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 (do not block if there is something to compute)
 209             data_receive(real_load > 0.0 ? 0.0 : opt::min_comp_iter_duration);
 210
 211             // send
 212             comm.data_flush(false);
 213             mutex.acquire();
 214             real_load += received_load;
 215             received_load = 0.0;
 216             std::for_each(neigh.begin(), neigh.end(),
 217                           std::bind(&process::data_send, this, _1));
 218             mutex.release();
 219
 220             ++all_comp_iter;
 221
 222         } while (real_load == 0.0);
 223
 224         convergence_check();
 225
 226         // compute
 227         ++comp_iter;
 228         double flops = opt::comp_cost(real_load);
 229         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 230         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 231         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 232         MSG_task_execute(task);
 233         add_comp_amount(flops);
 234         MSG_task_destroy(task);
 235
 236         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 237     }
 238
 239     XBT_VERB("Going to finalize for %s...", __func__);
 240     finalizing = true;
 241     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 242               neigh.size(), ESSE(neigh.size()));
 243     std::for_each(neigh.begin(), neigh.end(),
 244                   std::bind(&process::data_close, this, _1));
 245     while (data_close_pending) {
 246         comm.data_flush(false);
 247         XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);
 248         data_receive(-1.0);
 249     }
 250     real_load += received_load;
 251     received_load = 0.0;
 252     total_load_running -= real_load;
 253     convergence_check();
 254     comm.data_flush(true);
 255 }
 256
 257 void process::convergence_check()
 258 {
 259     double load_diff = std::fabs(real_load - total_load_average);
 260     bool converged = load_diff <= load_diff_threshold;
 261
 262     if (convergence >= 0.0) {
 263         if (!converged) {
 264             XBT_VERB("current load has diverged: %g (%.4g%%)",
 265                      real_load, 100.0 * load_diff / total_load_average);
 266             convergence = -1.0;
 267         }
 268     } else {
 269         if (converged) {
 270             XBT_VERB("current load has converged: %g (%.4g%%)",
 271                      real_load,  100.0 * load_diff / total_load_average);
 272             convergence = MSG_get_clock();
 273         }
 274     }
 275 }
 276
 277 bool process::still_running()
 278 {
 279     static bool last_status = true;
 280
 281     if (!last_status) {
 282         /* nop */
 283
 284     } else if (opt::exit_request) {
 285         XBT_VERB("Global exit requested");
 286         last_status = false;
 287
 288     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 289         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 290         last_status = false;
 291
 292     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 293         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 294         last_status = false;
 295
 296     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 297         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 298         last_status = false;
 299
 300     } else if (opt::exit_on_close && close_received) {
 301         XBT_VERB("Close received");
 302         last_status = false;
 303
 304     } else if (real_load == 0.0 && !data_close_pending) {
 305         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 306         last_status = false;
 307
 308     } else if (100.0 * total_load_running / total_load_init <=
 309                opt::load_ratio_threshold) {
 310         // fixme: this check should be implemented with a distributed
 311         // algorithm, and not a shared global variable!
 312         XBT_VERB("No more load to balance in system.");
 313         last_status = false;
 314     }
 315
 316     return last_status;
 317 }
 318
 319 double process::get_sum_of_to_send() const
 320 {
 321     using std::placeholders::_1;
 322     using std::placeholders::_2;
 323
 324     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 325                            std::bind(std::plus<double>(), _1,
 326                                      std::bind(&neighbor::get_to_send, _2)));
 327 }
 328
 329 void process::load_balance()
 330 {
 331     if (lb_iter == 1)           // warn only once
 332         XBT_WARN("process::load_balance() is a no-op!");
 333 }
 334
 335 void process::send(neighbor& nb, double amount)
 336 {
 337     expected_load -= amount;
 338     nb.set_to_send(nb.get_to_send() + amount);
 339     nb.set_load(nb.get_load() + amount);
 340 }
 341
 342 void process::ctrl_send(neighbor& nb)
 343 {
 344     double info_to_send = expected_load;
 345     double debt_to_send;
 346     if (opt::bookkeeping) {     // bookkeeping
 347         debt_to_send = nb.get_to_send();
 348         if (debt_to_send > 0.0) {
 349             nb.set_to_send(0.0);
 350             nb.set_debt(nb.get_debt() + debt_to_send);
 351         }
 352     } else {                    // !bookkeeping
 353         debt_to_send = 0.0;
 354     }
 355     if (info_to_send != prev_load_broadcast || debt_to_send > 0.0) {
 356         message* msg = new message(message::CTRL, info_to_send, debt_to_send);
 357         add_ctrl_send_mesg(msg->get_size());
 358         comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 359     }
 360 }
 361
 362 double process::compute_load_to_send(double desired)
 363 {
 364     if (opt::integer_transfer)
 365         desired = std::floor(desired);
 366     return desired >= opt::min_transfer_amount ? desired : 0.0;
 367 }
 368
 369 void process::data_send(neighbor& nb)
 370 {
 371     double load_to_send;
 372     if (opt::bookkeeping) {     // bookkeeping
 373         double excess_load;     // load amount we are able to send
 374         if (opt::egocentric)
 375             excess_load = std::max(0.0, real_load - expected_load);
 376         else
 377             excess_load = real_load;
 378
 379         double balance = nb.get_debt() - nb.get_credit();
 380         load_to_send = std::min(excess_load,
 381                                 std::max(0.0, balance));
 382
 383         // adjust load to send (rounding, truncation, etc.)
 384         load_to_send = compute_load_to_send(load_to_send);
 385         if (load_to_send > 0.0)
 386             nb.set_debt(nb.get_debt() - load_to_send);
 387     } else {                    // !bookkeeping
 388         load_to_send = compute_load_to_send(nb.get_to_send());
 389         if (load_to_send > 0.0)
 390             nb.set_to_send(nb.get_to_send() - load_to_send);
 391     }
 392     real_load -= load_to_send;
 393     while (load_to_send > 0.0) {
 394         double amount;
 395         if (opt::max_transfer_amount)
 396             amount = std::min(load_to_send, opt::max_transfer_amount);
 397         else
 398             amount = load_to_send;
 399         message* msg = new message(message::DATA, amount);
 400         add_data_send_mesg(msg->get_size());
 401         comm.data_send(nb.get_data_mbox(), msg);
 402         load_to_send -= amount;
 403     }
 404 }
 405
 406 void process::ctrl_close(neighbor& nb)
 407 {
 408     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 409 }
 410
 411 void process::data_close(neighbor& nb)
 412 {
 413     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 414 }
 415
 416 void process::ctrl_receive(double timeout)
 417 {
 418     message* msg;
 419     m_host_t from;
 420
 421     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 422     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 423         if (msg->get_type() != message::CTRL_CLOSE)
 424             add_ctrl_recv_mesg(msg->get_size());
 425         handle_message(msg, from);
 426         timeout = 0.0;
 427     }
 428 }
 429
 430 void process::data_receive(double timeout)
 431 {
 432     message* msg;
 433     m_host_t from;
 434
 435     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 436     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 437         if (msg->get_type() != message::DATA_CLOSE)
 438             add_data_recv_mesg(msg->get_size());
 439         handle_message(msg, from);
 440         timeout = 0.0;
 441     }
 442 }
 443
 444 void process::handle_message(message* msg, m_host_t from)
 445 {
 446     switch (msg->get_type()) {
 447     case message::CTRL: {
 448         neighbor* n = rev_neigh[from];
 449         n->set_load(msg->get_amount() + n->get_to_send());
 450         if (opt::bookkeeping) {
 451             double credit = msg->get_credit();
 452             expected_load += credit;
 453             n->set_credit(n->get_credit() + credit);
 454         }
 455         break;
 456     }
 457     case message::DATA: {
 458         neighbor* n = rev_neigh[from];
 459         double ld = msg->get_amount();
 460         received_load += ld;
 461         n->set_credit(n->get_credit() - ld);
 462         break;
 463     }
 464     case message::CTRL_CLOSE:
 465         ctrl_close_pending--;
 466         close_received = true;
 467         break;
 468     case message::DATA_CLOSE:
 469         data_close_pending--;
 470         close_received = true;
 471         break;
 472     }
 473     delete msg;
 474 }
 475
 476 #define print_loads_generic(vec, verbose, logp, cat)                    \
 477     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 478         using std::placeholders::_1;                                    \
 479         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 480                   "Neighbor loads:", real_load, expected_load);         \
 481         std::for_each(vec.begin(), vec.end(),                           \
 482                       std::bind(&neighbor::print, _1, verbose, logp, cat)); \
 483     } else ((void)0)
 484
 485 void process::print_loads(bool verbose,
 486                           e_xbt_log_priority_t logp,
 487                           xbt_log_category_t cat) const
 488 {
 489     print_loads_generic(neigh, verbose, logp, cat);
 490 }
 491
 492 void process::print_loads_p(bool verbose,
 493                             e_xbt_log_priority_t logp,
 494                             xbt_log_category_t cat) const
 495 {
 496     print_loads_generic(pneigh, verbose, logp, cat);
 497 }
 498
 499 #undef print_loads_generic
 500
 501 // Local variables:
 502 // mode: c++
 503 // End: