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