process.cpp

   1 #include <algorithm>
   2 #include <tr1/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 = 0.0)
  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     comp = 0.0;
  50
  51     prev_load_broadcast = -1;   // force sending of load on first send_all()
  52     expected_load = real_load;
  53     total_load_running += real_load;
  54     total_load_init += real_load;
  55
  56     ctrl_close_pending = data_close_pending = neigh.size();
  57     close_received = false;
  58     finalizing = false;
  59
  60     comp_iter = lb_iter = 0;
  61
  62     lb_thread = new_msg_thread("loba",
  63                                std::tr1::bind(&process::load_balance_loop,
  64                                               this));
  65
  66     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  67     if (!LOG_ISENABLED(logp))
  68         return;
  69     std::ostringstream oss;
  70     oss << neigh.size() << " neighbor";
  71     if (!neigh.empty()) {
  72         oss << ESSE(neigh.size()) << ": ";
  73         std::transform(neigh.begin(), neigh.end() - 1,
  74                        std::ostream_iterator<const char*>(oss, ", "),
  75                        std::tr1::mem_fn(&neighbor::get_name));
  76         oss << neigh.back().get_name();
  77     }
  78     XBT_LOG(logp, "Got %s.", oss.str().c_str());
  79     print_loads(false, logp);
  80 }
  81
  82 process::~process()
  83 {
  84     delete lb_thread;
  85     total_load_exit += real_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", comp);
  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::tr1::bind;
 118     using std::tr1::placeholders::_1;
 119
 120     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 121     while (still_running()) {
 122         if (lb_iter == opt::comp_iter_delay) {
 123             mutex.acquire();
 124             ++lb_iter;
 125             cond.signal();
 126             mutex.release();
 127         } else {
 128             ++lb_iter;
 129         }
 130
 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         std::for_each(neigh.begin(), neigh.end(),
 147                       bind(&process::ctrl_send, this, _1));
 148         prev_load_broadcast = expected_load;
 149
 150         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 151         ctrl_receive(0.0);
 152
 153         comm.ctrl_flush(false);
 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                   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::tr1::bind;
 172     using std::tr1::placeholders::_1;
 173
 174     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 175     while (still_running()) {
 176         // receive
 177         if (real_load > 0.0)
 178             data_receive(0.0);
 179         else
 180             data_receive(opt::min_comp_iter_duration);
 181
 182         comm.data_flush(false);
 183
 184         if (real_load == 0.0)
 185             continue;
 186
 187         // send
 188         std::for_each(neigh.begin(), neigh.end(),
 189                       bind(&process::data_send, this, _1));
 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         comp += 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     // last send, for not losing load scheduled to be sent
 206     std::for_each(neigh.begin(), neigh.end(),
 207                   bind(&process::data_send, this, _1));
 208     finalizing = true;
 209     total_load_running -= real_load;
 210     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 211               neigh.size(), ESSE(neigh.size()));
 212     std::for_each(neigh.begin(), neigh.end(),
 213                   bind(&process::data_close, this, _1));
 214     while (data_close_pending) {
 215         comm.data_flush(false);
 216         XBT_DEBUG("waiting for %d DATA CLOSE", data_close_pending);
 217         data_receive(-1.0);
 218     }
 219     comm.data_flush(true);
 220 }
 221
 222 bool process::still_running()
 223 {
 224     static bool last_status = true;
 225
 226     if (!last_status) {
 227         /* nop */
 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::tr1::bind;
 263     using std::tr1::placeholders::_1;
 264     using std::tr1::placeholders::_2;
 265
 266     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 267                            bind(std::plus<double>(),
 268                                 _1, bind(&neighbor::get_to_send, _2)));
 269 }
 270
 271 void process::load_balance()
 272 {
 273     if (lb_iter == 1)           // warn only once
 274         XBT_WARN("process::load_balance() is a no-op!");
 275 }
 276
 277 void process::send(neighbor& nb, double amount)
 278 {
 279     expected_load -= amount;
 280     nb.set_to_send(nb.get_to_send() + amount);
 281     nb.set_load(nb.get_load() + amount);
 282 }
 283
 284 void process::ctrl_send(neighbor& nb)
 285 {
 286     double info_to_send = expected_load;
 287     if (info_to_send != prev_load_broadcast)
 288         comm.ctrl_send(nb.get_ctrl_mbox(),
 289                        new message(message::INFO, info_to_send));
 290     if (opt::bookkeeping) {
 291         double 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             comm.ctrl_send(nb.get_ctrl_mbox(),
 296                            new message(message::CREDIT, debt_to_send));
 297         }
 298     }
 299 }
 300
 301 void process::data_send(neighbor& nb)
 302 {
 303     double load_to_send;
 304     if (opt::bookkeeping) {
 305         load_to_send = std::min(real_load, nb.get_debt());
 306         if (load_to_send >= opt::min_transfer_amount) {
 307             nb.set_debt(nb.get_debt() - load_to_send);
 308             real_load -= load_to_send;
 309         } else {
 310             load_to_send = 0.0;
 311         }
 312     } else {
 313         load_to_send = nb.get_to_send();
 314         if (load_to_send >= opt::min_transfer_amount) {
 315             nb.set_to_send(0.0);
 316             real_load -= load_to_send;
 317         } else {
 318             load_to_send = 0.0;
 319         }
 320     }
 321     while (load_to_send > 0.0) {
 322         double amount;
 323         if (opt::max_transfer_amount)
 324             amount = std::min(load_to_send, opt::max_transfer_amount);
 325         else
 326             amount = load_to_send;
 327         comm.data_send(nb.get_data_mbox(), new message(message::LOAD, amount));
 328         load_to_send -= amount;
 329     }
 330 }
 331
 332 void process::ctrl_close(neighbor& nb)
 333 {
 334     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 335 }
 336
 337 void process::data_close(neighbor& nb)
 338 {
 339     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 340 }
 341
 342 void process::ctrl_receive(double timeout)
 343 {
 344     message* msg;
 345     m_host_t from;
 346
 347     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 348     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 349         handle_message(msg, from);
 350         timeout = 0.0;
 351     }
 352 }
 353
 354 void process::data_receive(double timeout)
 355 {
 356     message* msg;
 357     m_host_t from;
 358
 359     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 360     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 361         handle_message(msg, from);
 362         timeout = 0.0;
 363     }
 364 }
 365
 366 void process::handle_message(message* msg, m_host_t from)
 367 {
 368     switch (msg->get_type()) {
 369     case message::INFO: {
 370         neighbor* n = rev_neigh[from];
 371         n->set_load(msg->get_amount() + n->get_to_send());
 372         break;
 373     }
 374     case message::CREDIT:
 375         expected_load += msg->get_amount();
 376         break;
 377     case message::LOAD: {
 378         double ld = msg->get_amount();
 379         real_load += ld;
 380         if (finalizing)
 381             total_load_running -= ld;
 382         break;
 383     }
 384     case message::CTRL_CLOSE:
 385         ctrl_close_pending--;
 386         close_received = true;
 387         break;
 388     case message::DATA_CLOSE:
 389         data_close_pending--;
 390         close_received = true;
 391         break;
 392     }
 393     delete msg;
 394 }
 395
 396 #define print_loads_generic(vec, verbose, logp, cat)                    \
 397     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 398         using std::tr1::bind;                                           \
 399         using std::tr1::placeholders::_1;                               \
 400         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 401                   "Neighbor loads:", real_load, expected_load);         \
 402         std::for_each(vec.begin(), vec.end(),                           \
 403                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 404     } else ((void)0)
 405
 406 void process::print_loads(bool verbose,
 407                           e_xbt_log_priority_t logp,
 408                           xbt_log_category_t cat) const
 409 {
 410     print_loads_generic(neigh, verbose, logp, cat);
 411 }
 412
 413 void process::print_loads_p(bool verbose,
 414                             e_xbt_log_priority_t logp,
 415                             xbt_log_category_t cat) const
 416 {
 417     print_loads_generic(pneigh, verbose, logp, cat);
 418 }
 419
 420 #undef print_loads_generic
 421
 422 // Local variables:
 423 // mode: c++
 424 // End: