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 namespace {
  24
  25     void sleep_until_date(double& date, double duration)
  26     {
  27         double sleep_duration = date - MSG_get_clock();
  28         if (sleep_duration > 0.0)
  29             MSG_process_sleep(sleep_duration);
  30         date = MSG_get_clock() + duration;
  31     }
  32
  33 }
  34
  35 process::process(int argc, char* argv[])
  36 {
  37     if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))
  38         throw std::invalid_argument("bad or missing initial load parameter");
  39
  40     double iload = std::trunc(real_load);
  41     if (opt::integer_transfer && real_load != iload) {
  42         XBT_WARN("Initial load %g is not an integer.  Truncate it.",
  43                  real_load);
  44         real_load = iload;
  45     }
  46
  47     neigh.assign(argv + 2, argv + argc);
  48
  49     pneigh.reserve(neigh.size());
  50     for (unsigned i = 0 ; i < neigh.size() ; i++) {
  51         neighbor* ptr = &neigh[i];
  52         m_host_t host = MSG_get_host_by_name(ptr->get_name());
  53         pneigh.push_back(ptr);
  54         rev_neigh.insert(std::make_pair(host, ptr));
  55     }
  56
  57     prev_load_broadcast = -1;   // force sending of load on first send_all()
  58     expected_load = real_load;
  59     total_load_running += real_load;
  60     total_load_init += real_load;
  61     received_load = 0.0;
  62
  63     ctrl_close_pending = data_close_pending = neigh.size();
  64     close_received = false;
  65     finalizing = false;
  66
  67     all_comp_iter = comp_iter = lb_iter = 0;
  68
  69     lb_thread = new_msg_thread("loba",
  70                                std::bind(&process::load_balance_loop, this));
  71
  72     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  73     if (!LOG_ISENABLED(logp))
  74         return;
  75     std::ostringstream oss;
  76     oss << neigh.size() << " neighbor";
  77     if (!neigh.empty()) {
  78         oss << ESSE(neigh.size()) << ": ";
  79         std::transform(neigh.begin(), neigh.end() - 1,
  80                        std::ostream_iterator<const char*>(oss, ", "),
  81                        std::mem_fn(&neighbor::get_name));
  82         oss << neigh.back().get_name();
  83     }
  84     XBT_LOG(logp, "Got %s.", oss.str().c_str());
  85     print_loads(false, logp);
  86 }
  87
  88 process::~process()
  89 {
  90     delete lb_thread;
  91     total_load_exit += real_load;
  92     xbt_assert(received_load == 0.0,
  93                "received_load is %g, but should be 0.0 !", received_load);
  94     if (opt::log_rate < 0)
  95         return;
  96     XBT_INFO("Final load after %d:%d:%d iterations: %g",
  97              lb_iter, comp_iter, all_comp_iter, real_load);
  98     XBT_VERB("Expected load was: %g", expected_load);
  99     XBT_VERB("Total computation for this process: %g", get_comp_amount());
 100     print_loads(true, xbt_log_priority_debug);
 101 }
 102
 103 int process::run()
 104 {
 105     if (opt::log_rate >= 0) {
 106         XBT_INFO("Initial load: %g", real_load);
 107         XBT_VERB("Initial expected load: %g", expected_load);
 108     }
 109     XBT_VERB("Starting...");
 110     mutex.acquire();
 111     lb_thread->start();
 112     while (lb_iter <= opt::comp_iter_delay)
 113         cond.wait(mutex);
 114     mutex.release();
 115     double sleep_duration = opt::comp_time_delay - MSG_get_clock();
 116     if (sleep_duration > 0.0)
 117         MSG_process_sleep(sleep_duration);
 118     compute_loop();
 119     lb_thread->wait();
 120     XBT_VERB("Done.");
 121     return 0;
 122 }
 123
 124 void process::load_balance_loop()
 125 {
 126     using std::placeholders::_1;
 127
 128     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 129     while (still_running()) {
 130         if (lb_iter == opt::comp_iter_delay) {
 131             mutex.acquire();
 132             ++lb_iter;
 133             cond.signal();
 134             mutex.release();
 135         } else {
 136             ++lb_iter;
 137         }
 138
 139         ctrl_receive(0.0);
 140
 141         mutex.acquire();
 142         if (!opt::bookkeeping)
 143             expected_load = real_load - get_sum_of_to_send();
 144         // nothing to do with opt::bookkeeping
 145
 146         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 147             XBT_INFO("(%u:%u:%u) current load: %g",
 148                      lb_iter, comp_iter, all_comp_iter, real_load);
 149             XBT_VERB("... expected load: %g", expected_load);
 150         }
 151
 152         if (expected_load > 0.0)
 153             load_balance();
 154
 155         print_loads(true, xbt_log_priority_debug);
 156
 157         // send
 158         comm.ctrl_flush(false);
 159         std::for_each(neigh.begin(), neigh.end(),
 160                       std::bind(&process::ctrl_send, this, _1));
 161         prev_load_broadcast = expected_load;
 162         mutex.release();
 163
 164         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 165     }
 166
 167     XBT_VERB("Going to finalize for %s...", __func__);
 168     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 169               neigh.size(), ESSE(neigh.size()));
 170     std::for_each(neigh.begin(), neigh.end(),
 171                   std::bind(&process::ctrl_close, this, _1));
 172     while (ctrl_close_pending) {
 173         comm.ctrl_flush(false);
 174         XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);
 175         ctrl_receive(-1.0);
 176     }
 177     comm.ctrl_flush(true);
 178 }
 179
 180 void process::compute_loop()
 181 {
 182     using std::placeholders::_1;
 183
 184     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 185     while (still_running()) {
 186         // receive (do not block if there is something to compute)
 187         data_receive(real_load > 0.0 ? 0.0 : opt::min_comp_iter_duration);
 188
 189         // send
 190         comm.data_flush(false);
 191         mutex.acquire();
 192         real_load += received_load;
 193         received_load = 0.0;
 194         std::for_each(neigh.begin(), neigh.end(),
 195                       std::bind(&process::data_send, this, _1));
 196         mutex.release();
 197
 198         ++all_comp_iter;
 199         if (real_load == 0.0)
 200             continue;
 201
 202         // compute
 203         ++comp_iter;
 204         double flops = opt::comp_cost(real_load);
 205         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 206         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 207         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 208         MSG_task_execute(task);
 209         add_comp_amount(flops);
 210         MSG_task_destroy(task);
 211
 212         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 213     }
 214
 215     XBT_VERB("Going to finalize for %s...", __func__);
 216     finalizing = true;
 217     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 218               neigh.size(), ESSE(neigh.size()));
 219     std::for_each(neigh.begin(), neigh.end(),
 220                   std::bind(&process::data_close, this, _1));
 221     while (data_close_pending) {
 222         comm.data_flush(false);
 223         XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);
 224         data_receive(-1.0);
 225     }
 226     real_load += received_load;
 227     received_load = 0.0;
 228     total_load_running -= real_load;
 229     comm.data_flush(true);
 230 }
 231
 232 bool process::still_running()
 233 {
 234     static bool last_status = true;
 235
 236     if (!last_status) {
 237         /* nop */
 238
 239     } else if (opt::exit_request) {
 240         XBT_VERB("Global exit requested");
 241         last_status = false;
 242
 243     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 244         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 245         last_status = false;
 246
 247     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 248         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 249         last_status = false;
 250
 251     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 252         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 253         last_status = false;
 254
 255     } else if (opt::exit_on_close && close_received) {
 256         XBT_VERB("Close received");
 257         last_status = false;
 258
 259     } else if (real_load == 0.0 && !data_close_pending) {
 260         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 261         last_status = false;
 262
 263     } else if (100.0 * total_load_running / total_load_init <=
 264                opt::load_ratio_threshold) {
 265         // fixme: this check should be implemented with a distributed
 266         // algorithm, and not a shared global variable!
 267         XBT_VERB("No more load to balance in system.");
 268         last_status = false;
 269     }
 270
 271     return last_status;
 272 }
 273
 274 double process::get_sum_of_to_send() const
 275 {
 276     using std::placeholders::_1;
 277     using std::placeholders::_2;
 278
 279     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 280                            std::bind(std::plus<double>(), _1,
 281                                      std::bind(&neighbor::get_to_send, _2)));
 282 }
 283
 284 void process::load_balance()
 285 {
 286     if (lb_iter == 1)           // warn only once
 287         XBT_WARN("process::load_balance() is a no-op!");
 288 }
 289
 290 void process::send(neighbor& nb, double amount)
 291 {
 292     expected_load -= amount;
 293     nb.set_to_send(nb.get_to_send() + amount);
 294     nb.set_load(nb.get_load() + amount);
 295 }
 296
 297 void process::ctrl_send(neighbor& nb)
 298 {
 299     double info_to_send = expected_load;
 300     double debt_to_send;
 301     if (opt::bookkeeping) {     // bookkeeping
 302         debt_to_send = nb.get_to_send();
 303         if (debt_to_send > 0.0) {
 304             nb.set_to_send(0.0);
 305             nb.set_debt(nb.get_debt() + debt_to_send);
 306         }
 307     } else {                    // !bookkeeping
 308         debt_to_send = 0.0;
 309     }
 310     if (info_to_send != prev_load_broadcast || debt_to_send > 0.0) {
 311         message* msg = new message(message::CTRL, info_to_send, debt_to_send);
 312         add_ctrl_send_mesg(msg->get_size());
 313         comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 314     }
 315 }
 316
 317 double process::compute_load_to_send(double desired)
 318 {
 319     if (opt::integer_transfer)
 320         desired = std::floor(desired);
 321     return desired >= opt::min_transfer_amount ? desired : 0.0;
 322 }
 323
 324 void process::data_send(neighbor& nb)
 325 {
 326     double load_to_send;
 327     if (opt::bookkeeping) {     // bookkeeping
 328         double excess_load;     // load amount we are able to send
 329         if (opt::egocentric)
 330             excess_load = std::max(0.0, real_load - expected_load);
 331         else
 332             excess_load = real_load;
 333
 334         double balance = nb.get_debt() - nb.get_credit();
 335         load_to_send = std::min(excess_load,
 336                                 std::max(0.0, balance));
 337
 338         // adjust load to send (rounding, truncation, etc.)
 339         load_to_send = compute_load_to_send(load_to_send);
 340         if (load_to_send > 0.0)
 341             nb.set_debt(nb.get_debt() - load_to_send);
 342     } else {                    // !bookkeeping
 343         load_to_send = compute_load_to_send(nb.get_to_send());
 344         if (load_to_send > 0.0)
 345             nb.set_to_send(nb.get_to_send() - load_to_send);
 346     }
 347     real_load -= load_to_send;
 348     while (load_to_send > 0.0) {
 349         double amount;
 350         if (opt::max_transfer_amount)
 351             amount = std::min(load_to_send, opt::max_transfer_amount);
 352         else
 353             amount = load_to_send;
 354         message* msg = new message(message::DATA, amount);
 355         add_data_send_mesg(msg->get_size());
 356         comm.data_send(nb.get_data_mbox(), msg);
 357         load_to_send -= amount;
 358     }
 359 }
 360
 361 void process::ctrl_close(neighbor& nb)
 362 {
 363     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 364 }
 365
 366 void process::data_close(neighbor& nb)
 367 {
 368     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 369 }
 370
 371 void process::ctrl_receive(double timeout)
 372 {
 373     message* msg;
 374     m_host_t from;
 375
 376     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 377     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 378         if (msg->get_type() != message::CTRL_CLOSE)
 379             add_ctrl_recv_mesg(msg->get_size());
 380         handle_message(msg, from);
 381         timeout = 0.0;
 382     }
 383 }
 384
 385 void process::data_receive(double timeout)
 386 {
 387     message* msg;
 388     m_host_t from;
 389
 390     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 391     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 392         if (msg->get_type() != message::DATA_CLOSE)
 393             add_data_recv_mesg(msg->get_size());
 394         handle_message(msg, from);
 395         timeout = 0.0;
 396     }
 397 }
 398
 399 void process::handle_message(message* msg, m_host_t from)
 400 {
 401     switch (msg->get_type()) {
 402     case message::CTRL: {
 403         neighbor* n = rev_neigh[from];
 404         n->set_load(msg->get_amount() + n->get_to_send());
 405         if (opt::bookkeeping) {
 406             double credit = msg->get_credit();
 407             expected_load += credit;
 408             n->set_credit(n->get_credit() + credit);
 409         }
 410         break;
 411     }
 412     case message::DATA: {
 413         neighbor* n = rev_neigh[from];
 414         double ld = msg->get_amount();
 415         received_load += ld;
 416         n->set_credit(n->get_credit() - ld);
 417         break;
 418     }
 419     case message::CTRL_CLOSE:
 420         ctrl_close_pending--;
 421         close_received = true;
 422         break;
 423     case message::DATA_CLOSE:
 424         data_close_pending--;
 425         close_received = true;
 426         break;
 427     }
 428     delete msg;
 429 }
 430
 431 #define print_loads_generic(vec, verbose, logp, cat)                    \
 432     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 433         using std::placeholders::_1;                                    \
 434         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 435                   "Neighbor loads:", real_load, expected_load);         \
 436         std::for_each(vec.begin(), vec.end(),                           \
 437                       std::bind(&neighbor::print, _1, verbose, logp, cat)); \
 438     } else ((void)0)
 439
 440 void process::print_loads(bool verbose,
 441                           e_xbt_log_priority_t logp,
 442                           xbt_log_category_t cat) const
 443 {
 444     print_loads_generic(neigh, verbose, logp, cat);
 445 }
 446
 447 void process::print_loads_p(bool verbose,
 448                             e_xbt_log_priority_t logp,
 449                             xbt_log_category_t cat) const
 450 {
 451     print_loads_generic(pneigh, verbose, logp, cat);
 452 }
 453
 454 #undef print_loads_generic
 455
 456 // Local variables:
 457 // mode: c++
 458 // End: