process.cpp

   1 #include <algorithm>
   2 #include <tr1/functional>
   3 #include <iterator>
   4 #include <numeric>
   5 #include <stdexcept>
   6 #include <sstream>
   7 #include <xbt/log.h>
   8 #include <xbt/time.h>
   9
  10 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
  11
  12 #include "misc.h"
  13 #include "options.h"
  14 #include "tracing.h"
  15
  16 #include "process.h"
  17
  18 double process::total_load_init = 0.0;
  19 double process::total_load_running = 0.0;
  20 double process::total_load_exit = 0.0;
  21
  22 namespace {
  23
  24     void sleep_until_date(double& date, double duration)
  25     {
  26         double sleep_duration = date - MSG_get_clock();
  27         if (sleep_duration > 0.0)
  28             MSG_process_sleep(sleep_duration);
  29         date = MSG_get_clock() + duration;
  30     }
  31
  32 }
  33
  34 process::process(int argc, char* argv[])
  35 {
  36     if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))
  37         throw std::invalid_argument("bad or missing initial load parameter");
  38
  39     neigh.assign(argv + 2, argv + argc);
  40
  41     pneigh.reserve(neigh.size());
  42     for (unsigned i = 0 ; i < neigh.size() ; i++) {
  43         neighbor* ptr = &neigh[i];
  44         m_host_t host = MSG_get_host_by_name(ptr->get_name());
  45         pneigh.push_back(ptr);
  46         rev_neigh.insert(std::make_pair(host, ptr));
  47     }
  48
  49     prev_load_broadcast = -1;   // force sending of load on first send_all()
  50     expected_load = real_load;
  51     total_load_running += real_load;
  52     total_load_init += real_load;
  53
  54     ctrl_close_pending = data_close_pending = neigh.size();
  55     close_received = false;
  56     finalizing = false;
  57
  58     comp_iter = lb_iter = 0;
  59
  60     lb_thread = new_msg_thread("loba",
  61                                std::tr1::bind(&process::load_balance_loop,
  62                                               this));
  63
  64     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  65     if (!LOG_ISENABLED(logp))
  66         return;
  67     std::ostringstream oss;
  68     oss << neigh.size() << " neighbor";
  69     if (!neigh.empty()) {
  70         oss << ESSE(neigh.size()) << ": ";
  71         std::transform(neigh.begin(), neigh.end() - 1,
  72                        std::ostream_iterator<const char*>(oss, ", "),
  73                        std::tr1::mem_fn(&neighbor::get_name));
  74         oss << neigh.back().get_name();
  75     }
  76     XBT_LOG(logp, "Got %s.", oss.str().c_str());
  77     print_loads(false, logp);
  78 }
  79
  80 process::~process()
  81 {
  82     delete lb_thread;
  83     total_load_exit += real_load;
  84     if (opt::log_rate < 0)
  85         return;
  86     XBT_INFO("Final load after %d:%d iterations: %g",
  87              lb_iter, comp_iter, real_load);
  88     XBT_VERB("Expected load was: %g", expected_load);
  89     XBT_VERB("Total computation for this process: %g", get_comp_amount());
  90 }
  91
  92 int process::run()
  93 {
  94     if (opt::log_rate >= 0) {
  95         XBT_INFO("Initial load: %g", real_load);
  96         XBT_VERB("Initial expected load: %g", expected_load);
  97     }
  98     XBT_VERB("Starting...");
  99     mutex.acquire();
 100     lb_thread->start();
 101     while (lb_iter <= opt::comp_iter_delay)
 102         cond.wait(mutex);
 103     mutex.release();
 104     double sleep_duration = opt::comp_time_delay - MSG_get_clock();
 105     if (sleep_duration > 0.0)
 106         MSG_process_sleep(sleep_duration);
 107     compute_loop();
 108     lb_thread->wait();
 109     XBT_VERB("Done.");
 110     return 0;
 111 }
 112
 113 void process::load_balance_loop()
 114 {
 115     using std::tr1::bind;
 116     using std::tr1::placeholders::_1;
 117
 118     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 119     while (still_running()) {
 120         if (lb_iter == opt::comp_iter_delay) {
 121             mutex.acquire();
 122             ++lb_iter;
 123             cond.signal();
 124             mutex.release();
 125         } else {
 126             ++lb_iter;
 127         }
 128
 129         mutex.acquire();
 130         if (!opt::bookkeeping)
 131             expected_load = real_load - get_sum_of_to_send();
 132         // nothing to do with opt::bookkeeping
 133
 134         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 135             XBT_INFO("(%u:%u) current load: %g", lb_iter, comp_iter, real_load);
 136             XBT_VERB("... expected load: %g", expected_load);
 137         }
 138         mutex.release();
 139
 140         if (expected_load > 0.0)
 141             load_balance();
 142
 143         print_loads(true, xbt_log_priority_debug);
 144
 145         // send
 146         std::for_each(neigh.begin(), neigh.end(),
 147                       bind(&process::ctrl_send, this, _1));
 148         prev_load_broadcast = expected_load;
 149
 150         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 151         ctrl_receive(0.0);
 152
 153         comm.ctrl_flush(false);
 154     }
 155
 156     XBT_VERB("Going to finalize for %s...", __func__);
 157     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 158               neigh.size(), ESSE(neigh.size()));
 159     std::for_each(neigh.begin(), neigh.end(),
 160                   bind(&process::ctrl_close, this, _1));
 161     while (ctrl_close_pending) {
 162         comm.ctrl_flush(false);
 163         XBT_DEBUG("waiting for %d CTRL CLOSE", ctrl_close_pending);
 164         ctrl_receive(-1.0);
 165     }
 166     comm.ctrl_flush(true);
 167 }
 168
 169 void process::compute_loop()
 170 {
 171     using std::tr1::bind;
 172     using std::tr1::placeholders::_1;
 173
 174     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 175     while (still_running()) {
 176         // receive
 177         mutex.acquire();
 178         if (real_load > 0.0)
 179             data_receive(0.0);
 180         else
 181             data_receive(opt::min_comp_iter_duration);
 182         mutex.release();
 183
 184         comm.data_flush(false);
 185
 186         // send
 187         mutex.acquire();
 188         std::for_each(neigh.begin(), neigh.end(),
 189                       bind(&process::data_send, this, _1));
 190         mutex.release();
 191
 192         if (real_load == 0.0)
 193             continue;
 194
 195         // compute
 196         ++comp_iter;
 197         double flops = opt::comp_cost(real_load);
 198         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 199         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 200         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 201         MSG_task_execute(task);
 202         add_comp_amount(flops);
 203         MSG_task_destroy(task);
 204
 205         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 206     }
 207
 208     XBT_VERB("Going to finalize for %s...", __func__);
 209     // last send, for not losing load scheduled to be sent
 210     std::for_each(neigh.begin(), neigh.end(),
 211                   bind(&process::data_send, this, _1));
 212     finalizing = true;
 213     total_load_running -= real_load;
 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                   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     comm.data_flush(true);
 224 }
 225
 226 bool process::still_running()
 227 {
 228     static bool last_status = true;
 229
 230     if (!last_status) {
 231         /* nop */
 232
 233     } else if (opt::exit_request) {
 234         XBT_VERB("Global exit requested");
 235         last_status = false;
 236
 237     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 238         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 239         last_status = false;
 240
 241     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 242         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 243         last_status = false;
 244
 245     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 246         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 247         last_status = false;
 248
 249     } else if (opt::exit_on_close && close_received) {
 250         XBT_VERB("Close received");
 251         last_status = false;
 252
 253     } else if (real_load == 0.0 && !data_close_pending) {
 254         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 255         last_status = false;
 256
 257     } else if (100.0 * total_load_running / total_load_init <=
 258                opt::load_ratio_threshold) {
 259         // fixme: this check should be implemented with a distributed
 260         // algorithm, and not a shared global variable!
 261         XBT_VERB("No more load to balance in system.");
 262         last_status = false;
 263     }
 264
 265     return last_status;
 266 }
 267
 268 double process::get_sum_of_to_send() const
 269 {
 270     using std::tr1::bind;
 271     using std::tr1::placeholders::_1;
 272     using std::tr1::placeholders::_2;
 273
 274     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 275                            bind(std::plus<double>(),
 276                                 _1, bind(&neighbor::get_to_send, _2)));
 277 }
 278
 279 void process::load_balance()
 280 {
 281     if (lb_iter == 1)           // warn only once
 282         XBT_WARN("process::load_balance() is a no-op!");
 283 }
 284
 285 void process::send(neighbor& nb, double amount)
 286 {
 287     expected_load -= amount;
 288     nb.set_to_send(nb.get_to_send() + amount);
 289     nb.set_load(nb.get_load() + amount);
 290 }
 291
 292 void process::ctrl_send(neighbor& nb)
 293 {
 294     double info_to_send = expected_load;
 295     if (info_to_send != prev_load_broadcast) {
 296         message* msg = new message(message::INFO, info_to_send);
 297         add_ctrl_send_mesg(msg->get_size());
 298         comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 299     }
 300     if (opt::bookkeeping) {
 301         double debt_to_send = nb.get_to_send();
 302         if (debt_to_send > 0.0) {
 303             nb.set_to_send(0.0);
 304             nb.set_debt(nb.get_debt() + debt_to_send);
 305             message* msg = new message(message::CREDIT, debt_to_send);
 306             add_ctrl_send_mesg(msg->get_size());
 307             comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 308         }
 309     }
 310 }
 311
 312 void process::data_send(neighbor& nb)
 313 {
 314     double load_to_send;
 315     if (opt::bookkeeping) {
 316         load_to_send = std::min(real_load, nb.get_debt());
 317         if (load_to_send >= opt::min_transfer_amount) {
 318             nb.set_debt(nb.get_debt() - load_to_send);
 319             real_load -= load_to_send;
 320         } else {
 321             load_to_send = 0.0;
 322         }
 323     } else {
 324         load_to_send = nb.get_to_send();
 325         if (load_to_send >= opt::min_transfer_amount) {
 326             nb.set_to_send(0.0);
 327             real_load -= load_to_send;
 328         } else {
 329             load_to_send = 0.0;
 330         }
 331     }
 332     while (load_to_send > 0.0) {
 333         double amount;
 334         if (opt::max_transfer_amount)
 335             amount = std::min(load_to_send, opt::max_transfer_amount);
 336         else
 337             amount = load_to_send;
 338         message* msg = new message(message::LOAD, amount);
 339         add_data_send_mesg(msg->get_size());
 340         comm.data_send(nb.get_data_mbox(), msg);
 341         load_to_send -= amount;
 342     }
 343 }
 344
 345 void process::ctrl_close(neighbor& nb)
 346 {
 347     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 348 }
 349
 350 void process::data_close(neighbor& nb)
 351 {
 352     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 353 }
 354
 355 void process::ctrl_receive(double timeout)
 356 {
 357     message* msg;
 358     m_host_t from;
 359
 360     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 361     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 362         if (msg->get_type() != message::CTRL_CLOSE)
 363             add_ctrl_recv_mesg(msg->get_size());
 364         handle_message(msg, from);
 365         timeout = 0.0;
 366     }
 367 }
 368
 369 void process::data_receive(double timeout)
 370 {
 371     message* msg;
 372     m_host_t from;
 373
 374     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 375     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 376         if (msg->get_type() != message::DATA_CLOSE)
 377             add_data_recv_mesg(msg->get_size());
 378         handle_message(msg, from);
 379         timeout = 0.0;
 380     }
 381 }
 382
 383 void process::handle_message(message* msg, m_host_t from)
 384 {
 385     switch (msg->get_type()) {
 386     case message::INFO: {
 387         neighbor* n = rev_neigh[from];
 388         n->set_load(msg->get_amount() + n->get_to_send());
 389         break;
 390     }
 391     case message::CREDIT:
 392         expected_load += msg->get_amount();
 393         break;
 394     case message::LOAD: {
 395         double ld = msg->get_amount();
 396         real_load += ld;
 397         if (finalizing)
 398             total_load_running -= ld;
 399         break;
 400     }
 401     case message::CTRL_CLOSE:
 402         ctrl_close_pending--;
 403         close_received = true;
 404         break;
 405     case message::DATA_CLOSE:
 406         data_close_pending--;
 407         close_received = true;
 408         break;
 409     }
 410     delete msg;
 411 }
 412
 413 #define print_loads_generic(vec, verbose, logp, cat)                    \
 414     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 415         using std::tr1::bind;                                           \
 416         using std::tr1::placeholders::_1;                               \
 417         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 418                   "Neighbor loads:", real_load, expected_load);         \
 419         std::for_each(vec.begin(), vec.end(),                           \
 420                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 421     } else ((void)0)
 422
 423 void process::print_loads(bool verbose,
 424                           e_xbt_log_priority_t logp,
 425                           xbt_log_category_t cat) const
 426 {
 427     print_loads_generic(neigh, verbose, logp, cat);
 428 }
 429
 430 void process::print_loads_p(bool verbose,
 431                             e_xbt_log_priority_t logp,
 432                             xbt_log_category_t cat) const
 433 {
 434     print_loads_generic(pneigh, verbose, logp, cat);
 435 }
 436
 437 #undef print_loads_generic
 438
 439 // Local variables:
 440 // mode: c++
 441 // End: