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