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 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::load_diff_threshold;
  26
  27 namespace {
  28
  29     void sleep_until_date(double& date, double duration)
  30     {
  31         double sleep_duration = date - MSG_get_clock();
  32         if (sleep_duration > 0.0)
  33             MSG_process_sleep(sleep_duration);
  34         date = MSG_get_clock() + duration;
  35     }
  36
  37 }
  38
  39 process::process(int argc, char* argv[])
  40 {
  41     if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))
  42         throw std::invalid_argument("bad or missing initial load parameter");
  43
  44     double iload = std::trunc(real_load);
  45     if (opt::integer_transfer && real_load != iload) {
  46         XBT_WARN("Initial load %g is not an integer.  Truncate it.",
  47                  real_load);
  48         real_load = iload;
  49     }
  50
  51     neigh.assign(argv + 2, argv + argc);
  52
  53     pneigh.reserve(neigh.size());
  54     for (unsigned i = 0 ; i < neigh.size() ; i++) {
  55         neighbor* ptr = &neigh[i];
  56         m_host_t host = MSG_get_host_by_name(ptr->get_name());
  57         pneigh.push_back(ptr);
  58         rev_neigh.insert(std::make_pair(host, ptr));
  59     }
  60
  61     // Note: there should not be race condition with the current
  62     // version of Simgrid, when updating the global variables.
  63
  64     prev_load_broadcast = -1;   // force sending of load on first send_all()
  65     expected_load = real_load;
  66     total_load_running += real_load;
  67     total_load_init += real_load;
  68     received_load = 0.0;
  69
  70     idle_duration = 0.0;
  71     convergence = -1.0;
  72
  73     process_counter++;
  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
  78     ctrl_close_pending = data_close_pending = neigh.size();
  79     close_received = false;
  80     finalizing = false;
  81
  82     all_comp_iter = comp_iter = lb_iter = 0;
  83
  84     lb_thread = new_msg_thread("loba",
  85                                std::bind(&process::load_balance_loop, this));
  86
  87     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  88     if (!LOG_ISENABLED(logp))
  89         return;
  90     std::ostringstream oss;
  91     oss << neigh.size() << " neighbor";
  92     if (!neigh.empty()) {
  93         oss << ESSE(neigh.size()) << ": ";
  94         std::transform(neigh.begin(), neigh.end() - 1,
  95                        std::ostream_iterator<const char*>(oss, ", "),
  96                        std::mem_fn(&neighbor::get_name));
  97         oss << neigh.back().get_name();
  98     }
  99     XBT_LOG(logp, "Got %s.", oss.str().c_str());
 100     print_loads(false, logp);
 101 }
 102
 103 process::~process()
 104 {
 105     delete lb_thread;
 106     total_load_exit += real_load;
 107     xbt_assert(received_load == 0.0,
 108                "received_load is %g, but should be 0.0 !", received_load);
 109     if (opt::log_rate < 0)
 110         return;
 111     XBT_INFO("Final load after %d:%d:%d iterations: %g",
 112              lb_iter, comp_iter, all_comp_iter, real_load);
 113     if (convergence >= 0.0)
 114         XBT_INFO("Convergence within %g%% was achieved at time %g",
 115                  opt::avg_load_ratio, convergence);
 116     else
 117         XBT_INFO("Convergence within %g%% was not achieved",
 118                  opt::avg_load_ratio);
 119     XBT_VERB("Expected load was: %g", expected_load);
 120     XBT_VERB("Total computation for this process: %g", get_comp_amount());
 121     print_loads(true, xbt_log_priority_debug);
 122 }
 123
 124 int process::run()
 125 {
 126     if (opt::log_rate >= 0) {
 127         XBT_INFO("Initial load: %g", real_load);
 128         XBT_VERB("Initial expected load: %g", expected_load);
 129     }
 130     XBT_VERB("Starting...");
 131     mutex.acquire();
 132     lb_thread->start();
 133     while (lb_iter <= opt::comp_iter_delay)
 134         cond.wait(mutex);
 135     mutex.release();
 136     double sleep_duration = opt::comp_time_delay - MSG_get_clock();
 137     if (sleep_duration > 0.0)
 138         MSG_process_sleep(sleep_duration);
 139     compute_loop();
 140     lb_thread->wait();
 141     XBT_VERB("Done.");
 142     return 0;
 143 }
 144
 145 void process::load_balance_loop()
 146 {
 147     using std::placeholders::_1;
 148
 149     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 150     while (still_running()) {
 151         if (lb_iter == opt::comp_iter_delay) {
 152             mutex.acquire();
 153             ++lb_iter;
 154             cond.signal();
 155             mutex.release();
 156         } else {
 157             ++lb_iter;
 158         }
 159
 160         ctrl_receive(0.0);
 161
 162         mutex.acquire();
 163         if (!opt::bookkeeping)
 164             expected_load = real_load - get_sum_of_to_send();
 165         // nothing to do with opt::bookkeeping
 166
 167         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 168             XBT_INFO("(%u:%u:%u) current load: %g",
 169                      lb_iter, comp_iter, all_comp_iter, real_load);
 170             XBT_VERB("... expected load: %g", expected_load);
 171         }
 172
 173         if (expected_load > 0.0)
 174             load_balance();
 175
 176         print_loads(true, xbt_log_priority_debug);
 177
 178         // send
 179         comm.ctrl_flush(false);
 180         std::for_each(neigh.begin(), neigh.end(),
 181                       std::bind(&process::ctrl_send, this, _1));
 182         prev_load_broadcast = expected_load;
 183         mutex.release();
 184
 185         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 186     }
 187
 188     XBT_VERB("Going to finalize for %s...", __func__);
 189     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 190               neigh.size(), ESSE(neigh.size()));
 191     std::for_each(neigh.begin(), neigh.end(),
 192                   std::bind(&process::ctrl_close, this, _1));
 193     while (ctrl_close_pending) {
 194         comm.ctrl_flush(false);
 195         XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);
 196         ctrl_receive(-1.0);
 197     }
 198     comm.ctrl_flush(true);
 199 }
 200
 201 void process::compute_loop()
 202 {
 203     using std::placeholders::_1;
 204
 205     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 206     double idle_since_date = 0.0;
 207     while (still_running()) {
 208
 209         do {
 210             // receive
 211             // if there is something to compute, do not block
 212             // else, block the duration of an *lb* iteration
 213             data_receive(real_load > 0.0 ? 0.0 : opt::min_lb_iter_duration);
 214
 215             // send
 216             comm.data_flush(false);
 217             mutex.acquire();
 218             real_load += received_load;
 219             received_load = 0.0;
 220             std::for_each(neigh.begin(), neigh.end(),
 221                           std::bind(&process::data_send, this, _1));
 222             mutex.release();
 223
 224             ++all_comp_iter;
 225
 226         } while (real_load == 0.0);
 227
 228         convergence_check();
 229
 230         // compute
 231         idle_duration += MSG_get_clock() - idle_since_date;
 232         ++comp_iter;
 233         double flops = opt::comp_cost(real_load);
 234         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 235         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 236         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 237         MSG_task_execute(task);
 238         add_comp_amount(flops);
 239         MSG_task_destroy(task);
 240
 241         idle_since_date = MSG_get_clock();
 242
 243         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 244     }
 245
 246     XBT_VERB("Going to finalize for %s...", __func__);
 247     // Note: idle duration is not counted during finalization
 248     finalizing = true;
 249     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 250               neigh.size(), ESSE(neigh.size()));
 251     std::for_each(neigh.begin(), neigh.end(),
 252                   std::bind(&process::data_close, this, _1));
 253     while (data_close_pending) {
 254         comm.data_flush(false);
 255         XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);
 256         data_receive(-1.0);
 257     }
 258     real_load += received_load;
 259     received_load = 0.0;
 260     total_load_running -= real_load;
 261     convergence_check();
 262     comm.data_flush(true);
 263 }
 264
 265 void process::convergence_check()
 266 {
 267     double load_diff = std::fabs(real_load - total_load_average);
 268     bool converged = load_diff <= load_diff_threshold;
 269
 270     if (convergence >= 0.0) {
 271         if (!converged) {
 272             XBT_VERB("current load has diverged: %g (%.4g%%)",
 273                      real_load, 100.0 * load_diff / total_load_average);
 274             convergence = -1.0;
 275         }
 276     } else {
 277         if (converged) {
 278             XBT_VERB("current load has converged: %g (%.4g%%)",
 279                      real_load,  100.0 * load_diff / total_load_average);
 280             convergence = MSG_get_clock();
 281         }
 282     }
 283 }
 284
 285 bool process::still_running()
 286 {
 287     static bool last_status = true;
 288
 289     if (!last_status) {
 290         /* nop */
 291
 292     } else if (opt::exit_request) {
 293         XBT_VERB("Global exit requested");
 294         last_status = false;
 295
 296     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 297         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 298         last_status = false;
 299
 300     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 301         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 302         last_status = false;
 303
 304     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 305         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 306         last_status = false;
 307
 308     } else if (opt::exit_on_close && close_received) {
 309         XBT_VERB("Close received");
 310         last_status = false;
 311
 312     } else if (real_load == 0.0 && !data_close_pending) {
 313         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 314         last_status = false;
 315
 316     } else if (100.0 * total_load_running / total_load_init <=
 317                opt::load_ratio_threshold) {
 318         // fixme: this check should be implemented with a distributed
 319         // algorithm, and not a shared global variable!
 320         XBT_VERB("No more load to balance in system.");
 321         last_status = false;
 322     }
 323
 324     return last_status;
 325 }
 326
 327 double process::get_sum_of_to_send() const
 328 {
 329     using std::placeholders::_1;
 330     using std::placeholders::_2;
 331
 332     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 333                            std::bind(std::plus<double>(), _1,
 334                                      std::bind(&neighbor::get_to_send, _2)));
 335 }
 336
 337 void process::load_balance()
 338 {
 339     if (lb_iter == 1)           // warn only once
 340         XBT_WARN("process::load_balance() is a no-op!");
 341 }
 342
 343 void process::send(neighbor& nb, double amount)
 344 {
 345     expected_load -= amount;
 346     nb.set_to_send(nb.get_to_send() + amount);
 347     nb.set_load(nb.get_load() + amount);
 348 }
 349
 350 void process::ctrl_send(neighbor& nb)
 351 {
 352     double info_to_send = expected_load;
 353     double debt_to_send;
 354     if (opt::bookkeeping) {     // bookkeeping
 355         debt_to_send = nb.get_to_send();
 356         if (debt_to_send > 0.0) {
 357             nb.set_to_send(0.0);
 358             nb.set_debt(nb.get_debt() + debt_to_send);
 359         }
 360     } else {                    // !bookkeeping
 361         debt_to_send = 0.0;
 362     }
 363     if (info_to_send != prev_load_broadcast || debt_to_send > 0.0) {
 364         message* msg = new message(message::CTRL, info_to_send, debt_to_send);
 365         add_ctrl_send_mesg(msg->get_size());
 366         comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 367     }
 368 }
 369
 370 double process::compute_load_to_send(double desired)
 371 {
 372     if (opt::integer_transfer)
 373         desired = std::floor(desired);
 374     return desired >= opt::min_transfer_amount ? desired : 0.0;
 375 }
 376
 377 void process::data_send(neighbor& nb)
 378 {
 379     double load_to_send;
 380     if (opt::bookkeeping) {     // bookkeeping
 381         double excess_load;     // load amount we are able to send
 382         if (opt::egocentric)
 383             excess_load = std::max(0.0, real_load - expected_load);
 384         else
 385             excess_load = real_load;
 386
 387         double balance = nb.get_debt() - nb.get_credit();
 388         load_to_send = std::min(excess_load,
 389                                 std::max(0.0, balance));
 390
 391         // adjust load to send (rounding, truncation, etc.)
 392         load_to_send = compute_load_to_send(load_to_send);
 393         if (load_to_send > 0.0)
 394             nb.set_debt(nb.get_debt() - load_to_send);
 395     } else {                    // !bookkeeping
 396         load_to_send = compute_load_to_send(nb.get_to_send());
 397         if (load_to_send > 0.0)
 398             nb.set_to_send(nb.get_to_send() - load_to_send);
 399     }
 400     real_load -= load_to_send;
 401     while (load_to_send > 0.0) {
 402         double amount;
 403         if (opt::max_transfer_amount)
 404             amount = std::min(load_to_send, opt::max_transfer_amount);
 405         else
 406             amount = load_to_send;
 407         message* msg = new message(message::DATA, amount);
 408         add_data_send_mesg(msg->get_size());
 409         comm.data_send(nb.get_data_mbox(), msg);
 410         load_to_send -= amount;
 411     }
 412 }
 413
 414 void process::ctrl_close(neighbor& nb)
 415 {
 416     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 417 }
 418
 419 void process::data_close(neighbor& nb)
 420 {
 421     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 422 }
 423
 424 void process::ctrl_receive(double timeout)
 425 {
 426     message* msg;
 427     m_host_t from;
 428
 429     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 430     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 431         if (msg->get_type() != message::CTRL_CLOSE)
 432             add_ctrl_recv_mesg(msg->get_size());
 433         handle_message(msg, from);
 434         timeout = 0.0;
 435     }
 436 }
 437
 438 void process::data_receive(double timeout)
 439 {
 440     message* msg;
 441     m_host_t from;
 442
 443     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 444     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 445         if (msg->get_type() != message::DATA_CLOSE)
 446             add_data_recv_mesg(msg->get_size());
 447         handle_message(msg, from);
 448         timeout = 0.0;
 449     }
 450 }
 451
 452 void process::handle_message(message* msg, m_host_t from)
 453 {
 454     switch (msg->get_type()) {
 455     case message::CTRL: {
 456         neighbor* n = rev_neigh[from];
 457         n->set_load(msg->get_amount() + n->get_to_send());
 458         if (opt::bookkeeping) {
 459             double credit = msg->get_credit();
 460             expected_load += credit;
 461             n->set_credit(n->get_credit() + credit);
 462         }
 463         break;
 464     }
 465     case message::DATA: {
 466         neighbor* n = rev_neigh[from];
 467         double ld = msg->get_amount();
 468         received_load += ld;
 469         n->set_credit(n->get_credit() - ld);
 470         break;
 471     }
 472     case message::CTRL_CLOSE:
 473         ctrl_close_pending--;
 474         close_received = true;
 475         break;
 476     case message::DATA_CLOSE:
 477         data_close_pending--;
 478         close_received = true;
 479         break;
 480     }
 481     delete msg;
 482 }
 483
 484 #define print_loads_generic(vec, verbose, logp, cat)                    \
 485     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 486         using std::placeholders::_1;                                    \
 487         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 488                   "Neighbor loads:", real_load, expected_load);         \
 489         std::for_each(vec.begin(), vec.end(),                           \
 490                       std::bind(&neighbor::print, _1, verbose, logp, cat)); \
 491     } else ((void)0)
 492
 493 void process::print_loads(bool verbose,
 494                           e_xbt_log_priority_t logp,
 495                           xbt_log_category_t cat) const
 496 {
 497     print_loads_generic(neigh, verbose, logp, cat);
 498 }
 499
 500 void process::print_loads_p(bool verbose,
 501                             e_xbt_log_priority_t logp,
 502                             xbt_log_category_t cat) const
 503 {
 504     print_loads_generic(pneigh, verbose, logp, cat);
 505 }
 506
 507 #undef print_loads_generic
 508
 509 // Local variables:
 510 // mode: c++
 511 // End: