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
 139         if (expected_load > 0.0)
 140             load_balance();
 141
 142         print_loads(true, xbt_log_priority_debug);
 143
 144         // send
 145         comm.ctrl_flush(false);
 146         std::for_each(neigh.begin(), neigh.end(),
 147                       bind(&process::ctrl_send, this, _1));
 148         prev_load_broadcast = expected_load;
 149         mutex.release();
 150
 151         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 152         ctrl_receive(0.0);
 153     }
 154
 155     XBT_VERB("Going to finalize for %s...", __func__);
 156     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 157               neigh.size(), ESSE(neigh.size()));
 158     std::for_each(neigh.begin(), neigh.end(),
 159                   bind(&process::ctrl_close, this, _1));
 160     while (ctrl_close_pending) {
 161         comm.ctrl_flush(false);
 162         XBT_DEBUG("waiting for %d CTRL CLOSE", ctrl_close_pending);
 163         ctrl_receive(-1.0);
 164     }
 165     comm.ctrl_flush(true);
 166 }
 167
 168 void process::compute_loop()
 169 {
 170     using std::tr1::bind;
 171     using std::tr1::placeholders::_1;
 172
 173     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 174     while (still_running()) {
 175         // receive
 176         mutex.acquire();
 177         if (real_load > 0.0)
 178             data_receive(0.0);
 179         else
 180             data_receive(opt::min_comp_iter_duration);
 181         mutex.release();
 182
 183         // send
 184         comm.data_flush(false);
 185         mutex.acquire();
 186         std::for_each(neigh.begin(), neigh.end(),
 187                       bind(&process::data_send, this, _1));
 188         mutex.release();
 189
 190         if (real_load == 0.0)
 191             continue;
 192
 193         // compute
 194         ++comp_iter;
 195         double flops = opt::comp_cost(real_load);
 196         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 197         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 198         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 199         MSG_task_execute(task);
 200         add_comp_amount(flops);
 201         MSG_task_destroy(task);
 202
 203         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 204     }
 205
 206     XBT_VERB("Going to finalize for %s...", __func__);
 207     // last send, for not losing load scheduled to be sent
 208     std::for_each(neigh.begin(), neigh.end(),
 209                   bind(&process::data_send, this, _1));
 210     finalizing = true;
 211     total_load_running -= real_load;
 212     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 213               neigh.size(), ESSE(neigh.size()));
 214     std::for_each(neigh.begin(), neigh.end(),
 215                   bind(&process::data_close, this, _1));
 216     while (data_close_pending) {
 217         comm.data_flush(false);
 218         XBT_DEBUG("waiting for %d DATA CLOSE", data_close_pending);
 219         data_receive(-1.0);
 220     }
 221     comm.data_flush(true);
 222 }
 223
 224 bool process::still_running()
 225 {
 226     static bool last_status = true;
 227
 228     if (!last_status) {
 229         /* nop */
 230
 231     } else if (opt::exit_request) {
 232         XBT_VERB("Global exit requested");
 233         last_status = false;
 234
 235     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 236         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 237         last_status = false;
 238
 239     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 240         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 241         last_status = false;
 242
 243     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 244         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 245         last_status = false;
 246
 247     } else if (opt::exit_on_close && close_received) {
 248         XBT_VERB("Close received");
 249         last_status = false;
 250
 251     } else if (real_load == 0.0 && !data_close_pending) {
 252         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 253         last_status = false;
 254
 255     } else if (100.0 * total_load_running / total_load_init <=
 256                opt::load_ratio_threshold) {
 257         // fixme: this check should be implemented with a distributed
 258         // algorithm, and not a shared global variable!
 259         XBT_VERB("No more load to balance in system.");
 260         last_status = false;
 261     }
 262
 263     return last_status;
 264 }
 265
 266 double process::get_sum_of_to_send() const
 267 {
 268     using std::tr1::bind;
 269     using std::tr1::placeholders::_1;
 270     using std::tr1::placeholders::_2;
 271
 272     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 273                            bind(std::plus<double>(),
 274                                 _1, bind(&neighbor::get_to_send, _2)));
 275 }
 276
 277 void process::load_balance()
 278 {
 279     if (lb_iter == 1)           // warn only once
 280         XBT_WARN("process::load_balance() is a no-op!");
 281 }
 282
 283 void process::send(neighbor& nb, double amount)
 284 {
 285     expected_load -= amount;
 286     nb.set_to_send(nb.get_to_send() + amount);
 287     nb.set_load(nb.get_load() + amount);
 288 }
 289
 290 void process::ctrl_send(neighbor& nb)
 291 {
 292     double info_to_send = expected_load;
 293     if (info_to_send != prev_load_broadcast) {
 294         message* msg = new message(message::INFO, info_to_send);
 295         add_ctrl_send_mesg(msg->get_size());
 296         comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 297     }
 298     if (opt::bookkeeping) {
 299         double 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             message* msg = new message(message::CREDIT, debt_to_send);
 304             add_ctrl_send_mesg(msg->get_size());
 305             comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 306         }
 307     }
 308 }
 309
 310 void process::data_send(neighbor& nb)
 311 {
 312     double load_to_send;
 313     if (opt::bookkeeping) {
 314         load_to_send = std::min(real_load, nb.get_debt());
 315         if (load_to_send >= opt::min_transfer_amount) {
 316             nb.set_debt(nb.get_debt() - load_to_send);
 317             real_load -= load_to_send;
 318         } else {
 319             load_to_send = 0.0;
 320         }
 321     } else {
 322         load_to_send = nb.get_to_send();
 323         if (load_to_send >= opt::min_transfer_amount) {
 324             nb.set_to_send(0.0);
 325             real_load -= load_to_send;
 326         } else {
 327             load_to_send = 0.0;
 328         }
 329     }
 330     while (load_to_send > 0.0) {
 331         double amount;
 332         if (opt::max_transfer_amount)
 333             amount = std::min(load_to_send, opt::max_transfer_amount);
 334         else
 335             amount = load_to_send;
 336         message* msg = new message(message::LOAD, amount);
 337         add_data_send_mesg(msg->get_size());
 338         comm.data_send(nb.get_data_mbox(), msg);
 339         load_to_send -= amount;
 340     }
 341 }
 342
 343 void process::ctrl_close(neighbor& nb)
 344 {
 345     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 346 }
 347
 348 void process::data_close(neighbor& nb)
 349 {
 350     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 351 }
 352
 353 void process::ctrl_receive(double timeout)
 354 {
 355     message* msg;
 356     m_host_t from;
 357
 358     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 359     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 360         if (msg->get_type() != message::CTRL_CLOSE)
 361             add_ctrl_recv_mesg(msg->get_size());
 362         handle_message(msg, from);
 363         timeout = 0.0;
 364     }
 365 }
 366
 367 void process::data_receive(double timeout)
 368 {
 369     message* msg;
 370     m_host_t from;
 371
 372     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 373     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 374         if (msg->get_type() != message::DATA_CLOSE)
 375             add_data_recv_mesg(msg->get_size());
 376         handle_message(msg, from);
 377         timeout = 0.0;
 378     }
 379 }
 380
 381 void process::handle_message(message* msg, m_host_t from)
 382 {
 383     switch (msg->get_type()) {
 384     case message::INFO: {
 385         neighbor* n = rev_neigh[from];
 386         n->set_load(msg->get_amount() + n->get_to_send());
 387         break;
 388     }
 389     case message::CREDIT:
 390         expected_load += msg->get_amount();
 391         break;
 392     case message::LOAD: {
 393         double ld = msg->get_amount();
 394         real_load += ld;
 395         if (finalizing)
 396             total_load_running -= ld;
 397         break;
 398     }
 399     case message::CTRL_CLOSE:
 400         ctrl_close_pending--;
 401         close_received = true;
 402         break;
 403     case message::DATA_CLOSE:
 404         data_close_pending--;
 405         close_received = true;
 406         break;
 407     }
 408     delete msg;
 409 }
 410
 411 #define print_loads_generic(vec, verbose, logp, cat)                    \
 412     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 413         using std::tr1::bind;                                           \
 414         using std::tr1::placeholders::_1;                               \
 415         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 416                   "Neighbor loads:", real_load, expected_load);         \
 417         std::for_each(vec.begin(), vec.end(),                           \
 418                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 419     } else ((void)0)
 420
 421 void process::print_loads(bool verbose,
 422                           e_xbt_log_priority_t logp,
 423                           xbt_log_category_t cat) const
 424 {
 425     print_loads_generic(neigh, verbose, logp, cat);
 426 }
 427
 428 void process::print_loads_p(bool verbose,
 429                             e_xbt_log_priority_t logp,
 430                             xbt_log_category_t cat) const
 431 {
 432     print_loads_generic(pneigh, verbose, logp, cat);
 433 }
 434
 435 #undef print_loads_generic
 436
 437 // Local variables:
 438 // mode: c++
 439 // End: