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
  26 namespace {
  27
  28     void sleep_until_date(double& date, double duration)
  29     {
  30         double sleep_duration = date - MSG_get_clock();
  31         if (sleep_duration > 0.0)
  32             MSG_process_sleep(sleep_duration);
  33         date = MSG_get_clock() + duration;
  34     }
  35
  36 }
  37
  38 process::process(int argc, char* argv[])
  39 {
  40     if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))
  41         throw std::invalid_argument("bad or missing initial load parameter");
  42
  43     double iload = std::trunc(real_load);
  44     if (opt::integer_transfer && real_load != iload) {
  45         XBT_WARN("Initial load %g is not an integer.  Truncate it.",
  46                  real_load);
  47         real_load = iload;
  48     }
  49
  50     neigh.assign(argv + 2, argv + argc);
  51
  52     pneigh.reserve(neigh.size());
  53     for (unsigned i = 0 ; i < neigh.size() ; i++) {
  54         neighbor* ptr = &neigh[i];
  55         m_host_t host = MSG_get_host_by_name(ptr->get_name());
  56         pneigh.push_back(ptr);
  57         rev_neigh.insert(std::make_pair(host, ptr));
  58     }
  59
  60     // Note: there should not be race condition with the current
  61     // version of Simgrid, when updating the global variables.
  62
  63     prev_load_broadcast = -1;   // force sending of load on first send_all()
  64     expected_load = real_load;
  65     total_load_running += real_load;
  66     total_load_init += real_load;
  67     received_load = 0.0;
  68
  69     convergence = -1.0;
  70
  71     process_counter++;
  72     total_load_average = total_load_running / process_counter;
  73
  74     ctrl_close_pending = data_close_pending = neigh.size();
  75     close_received = false;
  76     finalizing = false;
  77
  78     all_comp_iter = comp_iter = lb_iter = 0;
  79
  80     lb_thread = new_msg_thread("loba",
  81                                std::bind(&process::load_balance_loop, this));
  82
  83     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  84     if (!LOG_ISENABLED(logp))
  85         return;
  86     std::ostringstream oss;
  87     oss << neigh.size() << " neighbor";
  88     if (!neigh.empty()) {
  89         oss << ESSE(neigh.size()) << ": ";
  90         std::transform(neigh.begin(), neigh.end() - 1,
  91                        std::ostream_iterator<const char*>(oss, ", "),
  92                        std::mem_fn(&neighbor::get_name));
  93         oss << neigh.back().get_name();
  94     }
  95     XBT_LOG(logp, "Got %s.", oss.str().c_str());
  96     print_loads(false, logp);
  97 }
  98
  99 process::~process()
 100 {
 101     delete lb_thread;
 102     total_load_exit += real_load;
 103     xbt_assert(received_load == 0.0,
 104                "received_load is %g, but should be 0.0 !", received_load);
 105     if (opt::log_rate < 0)
 106         return;
 107     XBT_INFO("Final load after %d:%d:%d iterations: %g",
 108              lb_iter, comp_iter, all_comp_iter, real_load);
 109     if (convergence >= 0.0)
 110         XBT_INFO("Convergence within %g%% was achieved at time %g",
 111                  opt::avg_load_ratio, convergence);
 112     else
 113         XBT_INFO("Convergence within %g%% was not achieved",
 114                  opt::avg_load_ratio);
 115     XBT_VERB("Expected load was: %g", expected_load);
 116     XBT_VERB("Total computation for this process: %g", get_comp_amount());
 117     print_loads(true, xbt_log_priority_debug);
 118 }
 119
 120 int process::run()
 121 {
 122     if (opt::log_rate >= 0) {
 123         XBT_INFO("Initial load: %g", real_load);
 124         XBT_VERB("Initial expected load: %g", expected_load);
 125     }
 126     XBT_VERB("Starting...");
 127     mutex.acquire();
 128     lb_thread->start();
 129     while (lb_iter <= opt::comp_iter_delay)
 130         cond.wait(mutex);
 131     mutex.release();
 132     double sleep_duration = opt::comp_time_delay - MSG_get_clock();
 133     if (sleep_duration > 0.0)
 134         MSG_process_sleep(sleep_duration);
 135     compute_loop();
 136     lb_thread->wait();
 137     XBT_VERB("Done.");
 138     return 0;
 139 }
 140
 141 void process::load_balance_loop()
 142 {
 143     using std::placeholders::_1;
 144
 145     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 146     while (still_running()) {
 147         if (lb_iter == opt::comp_iter_delay) {
 148             mutex.acquire();
 149             ++lb_iter;
 150             cond.signal();
 151             mutex.release();
 152         } else {
 153             ++lb_iter;
 154         }
 155
 156         ctrl_receive(0.0);
 157
 158         mutex.acquire();
 159         if (!opt::bookkeeping)
 160             expected_load = real_load - get_sum_of_to_send();
 161         // nothing to do with opt::bookkeeping
 162
 163         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 164             XBT_INFO("(%u:%u:%u) current load: %g",
 165                      lb_iter, comp_iter, all_comp_iter, real_load);
 166             XBT_VERB("... expected load: %g", expected_load);
 167         }
 168
 169         if (expected_load > 0.0)
 170             load_balance();
 171
 172         print_loads(true, xbt_log_priority_debug);
 173
 174         // send
 175         comm.ctrl_flush(false);
 176         std::for_each(neigh.begin(), neigh.end(),
 177                       std::bind(&process::ctrl_send, this, _1));
 178         prev_load_broadcast = expected_load;
 179         mutex.release();
 180
 181         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 182     }
 183
 184     XBT_VERB("Going to finalize for %s...", __func__);
 185     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 186               neigh.size(), ESSE(neigh.size()));
 187     std::for_each(neigh.begin(), neigh.end(),
 188                   std::bind(&process::ctrl_close, this, _1));
 189     while (ctrl_close_pending) {
 190         comm.ctrl_flush(false);
 191         XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);
 192         ctrl_receive(-1.0);
 193     }
 194     comm.ctrl_flush(true);
 195 }
 196
 197 void process::compute_loop()
 198 {
 199     using std::placeholders::_1;
 200
 201     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 202     while (still_running()) {
 203         // receive (do not block if there is something to compute)
 204         data_receive(real_load > 0.0 ? 0.0 : opt::min_comp_iter_duration);
 205
 206         // send
 207         comm.data_flush(false);
 208         mutex.acquire();
 209         real_load += received_load;
 210         received_load = 0.0;
 211         std::for_each(neigh.begin(), neigh.end(),
 212                       std::bind(&process::data_send, this, _1));
 213         mutex.release();
 214
 215         ++all_comp_iter;
 216         if (real_load == 0.0)
 217             continue;
 218
 219         double load_ratio =
 220             100.0 * std::fabs(real_load / total_load_average - 1.0);
 221         if (convergence >= 0.0) {
 222             if (load_ratio > opt::avg_load_ratio) {
 223                 XBT_VERB("current load has diverged: %g (%.4g%%)",
 224                          real_load, load_ratio);
 225                 convergence = -1.0;
 226             }
 227         } else {
 228             if (load_ratio <= opt::avg_load_ratio) {
 229                 XBT_VERB("current load has converged: %g (%.4g%%)",
 230                          real_load, load_ratio);
 231                 convergence = MSG_get_clock();
 232             }
 233         }
 234
 235         // compute
 236         ++comp_iter;
 237         double flops = opt::comp_cost(real_load);
 238         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 239         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 240         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 241         MSG_task_execute(task);
 242         add_comp_amount(flops);
 243         MSG_task_destroy(task);
 244
 245         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 246     }
 247
 248     XBT_VERB("Going to finalize for %s...", __func__);
 249     finalizing = true;
 250     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 251               neigh.size(), ESSE(neigh.size()));
 252     std::for_each(neigh.begin(), neigh.end(),
 253                   std::bind(&process::data_close, this, _1));
 254     while (data_close_pending) {
 255         comm.data_flush(false);
 256         XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);
 257         data_receive(-1.0);
 258     }
 259     real_load += received_load;
 260     received_load = 0.0;
 261     total_load_running -= real_load;
 262     comm.data_flush(true);
 263 }
 264
 265 bool process::still_running()
 266 {
 267     static bool last_status = true;
 268
 269     if (!last_status) {
 270         /* nop */
 271
 272     } else if (opt::exit_request) {
 273         XBT_VERB("Global exit requested");
 274         last_status = false;
 275
 276     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 277         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 278         last_status = false;
 279
 280     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 281         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 282         last_status = false;
 283
 284     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 285         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 286         last_status = false;
 287
 288     } else if (opt::exit_on_close && close_received) {
 289         XBT_VERB("Close received");
 290         last_status = false;
 291
 292     } else if (real_load == 0.0 && !data_close_pending) {
 293         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 294         last_status = false;
 295
 296     } else if (100.0 * total_load_running / total_load_init <=
 297                opt::load_ratio_threshold) {
 298         // fixme: this check should be implemented with a distributed
 299         // algorithm, and not a shared global variable!
 300         XBT_VERB("No more load to balance in system.");
 301         last_status = false;
 302     }
 303
 304     return last_status;
 305 }
 306
 307 double process::get_sum_of_to_send() const
 308 {
 309     using std::placeholders::_1;
 310     using std::placeholders::_2;
 311
 312     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 313                            std::bind(std::plus<double>(), _1,
 314                                      std::bind(&neighbor::get_to_send, _2)));
 315 }
 316
 317 void process::load_balance()
 318 {
 319     if (lb_iter == 1)           // warn only once
 320         XBT_WARN("process::load_balance() is a no-op!");
 321 }
 322
 323 void process::send(neighbor& nb, double amount)
 324 {
 325     expected_load -= amount;
 326     nb.set_to_send(nb.get_to_send() + amount);
 327     nb.set_load(nb.get_load() + amount);
 328 }
 329
 330 void process::ctrl_send(neighbor& nb)
 331 {
 332     double info_to_send = expected_load;
 333     double debt_to_send;
 334     if (opt::bookkeeping) {     // bookkeeping
 335         debt_to_send = nb.get_to_send();
 336         if (debt_to_send > 0.0) {
 337             nb.set_to_send(0.0);
 338             nb.set_debt(nb.get_debt() + debt_to_send);
 339         }
 340     } else {                    // !bookkeeping
 341         debt_to_send = 0.0;
 342     }
 343     if (info_to_send != prev_load_broadcast || debt_to_send > 0.0) {
 344         message* msg = new message(message::CTRL, info_to_send, debt_to_send);
 345         add_ctrl_send_mesg(msg->get_size());
 346         comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 347     }
 348 }
 349
 350 double process::compute_load_to_send(double desired)
 351 {
 352     if (opt::integer_transfer)
 353         desired = std::floor(desired);
 354     return desired >= opt::min_transfer_amount ? desired : 0.0;
 355 }
 356
 357 void process::data_send(neighbor& nb)
 358 {
 359     double load_to_send;
 360     if (opt::bookkeeping) {     // bookkeeping
 361         double excess_load;     // load amount we are able to send
 362         if (opt::egocentric)
 363             excess_load = std::max(0.0, real_load - expected_load);
 364         else
 365             excess_load = real_load;
 366
 367         double balance = nb.get_debt() - nb.get_credit();
 368         load_to_send = std::min(excess_load,
 369                                 std::max(0.0, balance));
 370
 371         // adjust load to send (rounding, truncation, etc.)
 372         load_to_send = compute_load_to_send(load_to_send);
 373         if (load_to_send > 0.0)
 374             nb.set_debt(nb.get_debt() - load_to_send);
 375     } else {                    // !bookkeeping
 376         load_to_send = compute_load_to_send(nb.get_to_send());
 377         if (load_to_send > 0.0)
 378             nb.set_to_send(nb.get_to_send() - load_to_send);
 379     }
 380     real_load -= load_to_send;
 381     while (load_to_send > 0.0) {
 382         double amount;
 383         if (opt::max_transfer_amount)
 384             amount = std::min(load_to_send, opt::max_transfer_amount);
 385         else
 386             amount = load_to_send;
 387         message* msg = new message(message::DATA, amount);
 388         add_data_send_mesg(msg->get_size());
 389         comm.data_send(nb.get_data_mbox(), msg);
 390         load_to_send -= amount;
 391     }
 392 }
 393
 394 void process::ctrl_close(neighbor& nb)
 395 {
 396     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 397 }
 398
 399 void process::data_close(neighbor& nb)
 400 {
 401     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 402 }
 403
 404 void process::ctrl_receive(double timeout)
 405 {
 406     message* msg;
 407     m_host_t from;
 408
 409     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 410     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 411         if (msg->get_type() != message::CTRL_CLOSE)
 412             add_ctrl_recv_mesg(msg->get_size());
 413         handle_message(msg, from);
 414         timeout = 0.0;
 415     }
 416 }
 417
 418 void process::data_receive(double timeout)
 419 {
 420     message* msg;
 421     m_host_t from;
 422
 423     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 424     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 425         if (msg->get_type() != message::DATA_CLOSE)
 426             add_data_recv_mesg(msg->get_size());
 427         handle_message(msg, from);
 428         timeout = 0.0;
 429     }
 430 }
 431
 432 void process::handle_message(message* msg, m_host_t from)
 433 {
 434     switch (msg->get_type()) {
 435     case message::CTRL: {
 436         neighbor* n = rev_neigh[from];
 437         n->set_load(msg->get_amount() + n->get_to_send());
 438         if (opt::bookkeeping) {
 439             double credit = msg->get_credit();
 440             expected_load += credit;
 441             n->set_credit(n->get_credit() + credit);
 442         }
 443         break;
 444     }
 445     case message::DATA: {
 446         neighbor* n = rev_neigh[from];
 447         double ld = msg->get_amount();
 448         received_load += ld;
 449         n->set_credit(n->get_credit() - ld);
 450         break;
 451     }
 452     case message::CTRL_CLOSE:
 453         ctrl_close_pending--;
 454         close_received = true;
 455         break;
 456     case message::DATA_CLOSE:
 457         data_close_pending--;
 458         close_received = true;
 459         break;
 460     }
 461     delete msg;
 462 }
 463
 464 #define print_loads_generic(vec, verbose, logp, cat)                    \
 465     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 466         using std::placeholders::_1;                                    \
 467         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 468                   "Neighbor loads:", real_load, expected_load);         \
 469         std::for_each(vec.begin(), vec.end(),                           \
 470                       std::bind(&neighbor::print, _1, verbose, logp, cat)); \
 471     } else ((void)0)
 472
 473 void process::print_loads(bool verbose,
 474                           e_xbt_log_priority_t logp,
 475                           xbt_log_category_t cat) const
 476 {
 477     print_loads_generic(neigh, verbose, logp, cat);
 478 }
 479
 480 void process::print_loads_p(bool verbose,
 481                             e_xbt_log_priority_t logp,
 482                             xbt_log_category_t cat) const
 483 {
 484     print_loads_generic(pneigh, verbose, logp, cat);
 485 }
 486
 487 #undef print_loads_generic
 488
 489 // Local variables:
 490 // mode: c++
 491 // End: