process.cpp

   1 #include <algorithm>
   2 #include <tr1/functional>
   3 #include <iterator>
   4 #include <stdexcept>
   5 #include <sstream>
   6 #include <xbt/log.h>
   7 #include <xbt/time.h>
   8
   9 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
  10
  11 #include "misc.h"
  12 #include "options.h"
  13 #include "tracing.h"
  14
  15 #include "process.h"
  16
  17 double process::total_load_init = 0.0;
  18 double process::total_load_running = 0.0;
  19 double process::total_load_exit = 0.0;
  20
  21 namespace {
  22
  23     void sleep_until_date(double& date, double duration = 0.0)
  24     {
  25         double sleep_duration = date - MSG_get_clock();
  26         if (sleep_duration > 0.0)
  27             MSG_process_sleep(sleep_duration);
  28         date = MSG_get_clock() + duration;
  29     }
  30
  31 }
  32
  33 process::process(int argc, char* argv[])
  34 {
  35     if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))
  36         throw std::invalid_argument("bad or missing initial load parameter");
  37
  38     neigh.assign(argv + 2, argv + argc);
  39
  40     pneigh.reserve(neigh.size());
  41     for (unsigned i = 0 ; i < neigh.size() ; i++) {
  42         neighbor* ptr = &neigh[i];
  43         m_host_t host = MSG_get_host_by_name(ptr->get_name());
  44         pneigh.push_back(ptr);
  45         rev_neigh.insert(std::make_pair(host, ptr));
  46     }
  47
  48     comp = 0.0;
  49
  50     prev_load_broadcast = -1;   // force sending of load on first send_all()
  51     expected_load = real_load;
  52     total_load_running += real_load;
  53     total_load_init += real_load;
  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::tr1::bind(&process::load_balance_loop,
  63                                               this));
  64
  65     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  66     if (!LOG_ISENABLED(logp))
  67         return;
  68     std::ostringstream oss;
  69     oss << neigh.size() << " neighbor";
  70     if (!neigh.empty()) {
  71         oss << ESSE(neigh.size()) << ": ";
  72         std::transform(neigh.begin(), neigh.end() - 1,
  73                        std::ostream_iterator<const char*>(oss, ", "),
  74                        std::tr1::mem_fn(&neighbor::get_name));
  75         oss << neigh.back().get_name();
  76     }
  77     XBT_LOG(logp, "Got %s.", oss.str().c_str());
  78     print_loads(false, logp);
  79 }
  80
  81 process::~process()
  82 {
  83     delete lb_thread;
  84     total_load_exit += real_load;
  85     if (opt::log_rate < 0)
  86         return;
  87     XBT_INFO("Final load after %d:%d iterations: %g",
  88              lb_iter, comp_iter, real_load);
  89     XBT_VERB("Expected load was: %g", expected_load);
  90     XBT_VERB("Total computation for this process: %g", comp);
  91 }
  92
  93 int process::run()
  94 {
  95     if (opt::log_rate >= 0) {
  96         XBT_INFO("Initial load: %g", real_load);
  97         XBT_VERB("Initial expected load: %g", expected_load);
  98     }
  99     XBT_VERB("Starting...");
 100     mutex.acquire();
 101     lb_thread->start();
 102     while (lb_iter <= opt::comp_iter_delay)
 103         cond.wait(mutex);
 104     mutex.release();
 105     double sleep_duration = opt::comp_time_delay - MSG_get_clock();
 106     if (sleep_duration > 0.0)
 107         MSG_process_sleep(sleep_duration);
 108     compute_loop();
 109     lb_thread->wait();
 110     XBT_VERB("Done.");
 111     return 0;
 112 }
 113
 114 void process::load_balance_loop()
 115 {
 116     using std::tr1::bind;
 117     using std::tr1::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         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 131             XBT_INFO("(%u:%u) current load: %g", lb_iter, comp_iter, real_load);
 132             XBT_VERB("... expected load: %g", expected_load);
 133         }
 134
 135         if (get_load() > 0.0)
 136             load_balance();
 137
 138         print_loads(true, xbt_log_priority_debug);
 139
 140         // send
 141         std::for_each(neigh.begin(), neigh.end(),
 142                       bind(&process::ctrl_send, this, _1));
 143         prev_load_broadcast = get_load();
 144
 145         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 146         ctrl_receive(0.0);
 147
 148         comm.ctrl_flush(false);
 149     }
 150
 151     XBT_VERB("Going to finalize for %s...", __func__);
 152     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 153               neigh.size(), ESSE(neigh.size()));
 154     std::for_each(neigh.begin(), neigh.end(),
 155                   bind(&process::ctrl_close, this, _1));
 156     while (ctrl_close_pending) {
 157         comm.ctrl_flush(false);
 158         XBT_DEBUG("waiting for %d CTRL CLOSE", ctrl_close_pending);
 159         ctrl_receive(-1.0);
 160     }
 161     comm.ctrl_flush(true);
 162 }
 163
 164 void process::compute_loop()
 165 {
 166     using std::tr1::bind;
 167     using std::tr1::placeholders::_1;
 168
 169     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 170     while (still_running()) {
 171         // receive
 172         if (real_load > 0.0)
 173             data_receive(0.0);
 174         else
 175             data_receive(opt::min_comp_iter_duration);
 176
 177         comm.data_flush(false);
 178
 179         if (real_load == 0.0)
 180             continue;
 181
 182         // send
 183         std::for_each(neigh.begin(), neigh.end(),
 184                       bind(&process::data_send, this, _1));
 185
 186         // compute
 187         ++comp_iter;
 188         double flops = opt::comp_cost(real_load);
 189         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 190         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 191         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 192         MSG_task_execute(task);
 193         comp += flops;
 194         MSG_task_destroy(task);
 195
 196         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 197     }
 198
 199     XBT_VERB("Going to finalize for %s...", __func__);
 200     // last send, for not losing load scheduled to be sent
 201     std::for_each(neigh.begin(), neigh.end(),
 202                   bind(&process::data_send, this, _1));
 203     finalizing = true;
 204     total_load_running -= real_load;
 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                   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     comm.data_flush(true);
 215 }
 216
 217 bool process::still_running()
 218 {
 219     static bool last_status = true;
 220
 221     if (!last_status) {
 222         /* nop */
 223
 224     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 225         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 226         last_status = false;
 227
 228     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 229         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 230         last_status = false;
 231
 232     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 233         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 234         last_status = false;
 235
 236     } else if (opt::exit_on_close && close_received) {
 237         XBT_VERB("Close received");
 238         last_status = false;
 239
 240     } else if (real_load == 0.0 && !data_close_pending) {
 241         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 242         last_status = false;
 243
 244     } else if (100.0 * total_load_running / total_load_init <=
 245                opt::load_ratio_threshold) {
 246         // fixme: this check should be implemented with a distributed
 247         // algorithm, and not a shared global variable!
 248         XBT_VERB("No more load to balance in system.");
 249         last_status = false;
 250     }
 251
 252     return last_status;
 253 }
 254
 255 void process::load_balance()
 256 {
 257     if (lb_iter == 1)           // warn only once
 258         XBT_WARN("process::load_balance() is a no-op!");
 259 }
 260
 261 void process::send(neighbor& nb, double amount)
 262 {
 263     set_load(get_load() - amount);
 264     nb.set_to_send(nb.get_to_send() + amount);
 265     nb.set_load(nb.get_load() + amount);
 266 }
 267
 268 #define print_loads_generic(vec, verbose, logp, cat)                    \
 269     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 270         using std::tr1::bind;                                           \
 271         using std::tr1::placeholders::_1;                               \
 272         XBT_XCLOG(cat, logp, "Neighbor loads:");                        \
 273         std::for_each(vec.begin(), vec.end(),                           \
 274                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 275     } else ((void)0)
 276
 277 void process::print_loads(bool verbose,
 278                           e_xbt_log_priority_t logp,
 279                           xbt_log_category_t cat) const
 280 {
 281     print_loads_generic(neigh, verbose, logp, cat);
 282 }
 283
 284 void process::print_loads_p(bool verbose,
 285                             e_xbt_log_priority_t logp,
 286                             xbt_log_category_t cat) const
 287 {
 288     print_loads_generic(pneigh, verbose, logp, cat);
 289 }
 290
 291 #undef print_loads_generic
 292
 293 void process::ctrl_send(neighbor& nb)
 294 {
 295     double info_to_send = get_load();
 296     if (info_to_send != prev_load_broadcast)
 297         comm.ctrl_send(nb.get_ctrl_mbox(),
 298                        new message(message::INFO, info_to_send));
 299     if (opt::bookkeeping) {
 300         double debt_to_send = nb.get_to_send();
 301         if (debt_to_send > 0.0) {
 302             nb.set_to_send(0.0);
 303             nb.set_debt(nb.get_debt() + debt_to_send);
 304             comm.ctrl_send(nb.get_ctrl_mbox(),
 305                            new message(message::CREDIT, debt_to_send));
 306         }
 307     }
 308 }
 309
 310 void process::data_send(neighbor& nb)
 311 {
 312     double load_to_send;
 313     if (opt::bookkeeping) {
 314         if (real_load <= nb.get_debt()) {
 315             load_to_send = real_load;
 316             nb.set_debt(nb.get_debt() - load_to_send);
 317             real_load = 0.0;
 318         } else {
 319             load_to_send = nb.get_debt();
 320             nb.set_debt(0.0);
 321             real_load -= load_to_send;
 322         }
 323     } else {
 324         load_to_send = nb.get_to_send();
 325         nb.set_to_send(0.0);
 326         // do not update real_load here
 327     }
 328     if (load_to_send > 0.0)
 329         comm.data_send(nb.get_data_mbox(),
 330                        new message(message::LOAD, load_to_send));
 331 }
 332
 333 void process::ctrl_close(neighbor& nb)
 334 {
 335     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 336 }
 337
 338 void process::data_close(neighbor& nb)
 339 {
 340     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 341 }
 342
 343 void process::ctrl_receive(double timeout)
 344 {
 345     message* msg;
 346     m_host_t from;
 347
 348     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 349     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 350         handle_message(msg, from);
 351         timeout = 0.0;
 352     }
 353 }
 354
 355 void process::data_receive(double timeout)
 356 {
 357     message* msg;
 358     m_host_t from;
 359
 360     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 361     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 362         handle_message(msg, from);
 363         timeout = 0.0;
 364     }
 365 }
 366
 367 void process::handle_message(message* msg, m_host_t from)
 368 {
 369     switch (msg->get_type()) {
 370     case message::INFO: {
 371         neighbor* n = rev_neigh[from];
 372         n->set_load(msg->get_amount());
 373         break;
 374     }
 375     case message::CREDIT:
 376         expected_load += msg->get_amount();
 377         break;
 378     case message::LOAD: {
 379         double ld = msg->get_amount();
 380         real_load += ld;
 381         if (finalizing)
 382             total_load_running -= ld;
 383         break;
 384     }
 385     case message::CTRL_CLOSE:
 386         ctrl_close_pending--;
 387         close_received = true;
 388         break;
 389     case message::DATA_CLOSE:
 390         data_close_pending--;
 391         close_received = true;
 392         break;
 393     }
 394     delete msg;
 395 }
 396
 397 // Local variables:
 398 // mode: c++
 399 // End: