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