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     if (opt::bookkeeping) {
  88         XBT_INFO("Final load after %d:%d iterations: %g ; expected: %g",
  89                  lb_iter, comp_iter, real_load, expected_load);
  90     } else {
  91         XBT_INFO("Final load after %d:%d iterations: %g",
  92                  lb_iter, comp_iter, real_load);
  93     }
  94     XBT_VERB("Total computation for this process: %g", comp);
  95 }
  96
  97 int process::run()
  98 {
  99     if (opt::log_rate >= 0)
 100         XBT_INFO("Initial load: %g", real_load);
 101     XBT_VERB("Starting...");
 102     lb_thread->start();
 103     compute_loop();
 104     lb_thread->wait();
 105     XBT_VERB("Done.");
 106     return 0;
 107 }
 108
 109 void process::load_balance_loop()
 110 {
 111     using std::tr1::bind;
 112     using std::tr1::placeholders::_1;
 113
 114     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 115     while (still_running()) {
 116         ++lb_iter;
 117
 118         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 119             if (opt::bookkeeping)
 120                 XBT_INFO("(%u:%u) current load: %g ; expected: %g",
 121                          lb_iter, comp_iter, real_load, expected_load);
 122             else
 123                 XBT_INFO("(%u:%u) current load: %g",
 124                          lb_iter, comp_iter, real_load);
 125         }
 126
 127         if (get_load() > 0.0)
 128             load_balance();
 129
 130         print_loads(true, xbt_log_priority_debug);
 131
 132         // send
 133         std::for_each(neigh.begin(), neigh.end(),
 134                       bind(&process::ctrl_send, this, _1));
 135         prev_load_broadcast = get_load();
 136
 137         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 138         ctrl_receive(0.0);
 139
 140         comm.ctrl_flush(false);
 141     }
 142
 143     XBT_VERB("Going to finalize for %s...", __func__);
 144     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 145               neigh.size(), ESSE(neigh.size()));
 146     std::for_each(neigh.begin(), neigh.end(),
 147                   bind(&process::ctrl_close, this, _1));
 148     while (ctrl_close_pending) {
 149         comm.ctrl_flush(false);
 150         XBT_DEBUG("waiting for %d CTRL CLOSE", ctrl_close_pending);
 151         ctrl_receive(-1.0);
 152     }
 153     comm.ctrl_flush(true);
 154 }
 155
 156 void process::compute_loop()
 157 {
 158     using std::tr1::bind;
 159     using std::tr1::placeholders::_1;
 160
 161     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 162     while (still_running()) {
 163         // receive
 164         if (real_load > 0.0)
 165             data_receive(0.0);
 166         else
 167             data_receive(opt::min_comp_iter_duration);
 168
 169         comm.data_flush(false);
 170
 171         if (real_load == 0.0)
 172             continue;
 173
 174         // send
 175         std::for_each(neigh.begin(), neigh.end(),
 176                       bind(&process::data_send, this, _1));
 177
 178         // compute
 179         ++comp_iter;
 180         double flops = opt::comp_cost(real_load);
 181         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 182         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 183         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 184         MSG_task_execute(task);
 185         comp += flops;
 186         MSG_task_destroy(task);
 187
 188         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 189     }
 190
 191     XBT_VERB("Going to finalize for %s...", __func__);
 192     // last send, for not losing load scheduled to be sent
 193     std::for_each(neigh.begin(), neigh.end(),
 194                   bind(&process::data_send, this, _1));
 195     finalizing = true;
 196     total_load_running -= real_load;
 197     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 198               neigh.size(), ESSE(neigh.size()));
 199     std::for_each(neigh.begin(), neigh.end(),
 200                   bind(&process::data_close, this, _1));
 201     while (data_close_pending) {
 202         comm.data_flush(false);
 203         XBT_DEBUG("waiting for %d DATA CLOSE", data_close_pending);
 204         data_receive(-1.0);
 205     }
 206     comm.data_flush(true);
 207 }
 208
 209 bool process::still_running()
 210 {
 211     static bool last_status = true;
 212
 213     if (!last_status) {
 214         /* nop */
 215
 216     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 217         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 218         last_status = false;
 219
 220     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 221         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 222         last_status = false;
 223
 224     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 225         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 226         last_status = false;
 227
 228     } else if (opt::exit_on_close && close_received) {
 229         XBT_VERB("Close received");
 230         last_status = false;
 231
 232     } else if (real_load == 0.0 && !data_close_pending) {
 233         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 234         last_status = false;
 235
 236     } else if (100.0 * total_load_running / total_load_init <=
 237                opt::load_ratio_threshold) {
 238         // fixme: this check should be implemented with a distributed
 239         // algorithm, and not a shared global variable!
 240         XBT_VERB("No more load to balance in system.");
 241         last_status = false;
 242     }
 243
 244     return last_status;
 245 }
 246
 247 void process::load_balance()
 248 {
 249     if (lb_iter == 1)           // warn only once
 250         XBT_WARN("process::load_balance() is a no-op!");
 251 }
 252
 253 void process::send(neighbor& nb, double amount)
 254 {
 255     set_load(get_load() - amount);
 256     nb.set_to_send(nb.get_to_send() + amount);
 257     nb.set_load(nb.get_load() + amount);
 258 }
 259
 260 #define print_loads_generic(vec, verbose, logp, cat)                    \
 261     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 262         using std::tr1::bind;                                           \
 263         using std::tr1::placeholders::_1;                               \
 264         XBT_XCLOG(cat, logp, "Neighbor loads:");                        \
 265         std::for_each(vec.begin(), vec.end(),                           \
 266                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 267     } else ((void)0)
 268
 269 void process::print_loads(bool verbose,
 270                           e_xbt_log_priority_t logp,
 271                           xbt_log_category_t cat) const
 272 {
 273     print_loads_generic(neigh, verbose, logp, cat);
 274 }
 275
 276 void process::print_loads_p(bool verbose,
 277                             e_xbt_log_priority_t logp,
 278                             xbt_log_category_t cat) const
 279 {
 280     print_loads_generic(pneigh, verbose, logp, cat);
 281 }
 282
 283 #undef print_loads_generic
 284
 285 void process::ctrl_send(neighbor& nb)
 286 {
 287     double info_to_send = get_load();
 288     if (info_to_send != prev_load_broadcast)
 289         comm.ctrl_send(nb.get_ctrl_mbox(),
 290                        new message(message::INFO, info_to_send));
 291     if (opt::bookkeeping) {
 292         double debt_to_send = nb.get_to_send();
 293         if (debt_to_send > 0.0) {
 294             nb.set_to_send(0.0);
 295             nb.set_debt(nb.get_debt() + debt_to_send);
 296             comm.ctrl_send(nb.get_ctrl_mbox(),
 297                            new message(message::CREDIT, debt_to_send));
 298         }
 299     }
 300 }
 301
 302 void process::data_send(neighbor& nb)
 303 {
 304     double load_to_send;
 305     if (opt::bookkeeping) {
 306         if (real_load <= nb.get_debt()) {
 307             load_to_send = real_load;
 308             nb.set_debt(nb.get_debt() - load_to_send);
 309             real_load = 0.0;
 310         } else {
 311             load_to_send = nb.get_debt();
 312             nb.set_debt(0.0);
 313             real_load -= load_to_send;
 314         }
 315     } else {
 316         load_to_send = nb.get_to_send();
 317         nb.set_to_send(0.0);
 318         // do not update real_load here
 319     }
 320     if (load_to_send > 0.0)
 321         comm.data_send(nb.get_data_mbox(),
 322                        new message(message::LOAD, load_to_send));
 323 }
 324
 325 void process::ctrl_close(neighbor& nb)
 326 {
 327     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 328 }
 329
 330 void process::data_close(neighbor& nb)
 331 {
 332     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 333 }
 334
 335 void process::ctrl_receive(double timeout)
 336 {
 337     message* msg;
 338     m_host_t from;
 339
 340     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 341     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 342         handle_message(msg, from);
 343         timeout = 0.0;
 344     }
 345 }
 346
 347 void process::data_receive(double timeout)
 348 {
 349     message* msg;
 350     m_host_t from;
 351
 352     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 353     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 354         handle_message(msg, from);
 355         timeout = 0.0;
 356     }
 357 }
 358
 359 void process::handle_message(message* msg, m_host_t from)
 360 {
 361     switch (msg->get_type()) {
 362     case message::INFO: {
 363         neighbor* n = rev_neigh[from];
 364         n->set_load(msg->get_amount());
 365         break;
 366     }
 367     case message::CREDIT:
 368         expected_load += msg->get_amount();
 369         break;
 370     case message::LOAD: {
 371         double ld = msg->get_amount();
 372         real_load += ld;
 373         if (finalizing)
 374             total_load_running -= ld;
 375         break;
 376     }
 377     case message::CTRL_CLOSE:
 378         ctrl_close_pending--;
 379         close_received = true;
 380         break;
 381     case message::DATA_CLOSE:
 382         data_close_pending--;
 383         close_received = true;
 384         break;
 385     }
 386     delete msg;
 387 }
 388
 389 // Local variables:
 390 // mode: c++
 391 // End: