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 int process::process_counter = 0;
  24 double process::total_load_average;
  25 double process::load_diff_threshold;
  26
  27 namespace {
  28
  29     void sleep_until_date(double& date, double duration)
  30     {
  31         double sleep_duration = date - MSG_get_clock();
  32         if (sleep_duration > 0.0)
  33             MSG_process_sleep(sleep_duration);
  34         date = MSG_get_clock() + duration;
  35     }
  36
  37 }
  38
  39 process::process(int argc, char* argv[])
  40 {
  41     if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))
  42         throw std::invalid_argument("bad or missing initial load parameter");
  43
  44     double iload = std::trunc(real_load);
  45     if (opt::integer_transfer && real_load != iload) {
  46         XBT_WARN("Initial load %g is not an integer.  Truncate it.",
  47                  real_load);
  48         real_load = iload;
  49     }
  50
  51     neigh.assign(argv + 2, argv + argc);
  52
  53     pneigh.reserve(neigh.size());
  54     for (unsigned i = 0 ; i < neigh.size() ; i++) {
  55         neighbor* ptr = &neigh[i];
  56         m_host_t host = MSG_get_host_by_name(ptr->get_name());
  57         pneigh.push_back(ptr);
  58         rev_neigh.insert(std::make_pair(host, ptr));
  59     }
  60
  61     // Note: there should not be race condition with the current
  62     // version of Simgrid, when updating the global variables.
  63
  64     prev_load_broadcast = -1;   // force sending of load on first send_all()
  65     expected_load = real_load;
  66     total_load_running += real_load;
  67     total_load_init += real_load;
  68     received_load = 0.0;
  69
  70     convergence = -1.0;
  71
  72     process_counter++;
  73     total_load_average = total_load_running / process_counter;
  74     load_diff_threshold = (opt::load_ratio_threshold +
  75                            opt::avg_load_ratio * total_load_average) / 100.0;
  76
  77     ctrl_close_pending = data_close_pending = neigh.size();
  78     close_received = false;
  79     finalizing = false;
  80
  81     all_comp_iter = comp_iter = lb_iter = 0;
  82
  83     lb_thread = new_msg_thread("loba",
  84                                std::bind(&process::load_balance_loop, this));
  85
  86     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  87     if (!LOG_ISENABLED(logp))
  88         return;
  89     std::ostringstream oss;
  90     oss << neigh.size() << " neighbor";
  91     if (!neigh.empty()) {
  92         oss << ESSE(neigh.size()) << ": ";
  93         std::transform(neigh.begin(), neigh.end() - 1,
  94                        std::ostream_iterator<const char*>(oss, ", "),
  95                        std::mem_fn(&neighbor::get_name));
  96         oss << neigh.back().get_name();
  97     }
  98     XBT_LOG(logp, "Got %s.", oss.str().c_str());
  99     print_loads(false, logp);
 100 }
 101
 102 process::~process()
 103 {
 104     delete lb_thread;
 105     total_load_exit += real_load;
 106     xbt_assert(received_load == 0.0,
 107                "received_load is %g, but should be 0.0 !", received_load);
 108     if (opt::log_rate < 0)
 109         return;
 110     XBT_INFO("Final load after %d:%d:%d iterations: %g",
 111              lb_iter, comp_iter, all_comp_iter, real_load);
 112     if (convergence >= 0.0)
 113         XBT_INFO("Convergence within %g%% was achieved at time %g",
 114                  opt::avg_load_ratio, convergence);
 115     else
 116         XBT_INFO("Convergence within %g%% was not achieved",
 117                  opt::avg_load_ratio);
 118     XBT_VERB("Expected load was: %g", expected_load);
 119     XBT_VERB("Total computation for this process: %g", get_comp_amount());
 120     print_loads(true, xbt_log_priority_debug);
 121 }
 122
 123 int process::run()
 124 {
 125     if (opt::log_rate >= 0) {
 126         XBT_INFO("Initial load: %g", real_load);
 127         XBT_VERB("Initial expected load: %g", expected_load);
 128     }
 129     XBT_VERB("Starting...");
 130     mutex.acquire();
 131     lb_thread->start();
 132     while (lb_iter <= opt::comp_iter_delay)
 133         cond.wait(mutex);
 134     mutex.release();
 135     double sleep_duration = opt::comp_time_delay - MSG_get_clock();
 136     if (sleep_duration > 0.0)
 137         MSG_process_sleep(sleep_duration);
 138     compute_loop();
 139     lb_thread->wait();
 140     XBT_VERB("Done.");
 141     return 0;
 142 }
 143
 144 void process::load_balance_loop()
 145 {
 146     using std::placeholders::_1;
 147
 148     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 149     while (still_running()) {
 150         if (lb_iter == opt::comp_iter_delay) {
 151             mutex.acquire();
 152             ++lb_iter;
 153             cond.signal();
 154             mutex.release();
 155         } else {
 156             ++lb_iter;
 157         }
 158
 159         ctrl_receive(0.0);
 160
 161         mutex.acquire();
 162         if (!opt::bookkeeping)
 163             expected_load = real_load - get_sum_of_to_send();
 164         // nothing to do with opt::bookkeeping
 165
 166         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 167             XBT_INFO("(%u:%u:%u) current load: %g",
 168                      lb_iter, comp_iter, all_comp_iter, real_load);
 169             XBT_VERB("... expected load: %g", expected_load);
 170         }
 171
 172         if (expected_load > 0.0)
 173             load_balance();
 174
 175         print_loads(true, xbt_log_priority_debug);
 176
 177         // send
 178         comm.ctrl_flush(false);
 179         std::for_each(neigh.begin(), neigh.end(),
 180                       std::bind(&process::ctrl_send, this, _1));
 181         prev_load_broadcast = expected_load;
 182         mutex.release();
 183
 184         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 185     }
 186
 187     XBT_VERB("Going to finalize for %s...", __func__);
 188     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 189               neigh.size(), ESSE(neigh.size()));
 190     std::for_each(neigh.begin(), neigh.end(),
 191                   std::bind(&process::ctrl_close, this, _1));
 192     while (ctrl_close_pending) {
 193         comm.ctrl_flush(false);
 194         XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);
 195         ctrl_receive(-1.0);
 196     }
 197     comm.ctrl_flush(true);
 198 }
 199
 200 void process::compute_loop()
 201 {
 202     using std::placeholders::_1;
 203
 204     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 205     while (still_running()) {
 206
 207         do {
 208             // receive
 209             // if there is something to compute, do not block
 210             // else, block the duration of an *lb* iteration
 211             data_receive(real_load > 0.0 ? 0.0 : opt::min_lb_iter_duration);
 212
 213             // send
 214             comm.data_flush(false);
 215             mutex.acquire();
 216             real_load += received_load;
 217             received_load = 0.0;
 218             std::for_each(neigh.begin(), neigh.end(),
 219                           std::bind(&process::data_send, this, _1));
 220             mutex.release();
 221
 222             ++all_comp_iter;
 223
 224         } while (real_load == 0.0);
 225
 226         convergence_check();
 227
 228         // compute
 229         ++comp_iter;
 230         double flops = opt::comp_cost(real_load);
 231         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 232         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 233         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 234         MSG_task_execute(task);
 235         add_comp_amount(flops);
 236         MSG_task_destroy(task);
 237
 238         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 239     }
 240
 241     XBT_VERB("Going to finalize for %s...", __func__);
 242     finalizing = true;
 243     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 244               neigh.size(), ESSE(neigh.size()));
 245     std::for_each(neigh.begin(), neigh.end(),
 246                   std::bind(&process::data_close, this, _1));
 247     while (data_close_pending) {
 248         comm.data_flush(false);
 249         XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);
 250         data_receive(-1.0);
 251     }
 252     real_load += received_load;
 253     received_load = 0.0;
 254     total_load_running -= real_load;
 255     convergence_check();
 256     comm.data_flush(true);
 257 }
 258
 259 void process::convergence_check()
 260 {
 261     double load_diff = std::fabs(real_load - total_load_average);
 262     bool converged = load_diff <= load_diff_threshold;
 263
 264     if (convergence >= 0.0) {
 265         if (!converged) {
 266             XBT_VERB("current load has diverged: %g (%.4g%%)",
 267                      real_load, 100.0 * load_diff / total_load_average);
 268             convergence = -1.0;
 269         }
 270     } else {
 271         if (converged) {
 272             XBT_VERB("current load has converged: %g (%.4g%%)",
 273                      real_load,  100.0 * load_diff / total_load_average);
 274             convergence = MSG_get_clock();
 275         }
 276     }
 277 }
 278
 279 bool process::still_running()
 280 {
 281     static bool last_status = true;
 282
 283     if (!last_status) {
 284         /* nop */
 285
 286     } else if (opt::exit_request) {
 287         XBT_VERB("Global exit requested");
 288         last_status = false;
 289
 290     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 291         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 292         last_status = false;
 293
 294     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 295         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 296         last_status = false;
 297
 298     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 299         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 300         last_status = false;
 301
 302     } else if (opt::exit_on_close && close_received) {
 303         XBT_VERB("Close received");
 304         last_status = false;
 305
 306     } else if (real_load == 0.0 && !data_close_pending) {
 307         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 308         last_status = false;
 309
 310     } else if (100.0 * total_load_running / total_load_init <=
 311                opt::load_ratio_threshold) {
 312         // fixme: this check should be implemented with a distributed
 313         // algorithm, and not a shared global variable!
 314         XBT_VERB("No more load to balance in system.");
 315         last_status = false;
 316     }
 317
 318     return last_status;
 319 }
 320
 321 double process::get_sum_of_to_send() const
 322 {
 323     using std::placeholders::_1;
 324     using std::placeholders::_2;
 325
 326     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 327                            std::bind(std::plus<double>(), _1,
 328                                      std::bind(&neighbor::get_to_send, _2)));
 329 }
 330
 331 void process::load_balance()
 332 {
 333     if (lb_iter == 1)           // warn only once
 334         XBT_WARN("process::load_balance() is a no-op!");
 335 }
 336
 337 void process::send(neighbor& nb, double amount)
 338 {
 339     expected_load -= amount;
 340     nb.set_to_send(nb.get_to_send() + amount);
 341     nb.set_load(nb.get_load() + amount);
 342 }
 343
 344 void process::ctrl_send(neighbor& nb)
 345 {
 346     double info_to_send = expected_load;
 347     double debt_to_send;
 348     if (opt::bookkeeping) {     // bookkeeping
 349         debt_to_send = nb.get_to_send();
 350         if (debt_to_send > 0.0) {
 351             nb.set_to_send(0.0);
 352             nb.set_debt(nb.get_debt() + debt_to_send);
 353         }
 354     } else {                    // !bookkeeping
 355         debt_to_send = 0.0;
 356     }
 357     if (info_to_send != prev_load_broadcast || debt_to_send > 0.0) {
 358         message* msg = new message(message::CTRL, info_to_send, debt_to_send);
 359         add_ctrl_send_mesg(msg->get_size());
 360         comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 361     }
 362 }
 363
 364 double process::compute_load_to_send(double desired)
 365 {
 366     if (opt::integer_transfer)
 367         desired = std::floor(desired);
 368     return desired >= opt::min_transfer_amount ? desired : 0.0;
 369 }
 370
 371 void process::data_send(neighbor& nb)
 372 {
 373     double load_to_send;
 374     if (opt::bookkeeping) {     // bookkeeping
 375         double excess_load;     // load amount we are able to send
 376         if (opt::egocentric)
 377             excess_load = std::max(0.0, real_load - expected_load);
 378         else
 379             excess_load = real_load;
 380
 381         double balance = nb.get_debt() - nb.get_credit();
 382         load_to_send = std::min(excess_load,
 383                                 std::max(0.0, balance));
 384
 385         // adjust load to send (rounding, truncation, etc.)
 386         load_to_send = compute_load_to_send(load_to_send);
 387         if (load_to_send > 0.0)
 388             nb.set_debt(nb.get_debt() - load_to_send);
 389     } else {                    // !bookkeeping
 390         load_to_send = compute_load_to_send(nb.get_to_send());
 391         if (load_to_send > 0.0)
 392             nb.set_to_send(nb.get_to_send() - load_to_send);
 393     }
 394     real_load -= load_to_send;
 395     while (load_to_send > 0.0) {
 396         double amount;
 397         if (opt::max_transfer_amount)
 398             amount = std::min(load_to_send, opt::max_transfer_amount);
 399         else
 400             amount = load_to_send;
 401         message* msg = new message(message::DATA, amount);
 402         add_data_send_mesg(msg->get_size());
 403         comm.data_send(nb.get_data_mbox(), msg);
 404         load_to_send -= amount;
 405     }
 406 }
 407
 408 void process::ctrl_close(neighbor& nb)
 409 {
 410     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 411 }
 412
 413 void process::data_close(neighbor& nb)
 414 {
 415     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 416 }
 417
 418 void process::ctrl_receive(double timeout)
 419 {
 420     message* msg;
 421     m_host_t from;
 422
 423     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 424     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 425         if (msg->get_type() != message::CTRL_CLOSE)
 426             add_ctrl_recv_mesg(msg->get_size());
 427         handle_message(msg, from);
 428         timeout = 0.0;
 429     }
 430 }
 431
 432 void process::data_receive(double timeout)
 433 {
 434     message* msg;
 435     m_host_t from;
 436
 437     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 438     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 439         if (msg->get_type() != message::DATA_CLOSE)
 440             add_data_recv_mesg(msg->get_size());
 441         handle_message(msg, from);
 442         timeout = 0.0;
 443     }
 444 }
 445
 446 void process::handle_message(message* msg, m_host_t from)
 447 {
 448     switch (msg->get_type()) {
 449     case message::CTRL: {
 450         neighbor* n = rev_neigh[from];
 451         n->set_load(msg->get_amount() + n->get_to_send());
 452         if (opt::bookkeeping) {
 453             double credit = msg->get_credit();
 454             expected_load += credit;
 455             n->set_credit(n->get_credit() + credit);
 456         }
 457         break;
 458     }
 459     case message::DATA: {
 460         neighbor* n = rev_neigh[from];
 461         double ld = msg->get_amount();
 462         received_load += ld;
 463         n->set_credit(n->get_credit() - ld);
 464         break;
 465     }
 466     case message::CTRL_CLOSE:
 467         ctrl_close_pending--;
 468         close_received = true;
 469         break;
 470     case message::DATA_CLOSE:
 471         data_close_pending--;
 472         close_received = true;
 473         break;
 474     }
 475     delete msg;
 476 }
 477
 478 #define print_loads_generic(vec, verbose, logp, cat)                    \
 479     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 480         using std::placeholders::_1;                                    \
 481         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 482                   "Neighbor loads:", real_load, expected_load);         \
 483         std::for_each(vec.begin(), vec.end(),                           \
 484                       std::bind(&neighbor::print, _1, verbose, logp, cat)); \
 485     } else ((void)0)
 486
 487 void process::print_loads(bool verbose,
 488                           e_xbt_log_priority_t logp,
 489                           xbt_log_category_t cat) const
 490 {
 491     print_loads_generic(neigh, verbose, logp, cat);
 492 }
 493
 494 void process::print_loads_p(bool verbose,
 495                             e_xbt_log_priority_t logp,
 496                             xbt_log_category_t cat) const
 497 {
 498     print_loads_generic(pneigh, verbose, logp, cat);
 499 }
 500
 501 #undef print_loads_generic
 502
 503 // Local variables:
 504 // mode: c++
 505 // End: