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