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     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 iterations: %g",
  97              lb_iter, 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) current load: %g", lb_iter, comp_iter, real_load);
 148             XBT_VERB("... expected load: %g", expected_load);
 149         }
 150
 151         if (expected_load > 0.0)
 152             load_balance();
 153
 154         print_loads(true, xbt_log_priority_debug);
 155
 156         // send
 157         comm.ctrl_flush(false);
 158         std::for_each(neigh.begin(), neigh.end(),
 159                       std::bind(&process::ctrl_send, this, _1));
 160         prev_load_broadcast = expected_load;
 161         mutex.release();
 162
 163         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 164     }
 165
 166     XBT_VERB("Going to finalize for %s...", __func__);
 167     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 168               neigh.size(), ESSE(neigh.size()));
 169     std::for_each(neigh.begin(), neigh.end(),
 170                   std::bind(&process::ctrl_close, this, _1));
 171     while (ctrl_close_pending) {
 172         comm.ctrl_flush(false);
 173         XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);
 174         ctrl_receive(-1.0);
 175     }
 176     comm.ctrl_flush(true);
 177 }
 178
 179 void process::compute_loop()
 180 {
 181     using std::placeholders::_1;
 182
 183     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 184     while (still_running()) {
 185         // receive (do not block if there is something to compute)
 186         data_receive(real_load > 0.0 ? 0.0 : opt::min_comp_iter_duration);
 187
 188         // send
 189         comm.data_flush(false);
 190         mutex.acquire();
 191         real_load += received_load;
 192         received_load = 0.0;
 193         std::for_each(neigh.begin(), neigh.end(),
 194                       std::bind(&process::data_send, this, _1));
 195         mutex.release();
 196
 197         if (real_load == 0.0)
 198             continue;
 199
 200         // compute
 201         ++comp_iter;
 202         double flops = opt::comp_cost(real_load);
 203         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 204         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 205         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 206         MSG_task_execute(task);
 207         add_comp_amount(flops);
 208         MSG_task_destroy(task);
 209
 210         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 211     }
 212
 213     XBT_VERB("Going to finalize for %s...", __func__);
 214     finalizing = true;
 215     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 216               neigh.size(), ESSE(neigh.size()));
 217     std::for_each(neigh.begin(), neigh.end(),
 218                   std::bind(&process::data_close, this, _1));
 219     while (data_close_pending) {
 220         comm.data_flush(false);
 221         XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);
 222         data_receive(-1.0);
 223     }
 224     real_load += received_load;
 225     received_load = 0.0;
 226     total_load_running -= real_load;
 227     comm.data_flush(true);
 228 }
 229
 230 bool process::still_running()
 231 {
 232     static bool last_status = true;
 233
 234     if (!last_status) {
 235         /* nop */
 236
 237     } else if (opt::exit_request) {
 238         XBT_VERB("Global exit requested");
 239         last_status = false;
 240
 241     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 242         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 243         last_status = false;
 244
 245     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 246         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 247         last_status = false;
 248
 249     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 250         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 251         last_status = false;
 252
 253     } else if (opt::exit_on_close && close_received) {
 254         XBT_VERB("Close received");
 255         last_status = false;
 256
 257     } else if (real_load == 0.0 && !data_close_pending) {
 258         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 259         last_status = false;
 260
 261     } else if (100.0 * total_load_running / total_load_init <=
 262                opt::load_ratio_threshold) {
 263         // fixme: this check should be implemented with a distributed
 264         // algorithm, and not a shared global variable!
 265         XBT_VERB("No more load to balance in system.");
 266         last_status = false;
 267     }
 268
 269     return last_status;
 270 }
 271
 272 double process::get_sum_of_to_send() const
 273 {
 274     using std::placeholders::_1;
 275     using std::placeholders::_2;
 276
 277     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 278                            std::bind(std::plus<double>(), _1,
 279                                      std::bind(&neighbor::get_to_send, _2)));
 280 }
 281
 282 void process::load_balance()
 283 {
 284     if (lb_iter == 1)           // warn only once
 285         XBT_WARN("process::load_balance() is a no-op!");
 286 }
 287
 288 void process::send(neighbor& nb, double amount)
 289 {
 290     expected_load -= amount;
 291     nb.set_to_send(nb.get_to_send() + amount);
 292     nb.set_load(nb.get_load() + amount);
 293 }
 294
 295 void process::ctrl_send(neighbor& nb)
 296 {
 297     double info_to_send = expected_load;
 298     double debt_to_send;
 299     if (opt::bookkeeping) {     // bookkeeping
 300         debt_to_send = nb.get_to_send();
 301         if (debt_to_send > 0.0) {
 302             nb.set_to_send(0.0);
 303             nb.set_debt(nb.get_debt() + debt_to_send);
 304         }
 305     } else {                    // !bookkeeping
 306         debt_to_send = 0.0;
 307     }
 308     if (info_to_send != prev_load_broadcast || debt_to_send > 0.0) {
 309         message* msg = new message(message::CTRL, info_to_send, debt_to_send);
 310         add_ctrl_send_mesg(msg->get_size());
 311         comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 312     }
 313 }
 314
 315 double process::compute_load_to_send(double desired)
 316 {
 317     if (opt::integer_transfer)
 318         desired = std::floor(desired);
 319     return desired >= opt::min_transfer_amount ? desired : 0.0;
 320 }
 321
 322 void process::data_send(neighbor& nb)
 323 {
 324     double load_to_send;
 325     if (opt::bookkeeping) {     // bookkeeping
 326         double excess_load;     // load amount we are able to send
 327         if (opt::egocentric)
 328             excess_load = std::max(0.0, real_load - expected_load);
 329         else
 330             excess_load = real_load;
 331
 332         double balance = nb.get_debt() - nb.get_credit();
 333         load_to_send = std::min(excess_load,
 334                                 std::max(0.0, balance));
 335
 336         // adjust load to send (rounding, truncation, etc.)
 337         load_to_send = compute_load_to_send(load_to_send);
 338         if (load_to_send > 0.0)
 339             nb.set_debt(nb.get_debt() - load_to_send);
 340     } else {                    // !bookkeeping
 341         load_to_send = compute_load_to_send(nb.get_to_send());
 342         if (load_to_send > 0.0)
 343             nb.set_to_send(nb.get_to_send() - load_to_send);
 344     }
 345     real_load -= load_to_send;
 346     while (load_to_send > 0.0) {
 347         double amount;
 348         if (opt::max_transfer_amount)
 349             amount = std::min(load_to_send, opt::max_transfer_amount);
 350         else
 351             amount = load_to_send;
 352         message* msg = new message(message::DATA, amount);
 353         add_data_send_mesg(msg->get_size());
 354         comm.data_send(nb.get_data_mbox(), msg);
 355         load_to_send -= amount;
 356     }
 357 }
 358
 359 void process::ctrl_close(neighbor& nb)
 360 {
 361     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 362 }
 363
 364 void process::data_close(neighbor& nb)
 365 {
 366     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 367 }
 368
 369 void process::ctrl_receive(double timeout)
 370 {
 371     message* msg;
 372     m_host_t from;
 373
 374     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 375     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 376         if (msg->get_type() != message::CTRL_CLOSE)
 377             add_ctrl_recv_mesg(msg->get_size());
 378         handle_message(msg, from);
 379         timeout = 0.0;
 380     }
 381 }
 382
 383 void process::data_receive(double timeout)
 384 {
 385     message* msg;
 386     m_host_t from;
 387
 388     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 389     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 390         if (msg->get_type() != message::DATA_CLOSE)
 391             add_data_recv_mesg(msg->get_size());
 392         handle_message(msg, from);
 393         timeout = 0.0;
 394     }
 395 }
 396
 397 void process::handle_message(message* msg, m_host_t from)
 398 {
 399     switch (msg->get_type()) {
 400     case message::CTRL: {
 401         neighbor* n = rev_neigh[from];
 402         n->set_load(msg->get_amount() + n->get_to_send());
 403         if (opt::bookkeeping) {
 404             double credit = msg->get_credit();
 405             expected_load += credit;
 406             n->set_credit(n->get_credit() + credit);
 407         }
 408         break;
 409     }
 410     case message::DATA: {
 411         neighbor* n = rev_neigh[from];
 412         double ld = msg->get_amount();
 413         received_load += ld;
 414         n->set_credit(n->get_credit() - ld);
 415         break;
 416     }
 417     case message::CTRL_CLOSE:
 418         ctrl_close_pending--;
 419         close_received = true;
 420         break;
 421     case message::DATA_CLOSE:
 422         data_close_pending--;
 423         close_received = true;
 424         break;
 425     }
 426     delete msg;
 427 }
 428
 429 #define print_loads_generic(vec, verbose, logp, cat)                    \
 430     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 431         using std::placeholders::_1;                                    \
 432         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 433                   "Neighbor loads:", real_load, expected_load);         \
 434         std::for_each(vec.begin(), vec.end(),                           \
 435                       std::bind(&neighbor::print, _1, verbose, logp, cat)); \
 436     } else ((void)0)
 437
 438 void process::print_loads(bool verbose,
 439                           e_xbt_log_priority_t logp,
 440                           xbt_log_category_t cat) const
 441 {
 442     print_loads_generic(neigh, verbose, logp, cat);
 443 }
 444
 445 void process::print_loads_p(bool verbose,
 446                             e_xbt_log_priority_t logp,
 447                             xbt_log_category_t cat) const
 448 {
 449     print_loads_generic(pneigh, verbose, logp, cat);
 450 }
 451
 452 #undef print_loads_generic
 453
 454 // Local variables:
 455 // mode: c++
 456 // End: