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
  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::bind(&process::load_balance_loop, this));
  62
  63     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  64     if (!LOG_ISENABLED(logp))
  65         return;
  66     std::ostringstream oss;
  67     oss << neigh.size() << " neighbor";
  68     if (!neigh.empty()) {
  69         oss << ESSE(neigh.size()) << ": ";
  70         std::transform(neigh.begin(), neigh.end() - 1,
  71                        std::ostream_iterator<const char*>(oss, ", "),
  72                        std::mem_fn(&neighbor::get_name));
  73         oss << neigh.back().get_name();
  74     }
  75     XBT_LOG(logp, "Got %s.", oss.str().c_str());
  76     print_loads(false, logp);
  77 }
  78
  79 process::~process()
  80 {
  81     delete lb_thread;
  82     total_load_exit += real_load;
  83     if (opt::log_rate < 0)
  84         return;
  85     XBT_INFO("Final load after %d:%d iterations: %g",
  86              lb_iter, comp_iter, real_load);
  87     XBT_VERB("Expected load was: %g", expected_load);
  88     XBT_VERB("Total computation for this process: %g", get_comp_amount());
  89 }
  90
  91 int process::run()
  92 {
  93     if (opt::log_rate >= 0) {
  94         XBT_INFO("Initial load: %g", real_load);
  95         XBT_VERB("Initial expected load: %g", expected_load);
  96     }
  97     XBT_VERB("Starting...");
  98     mutex.acquire();
  99     lb_thread->start();
 100     while (lb_iter <= opt::comp_iter_delay)
 101         cond.wait(mutex);
 102     mutex.release();
 103     double sleep_duration = opt::comp_time_delay - MSG_get_clock();
 104     if (sleep_duration > 0.0)
 105         MSG_process_sleep(sleep_duration);
 106     compute_loop();
 107     lb_thread->wait();
 108     XBT_VERB("Done.");
 109     return 0;
 110 }
 111
 112 void process::load_balance_loop()
 113 {
 114     using std::placeholders::_1;
 115
 116     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 117     while (still_running()) {
 118         if (lb_iter == opt::comp_iter_delay) {
 119             mutex.acquire();
 120             ++lb_iter;
 121             cond.signal();
 122             mutex.release();
 123         } else {
 124             ++lb_iter;
 125         }
 126
 127         mutex.acquire();
 128         if (!opt::bookkeeping)
 129             expected_load = real_load - get_sum_of_to_send();
 130         // nothing to do with opt::bookkeeping
 131
 132         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 133             XBT_INFO("(%u:%u) current load: %g", lb_iter, comp_iter, real_load);
 134             XBT_VERB("... expected load: %g", expected_load);
 135         }
 136
 137         if (expected_load > 0.0)
 138             load_balance();
 139
 140         print_loads(true, xbt_log_priority_debug);
 141
 142         // send
 143         comm.ctrl_flush(false);
 144         std::for_each(neigh.begin(), neigh.end(),
 145                       std::bind(&process::ctrl_send, this, _1));
 146         prev_load_broadcast = expected_load;
 147         mutex.release();
 148
 149         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 150         ctrl_receive(0.0);
 151     }
 152
 153     XBT_VERB("Going to finalize for %s...", __func__);
 154     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 155               neigh.size(), ESSE(neigh.size()));
 156     std::for_each(neigh.begin(), neigh.end(),
 157                   std::bind(&process::ctrl_close, this, _1));
 158     while (ctrl_close_pending) {
 159         comm.ctrl_flush(false);
 160         XBT_DEBUG("waiting for %d CTRL_CLOSE", ctrl_close_pending);
 161         ctrl_receive(-1.0);
 162     }
 163     comm.ctrl_flush(true);
 164 }
 165
 166 void process::compute_loop()
 167 {
 168     using std::placeholders::_1;
 169
 170     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 171     while (still_running()) {
 172         // receive
 173         mutex.acquire();
 174         if (real_load > 0.0)
 175             data_receive(0.0);
 176         else
 177             data_receive(opt::min_comp_iter_duration);
 178         mutex.release();
 179
 180         // send
 181         comm.data_flush(false);
 182         mutex.acquire();
 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     total_load_running -= real_load;
 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     comm.data_flush(true);
 216 }
 217
 218 bool process::still_running()
 219 {
 220     static bool last_status = true;
 221
 222     if (!last_status) {
 223         /* nop */
 224
 225     } else if (opt::exit_request) {
 226         XBT_VERB("Global exit requested");
 227         last_status = false;
 228
 229     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 230         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 231         last_status = false;
 232
 233     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 234         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 235         last_status = false;
 236
 237     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 238         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 239         last_status = false;
 240
 241     } else if (opt::exit_on_close && close_received) {
 242         XBT_VERB("Close received");
 243         last_status = false;
 244
 245     } else if (real_load == 0.0 && !data_close_pending) {
 246         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 247         last_status = false;
 248
 249     } else if (100.0 * total_load_running / total_load_init <=
 250                opt::load_ratio_threshold) {
 251         // fixme: this check should be implemented with a distributed
 252         // algorithm, and not a shared global variable!
 253         XBT_VERB("No more load to balance in system.");
 254         last_status = false;
 255     }
 256
 257     return last_status;
 258 }
 259
 260 double process::get_sum_of_to_send() const
 261 {
 262     using std::placeholders::_1;
 263     using std::placeholders::_2;
 264
 265     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 266                            std::bind(std::plus<double>(), _1,
 267                                      std::bind(&neighbor::get_to_send, _2)));
 268 }
 269
 270 void process::load_balance()
 271 {
 272     if (lb_iter == 1)           // warn only once
 273         XBT_WARN("process::load_balance() is a no-op!");
 274 }
 275
 276 void process::send(neighbor& nb, double amount)
 277 {
 278     expected_load -= amount;
 279     nb.set_to_send(nb.get_to_send() + amount);
 280     nb.set_load(nb.get_load() + amount);
 281 }
 282
 283 void process::ctrl_send(neighbor& nb)
 284 {
 285     double info_to_send = expected_load;
 286     double debt_to_send;
 287     if (opt::bookkeeping) {     // bookkeeping
 288         debt_to_send = nb.get_to_send();
 289         if (debt_to_send > 0.0) {
 290             nb.set_to_send(0.0);
 291             nb.set_debt(nb.get_debt() + debt_to_send);
 292         }
 293     } else {                    // !bookkeeping
 294         debt_to_send = 0.0;
 295     }
 296     if (info_to_send != prev_load_broadcast || debt_to_send > 0.0) {
 297         message* msg = new message(message::CTRL, info_to_send, debt_to_send);
 298         add_ctrl_send_mesg(msg->get_size());
 299         comm.ctrl_send(nb.get_ctrl_mbox(), msg);
 300     }
 301 }
 302
 303 void process::data_send(neighbor& nb)
 304 {
 305     double load_to_send;
 306     if (opt::bookkeeping) {
 307         load_to_send = std::min(real_load, nb.get_debt());
 308         if (load_to_send >= opt::min_transfer_amount) {
 309             nb.set_debt(nb.get_debt() - load_to_send);
 310             real_load -= load_to_send;
 311         } else {
 312             load_to_send = 0.0;
 313         }
 314     } else {
 315         load_to_send = nb.get_to_send();
 316         if (load_to_send >= opt::min_transfer_amount) {
 317             nb.set_to_send(0.0);
 318             real_load -= load_to_send;
 319         } else {
 320             load_to_send = 0.0;
 321         }
 322     }
 323     while (load_to_send > 0.0) {
 324         double amount;
 325         if (opt::max_transfer_amount)
 326             amount = std::min(load_to_send, opt::max_transfer_amount);
 327         else
 328             amount = load_to_send;
 329         message* msg = new message(message::DATA, amount);
 330         add_data_send_mesg(msg->get_size());
 331         comm.data_send(nb.get_data_mbox(), msg);
 332         load_to_send -= amount;
 333     }
 334 }
 335
 336 void process::ctrl_close(neighbor& nb)
 337 {
 338     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 339 }
 340
 341 void process::data_close(neighbor& nb)
 342 {
 343     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 344 }
 345
 346 void process::ctrl_receive(double timeout)
 347 {
 348     message* msg;
 349     m_host_t from;
 350
 351     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 352     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 353         if (msg->get_type() != message::CTRL_CLOSE)
 354             add_ctrl_recv_mesg(msg->get_size());
 355         handle_message(msg, from);
 356         timeout = 0.0;
 357     }
 358 }
 359
 360 void process::data_receive(double timeout)
 361 {
 362     message* msg;
 363     m_host_t from;
 364
 365     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 366     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 367         if (msg->get_type() != message::DATA_CLOSE)
 368             add_data_recv_mesg(msg->get_size());
 369         handle_message(msg, from);
 370         timeout = 0.0;
 371     }
 372 }
 373
 374 void process::handle_message(message* msg, m_host_t from)
 375 {
 376     switch (msg->get_type()) {
 377     case message::CTRL: {
 378         neighbor* n = rev_neigh[from];
 379         n->set_load(msg->get_amount() + n->get_to_send());
 380         expected_load += msg->get_credit(); // may be 0.0 if !opt::bookkeeping
 381         break;
 382     }
 383     case message::DATA: {
 384         double ld = msg->get_amount();
 385         real_load += ld;
 386         if (finalizing)
 387             total_load_running -= 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: