XBT_VERB("Going to finalize for %s...", __func__);
 215     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 216               neigh.size(), ESSE(neigh.size()));
 217     std::for_each(neigh.begin(), neigh.end(),
 218                   std::bind(&process::ctrl_close, this, _1));
 219     while (ctrl_close_pending) {
 220         comm.ctrl_flush(false);
 221         XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);
 222         ctrl_receive(-1.0);
 223     }
 224     comm.ctrl_flush(true);
 225 }
 226
 227 void process::compute_loop()
 228 {
 229     using std::placeholders::_1;
 230
 231     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 232     double idle_since_date = 0.0;
 233     while (still_running()) {
 234         // receive
 235         // if there is something to compute, do not block
 236         // else, block the duration of an *lb* iteration
 237         data_receive(real_load > 0.0 ? 0.0 : opt::min_lb_iter_duration);
 238
 239         // send
 240         comm.data_flush(false);
 241         mutex.acquire();
 242         real_load += received_load;
 243         received_load = 0.0;
 244         std::for_each(neigh.begin(), neigh.end(),
 245                       std::bind(&process::data_send, this, _1));
 246         mutex.release();
 247
 248         ++all_comp_iter;
 249         if (real_load == 0.0)
 250             continue;
 251
 252         convergence_check();
 253
 254         // compute
 255         idle_duration += MSG_get_clock() - idle_since_date;
 256         ++comp_iter;
 257         double flops = opt::comp_cost(real_load);
 258         msg_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 259         // MSG_task_set_category(task, TRACE_CAT_COMP);
 260         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 261         MSG_task_execute(task);
 262         add_comp_amount(flops);
 263         MSG_task_destroy(task);
 264
 265         idle_since_date = MSG_get_clock();
 266
 267         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 268     }
 269
 270     XBT_VERB("Going to finalize for %s...", __func__);
 271     // Note: idle duration is not counted during finalization
 272     finalizing = true;
 273     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 274               neigh.size(), ESSE(neigh.size()));
 275     std::for_each(neigh.begin(), neigh.end(),
 276                   std::bind(&process::data_close, this, _1));
 277     while (data_close_pending) {
 278         comm.data_flush(false);
 279         XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);
 280         data_receive(-1.0);
 281     }
 282     real_load += received_load;
 283     received_load = 0.0;
 284     proc_mutex->acquire();
 285     total_load_running -= real_load;
 286     proc_mutex->release();
 287     convergence_check();
 288     comm.data_flush(true);
 289 }
 290
 291 void process::convergence_check()
 292 {
 293     double average = total_load_average; // fixme: get locked?
 294     double load_diff = std::fabs(real_load - average);
 295     bool converged = load_diff <= load_diff_threshold;
 296
 297     if (converged) {
 298         if (convergence < 0) {
 299             XBT_VERB("current load has converged: %g (%.4g%%)",
 300                      real_load,  100.0 * load_diff / average);
 301             convergence = MSG_get_clock();
 302             local_convergence_counter = opt::exit_on_convergence;
 303         }
 304         if (local_convergence_counter > 0 && --local_convergence_counter == 0)
 305                 --convergence_counter;
 306     } else {
 307         if (convergence >= 0.0) {
 308             XBT_VERB("current load has diverged: %g (%.4g%%)",
 309                      real_load, 100.0 * load_diff / average);
 310             convergence = -1.0;
 311             if (local_convergence_counter == 0)
 312                 ++convergence_counter;
 313         }
 314     }
 315 }
 316
 317 bool process::still_running()
 318 {
 319     static bool last_status = true;
 320
 321     if (!last_status) {
 322         /* nop */
 323
 324     } else if (opt::exit_request) {
 325         XBT_VERB("Global exit requested");
 326         last_status = false;
 327
 328     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 329         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 330         last_status = false;
 331
 332     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 333         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 334         last_status = false;
 335
 336     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 337         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 338         last_status = false;
 339
 340     } else if (opt::exit_on_convergence && convergence_counter == 0) {
 341         XBT_VERB("Global convergence detected");
 342         last_status = false;
 343
 344     } else if (opt::exit_on_close && close_received) {
 345         XBT_VERB("Close received");
 346         last_status = false;
 347
 348     } else if (real_load == 0.0 && !data_close_pending) {
 349         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 350         last_status = false;
 351
 352     } else if (100.0 * total_load_running / total_load_init <=
 353                opt::load_ratio_threshold) { // fixme: get locked?
 354         // fixme: this check should be implemented with a distributed
 355         // algorithm, and not a shared global variable!
 356         XBT_VERB("No more load to balance in system.");
 357         last_status = false;
 358     }
 359
 360     return last_status;
 361 }
 362
 363 double process::get_sum_of_to_send() const
 364 {
 365     using std::placeholders::_1;
 366     using std::placeholders::_2;
 367
 368     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 369                            std::bind(std::plus<double>(), _1,
 370                                      std::bind(&neighbor::get_to_send, _2)));
 371 }
 372
 373 void process::load_balance()
 374 {
 375     if (lb_iter == 1)           // warn only once
 376         XBT_WARN("process::load_balance() is a no-op!");
 377 }
 378
 379 void process::send(neighbor& nb, double amount)
 380 {
 381     expected_load -= amount;
 382     nb.set_to_send(nb.get_to_send() + amount);
 383     nb.set_load(nb.get_load() + amount);
 384 }
 385
 386 void process::ctrl_send(neighbor& nb)
 387 {
 388     double info_to_send = expected_load;
 389     double debt_to_send;
 390     if (opt::bookkeeping) {     // bookkeeping
 391         debt_to_send = nb.get_to_send();
 392         if (debt_to_send > 0.0) {
 393             nb.set_to_send(0.0);
 394             nb.set_debt(nb.get_debt() + debt_to_send);
 395         }
 396     } else {                    // !bookkeeping
 397         debt_to_send = 0.0;
 398     }
 399     if (info_to_send != prev_load_broadcast || debt_to_send > 0.0) {
 400         message* msg = new message(message::CTRL, info_to_send, debt_to_send);
 401         add_ctrl_send_mesg(msg->get_size());
 402         comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 403     }
 404 }
 405
 406 double process::compute_load_to_send(double desired)
 407 {
 408     if (opt::integer_transfer)
 409         desired = std::floor(desired);
 410     return desired >= opt::min_transfer_amount ? desired : 0.0;
 411 }
 412
 413 void process::data_send(neighbor& nb)
 414 {
 415     double load_to_send;
 416     if (opt::bookkeeping) {     // bookkeeping
 417         double excess_load;     // load amount we are able to send
 418         if (opt::egocentric)
 419             excess_load = std::max(0.0, real_load - expected_load);
 420         else
 421             excess_load = real_load;
 422
 423         double balance;
 424         if (nb.get_credit() > 0.0)
 425             balance = nb.get_debt() - nb.get_credit();
 426         else
 427             balance = nb.get_debt();
 428         load_to_send = std::min(excess_load,
 429                                 std::max(0.0, balance));
 430
 431         // adjust load to send (rounding, truncation, etc.)
 432         load_to_send = compute_load_to_send(load_to_send);
 433         if (load_to_send > 0.0)
 434             nb.set_debt(nb.get_debt() - load_to_send);
 435     } else {                    // !bookkeeping
 436         load_to_send = compute_load_to_send(nb.get_to_send());
 437         if (load_to_send > 0.0)
 438             nb.set_to_send(nb.get_to_send() - load_to_send);
 439     }
 440     real_load -= load_to_send;
 441     while (load_to_send > 0.0) {
 442         double amount;
 443         if (opt::max_transfer_amount)
 444             amount = std::min(load_to_send, opt::max_transfer_amount);
 445         else
 446             amount = load_to_send;
 447         message* msg = new message(message::DATA, amount);
 448         add_data_send_mesg(msg->get_size());
 449         comm.data_send(nb.get_data_mbox(), msg);
 450         load_to_send -= amount;
 451     }
 452 }
 453
 454 void process::ctrl_close(neighbor& nb)
 455 {
 456     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 457 }
 458
 459 void process::data_close(neighbor& nb)
 460 {
 461     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 462 }
 463
 464 void process::ctrl_receive(double timeout)
 465 {
 466     message* msg;
 467     msg_host_t from;
 468
 469     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 470     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 471         if (msg->get_type() != message::CTRL_CLOSE)
 472             add_ctrl_recv_mesg(msg->get_size());
 473         handle_message(msg, from);
 474         timeout = 0.0;
 475     }
 476 }
 477
 478 void process::data_receive(double timeout)
 479 {
 480     message* msg;
 481     msg_host_t from;
 482
 483     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 484     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 485         if (msg->get_type() != message::DATA_CLOSE)
 486             add_data_recv_mesg(msg->get_size());
 487         handle_message(msg, from);
 488         timeout = 0.0;
 489     }
 490 }
 491
 492 void process::handle_message(message* msg, msg_host_t from)
 493 {
 494     switch (msg->get_type()) {
 495     case message::CTRL: {
 496         neighbor* n = rev_neigh[from];
 497         n->set_load(msg->get_amount() + n->get_to_send());
 498         if (opt::bookkeeping) {
 499             double credit = msg->get_credit();
 500             expected_load += credit;
 501             n->set_credit(n->get_credit() + credit);
 502         }
 503         break;
 504     }
 505     case message::DATA: {
 506         neighbor* n = rev_neigh[from];
 507         double ld = msg->get_amount();
 508         received_load += ld;
 509         n->set_credit(n->get_credit() - ld);
 510         break;
 511     }
 512     case message::CTRL_CLOSE:
 513         ctrl_close_pending--;
 514         close_received = true;
 515         break;
 516     case message::DATA_CLOSE:
 517         data_close_pending--;
 518         close_received = true;
 519         break;
 520     }
 521     delete msg;
 522 }
 523
 524 #define print_loads_generic(vec, verbose, logp, cat)                    \
 525     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 526         using std::placeholders::_1;                                    \
 527         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 528                   "Neighbor loads:", real_load, expected_load);         \
 529         std::for_each(vec.begin(), vec.end(),                           \
 530                       std::bind(&neighbor::print, _1, verbose, logp, cat)); \
 531     } else ((void)0)
 532
 533 void process::print_loads(bool verbose,
 534                           e_xbt_log_priority_t logp,
 535                           xbt_log_category_t cat) const
 536 {
 537     print_loads_generic(neigh, verbose, logp, cat);
 538 }
 539
 540 void process::print_loads_p(bool verbose,
 541                             e_xbt_log_priority_t logp,
 542                             xbt_log_category_t cat) const
 543 {
 544     print_loads_generic(pneigh, verbose, logp, cat);
 545 }
 546
 547 #undef print_loads_generic
 548
 549 // Local variables:
 550 // mode: c++
 551 // End: