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