process.cpp

   1 #include <algorithm>
   2 #include <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     received_load = 0.0;
  54
  55     ctrl_close_pending = data_close_pending = neigh.size();
  56     close_received = false;
  57     finalizing = false;
  58
  59     comp_iter = lb_iter = 0;
  60
  61     lb_thread = new_msg_thread("loba",
  62                                std::bind(&process::load_balance_loop, 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::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     xbt_assert(received_load == 0.0,
  85                "received_load is %g, but should be 0.0 !", received_load);
  86     if (opt::log_rate < 0)
  87         return;
  88     XBT_INFO("Final load after %d:%d iterations: %g",
  89              lb_iter, comp_iter, real_load);
  90     XBT_VERB("Expected load was: %g", expected_load);
  91     XBT_VERB("Total computation for this process: %g", get_comp_amount());
  92 }
  93
  94 int process::run()
  95 {
  96     if (opt::log_rate >= 0) {
  97         XBT_INFO("Initial load: %g", real_load);
  98         XBT_VERB("Initial expected load: %g", expected_load);
  99     }
 100     XBT_VERB("Starting...");
 101     mutex.acquire();
 102     lb_thread->start();
 103     while (lb_iter <= opt::comp_iter_delay)
 104         cond.wait(mutex);
 105     mutex.release();
 106     double sleep_duration = opt::comp_time_delay - MSG_get_clock();
 107     if (sleep_duration > 0.0)
 108         MSG_process_sleep(sleep_duration);
 109     compute_loop();
 110     lb_thread->wait();
 111     XBT_VERB("Done.");
 112     return 0;
 113 }
 114
 115 void process::load_balance_loop()
 116 {
 117     using std::placeholders::_1;
 118
 119     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 120     while (still_running()) {
 121         if (lb_iter == opt::comp_iter_delay) {
 122             mutex.acquire();
 123             ++lb_iter;
 124             cond.signal();
 125             mutex.release();
 126         } else {
 127             ++lb_iter;
 128         }
 129
 130         ctrl_receive(0.0);
 131
 132         mutex.acquire();
 133         if (!opt::bookkeeping)
 134             expected_load = real_load - get_sum_of_to_send();
 135         // nothing to do with opt::bookkeeping
 136
 137         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 138             XBT_INFO("(%u:%u) current load: %g", lb_iter, comp_iter, real_load);
 139             XBT_VERB("... expected load: %g", expected_load);
 140         }
 141
 142         if (expected_load > 0.0)
 143             load_balance();
 144
 145         print_loads(true, xbt_log_priority_debug);
 146
 147         // send
 148         comm.ctrl_flush(false);
 149         std::for_each(neigh.begin(), neigh.end(),
 150                       std::bind(&process::ctrl_send, this, _1));
 151         prev_load_broadcast = expected_load;
 152         mutex.release();
 153
 154         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 155     }
 156
 157     XBT_VERB("Going to finalize for %s...", __func__);
 158     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 159               neigh.size(), ESSE(neigh.size()));
 160     std::for_each(neigh.begin(), neigh.end(),
 161                   std::bind(&process::ctrl_close, this, _1));
 162     while (ctrl_close_pending) {
 163         comm.ctrl_flush(false);
 164         XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);
 165         ctrl_receive(-1.0);
 166     }
 167     comm.ctrl_flush(true);
 168 }
 169
 170 void process::compute_loop()
 171 {
 172     using std::placeholders::_1;
 173
 174     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 175     while (still_running()) {
 176         // receive (do not block if there is something to compute)
 177         data_receive(real_load > 0.0 ? 0.0 : opt::min_comp_iter_duration);
 178
 179         // send
 180         comm.data_flush(false);
 181         mutex.acquire();
 182         real_load += received_load;
 183         received_load = 0.0;
 184         std::for_each(neigh.begin(), neigh.end(),
 185                       std::bind(&process::data_send, this, _1));
 186         mutex.release();
 187
 188         if (real_load == 0.0)
 189             continue;
 190
 191         // compute
 192         ++comp_iter;
 193         double flops = opt::comp_cost(real_load);
 194         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 195         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 196         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 197         MSG_task_execute(task);
 198         add_comp_amount(flops);
 199         MSG_task_destroy(task);
 200
 201         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 202     }
 203
 204     XBT_VERB("Going to finalize for %s...", __func__);
 205     finalizing = true;
 206     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 207               neigh.size(), ESSE(neigh.size()));
 208     std::for_each(neigh.begin(), neigh.end(),
 209                   std::bind(&process::data_close, this, _1));
 210     while (data_close_pending) {
 211         comm.data_flush(false);
 212         XBT_DEBUG("waiting for %d DATA_CLOSE", data_close_pending);
 213         data_receive(-1.0);
 214     }
 215     real_load += received_load;
 216     received_load = 0.0;
 217     total_load_running -= real_load;
 218     comm.data_flush(true);
 219 }
 220
 221 bool process::still_running()
 222 {
 223     static bool last_status = true;
 224
 225     if (!last_status) {
 226         /* nop */
 227
 228     } else if (opt::exit_request) {
 229         XBT_VERB("Global exit requested");
 230         last_status = false;
 231
 232     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 233         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 234         last_status = false;
 235
 236     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 237         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 238         last_status = false;
 239
 240     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 241         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 242         last_status = false;
 243
 244     } else if (opt::exit_on_close && close_received) {
 245         XBT_VERB("Close received");
 246         last_status = false;
 247
 248     } else if (real_load == 0.0 && !data_close_pending) {
 249         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 250         last_status = false;
 251
 252     } else if (100.0 * total_load_running / total_load_init <=
 253                opt::load_ratio_threshold) {
 254         // fixme: this check should be implemented with a distributed
 255         // algorithm, and not a shared global variable!
 256         XBT_VERB("No more load to balance in system.");
 257         last_status = false;
 258     }
 259
 260     return last_status;
 261 }
 262
 263 double process::get_sum_of_to_send() const
 264 {
 265     using std::placeholders::_1;
 266     using std::placeholders::_2;
 267
 268     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 269                            std::bind(std::plus<double>(), _1,
 270                                      std::bind(&neighbor::get_to_send, _2)));
 271 }
 272
 273 void process::load_balance()
 274 {
 275     if (lb_iter == 1)           // warn only once
 276         XBT_WARN("process::load_balance() is a no-op!");
 277 }
 278
 279 void process::send(neighbor& nb, double amount)
 280 {
 281     expected_load -= amount;
 282     nb.set_to_send(nb.get_to_send() + amount);
 283     nb.set_load(nb.get_load() + amount);
 284 }
 285
 286 void process::ctrl_send(neighbor& nb)
 287 {
 288     double info_to_send = expected_load;
 289     double debt_to_send;
 290     if (opt::bookkeeping) {     // bookkeeping
 291         debt_to_send = nb.get_to_send();
 292         if (debt_to_send > 0.0) {
 293             nb.set_to_send(0.0);
 294             nb.set_debt(nb.get_debt() + debt_to_send);
 295         }
 296     } else {                    // !bookkeeping
 297         debt_to_send = 0.0;
 298     }
 299     if (info_to_send != prev_load_broadcast || debt_to_send > 0.0) {
 300         message* msg = new message(message::CTRL, info_to_send, debt_to_send);
 301         add_ctrl_send_mesg(msg->get_size());
 302         comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 303     }
 304 }
 305
 306 double process::compute_load_to_send(double desired)
 307 {
 308     return desired >= opt::min_transfer_amount ? desired : 0.0;
 309 }
 310
 311 void process::data_send(neighbor& nb)
 312 {
 313     double load_to_send;
 314     if (opt::bookkeeping) {
 315         load_to_send = compute_load_to_send(std::min(real_load, nb.get_debt()));
 316         if (load_to_send > 0.0)
 317             nb.set_debt(nb.get_debt() - load_to_send);
 318     } else {
 319         load_to_send = compute_load_to_send(nb.get_to_send());
 320         if (load_to_send > 0.0)
 321             nb.set_to_send(nb.get_to_send() - load_to_send);
 322     }
 323     real_load -= load_to_send;
 324     while (load_to_send > 0.0) {
 325         double amount;
 326         if (opt::max_transfer_amount)
 327             amount = std::min(load_to_send, opt::max_transfer_amount);
 328         else
 329             amount = load_to_send;
 330         message* msg = new message(message::DATA, amount);
 331         add_data_send_mesg(msg->get_size());
 332         comm.data_send(nb.get_data_mbox(), msg);
 333         load_to_send -= amount;
 334     }
 335 }
 336
 337 void process::ctrl_close(neighbor& nb)
 338 {
 339     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 340 }
 341
 342 void process::data_close(neighbor& nb)
 343 {
 344     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 345 }
 346
 347 void process::ctrl_receive(double timeout)
 348 {
 349     message* msg;
 350     m_host_t from;
 351
 352     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 353     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 354         if (msg->get_type() != message::CTRL_CLOSE)
 355             add_ctrl_recv_mesg(msg->get_size());
 356         handle_message(msg, from);
 357         timeout = 0.0;
 358     }
 359 }
 360
 361 void process::data_receive(double timeout)
 362 {
 363     message* msg;
 364     m_host_t from;
 365
 366     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 367     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 368         if (msg->get_type() != message::DATA_CLOSE)
 369             add_data_recv_mesg(msg->get_size());
 370         handle_message(msg, from);
 371         timeout = 0.0;
 372     }
 373 }
 374
 375 void process::handle_message(message* msg, m_host_t from)
 376 {
 377     switch (msg->get_type()) {
 378     case message::CTRL: {
 379         neighbor* n = rev_neigh[from];
 380         n->set_load(msg->get_amount() + n->get_to_send());
 381         expected_load += msg->get_credit(); // may be 0.0 if !opt::bookkeeping
 382         break;
 383     }
 384     case message::DATA: {
 385         double ld = msg->get_amount();
 386         received_load += ld;
 387         break;
 388     }
 389     case message::CTRL_CLOSE:
 390         ctrl_close_pending--;
 391         close_received = true;
 392         break;
 393     case message::DATA_CLOSE:
 394         data_close_pending--;
 395         close_received = true;
 396         break;
 397     }
 398     delete msg;
 399 }
 400
 401 #define print_loads_generic(vec, verbose, logp, cat)                    \
 402     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 403         using std::placeholders::_1;                                    \
 404         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 405                   "Neighbor loads:", real_load, expected_load);         \
 406         std::for_each(vec.begin(), vec.end(),                           \
 407                       std::bind(&neighbor::print, _1, verbose, logp, cat)); \
 408     } else ((void)0)
 409
 410 void process::print_loads(bool verbose,
 411                           e_xbt_log_priority_t logp,
 412                           xbt_log_category_t cat) const
 413 {
 414     print_loads_generic(neigh, verbose, logp, cat);
 415 }
 416
 417 void process::print_loads_p(bool verbose,
 418                             e_xbt_log_priority_t logp,
 419                             xbt_log_category_t cat) const
 420 {
 421     print_loads_generic(pneigh, verbose, logp, cat);
 422 }
 423
 424 #undef print_loads_generic
 425
 426 // Local variables:
 427 // mode: c++
 428 // End: