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::log_rate && lb_iter % opt::log_rate == 0) {
 132             XBT_INFO("(%u:%u) current load: %g", lb_iter, comp_iter, real_load);
 133             XBT_VERB("... expected load: %g", expected_load);
 134         }
 135
 136         if (get_load() > 0.0)
 137             load_balance();
 138
 139         print_loads(true, xbt_log_priority_debug);
 140
 141         // send
 142         std::for_each(neigh.begin(), neigh.end(),
 143                       bind(&process::ctrl_send, this, _1));
 144         prev_load_broadcast = get_load();
 145
 146         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 147         ctrl_receive(0.0);
 148
 149         comm.ctrl_flush(false);
 150     }
 151
 152     XBT_VERB("Going to finalize for %s...", __func__);
 153     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 154               neigh.size(), ESSE(neigh.size()));
 155     std::for_each(neigh.begin(), neigh.end(),
 156                   bind(&process::ctrl_close, this, _1));
 157     while (ctrl_close_pending) {
 158         comm.ctrl_flush(false);
 159         XBT_DEBUG("waiting for %d CTRL CLOSE", ctrl_close_pending);
 160         ctrl_receive(-1.0);
 161     }
 162     comm.ctrl_flush(true);
 163 }
 164
 165 void process::compute_loop()
 166 {
 167     using std::tr1::bind;
 168     using std::tr1::placeholders::_1;
 169
 170     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 171     while (still_running()) {
 172         // receive
 173         if (real_load > 0.0)
 174             data_receive(0.0);
 175         else
 176             data_receive(opt::min_comp_iter_duration);
 177
 178         comm.data_flush(false);
 179
 180         if (real_load == 0.0)
 181             continue;
 182
 183         // send
 184         std::for_each(neigh.begin(), neigh.end(),
 185                       bind(&process::data_send, this, _1));
 186
 187         // compute
 188         ++comp_iter;
 189         double flops = opt::comp_cost(real_load);
 190         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 191         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 192         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 193         MSG_task_execute(task);
 194         comp += flops;
 195         MSG_task_destroy(task);
 196
 197         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 198     }
 199
 200     XBT_VERB("Going to finalize for %s...", __func__);
 201     // last send, for not losing load scheduled to be sent
 202     std::for_each(neigh.begin(), neigh.end(),
 203                   bind(&process::data_send, this, _1));
 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                   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::time_limit && MSG_get_clock() >= opt::time_limit) {
 226         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 227         last_status = false;
 228
 229     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 230         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 231         last_status = false;
 232
 233     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 234         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 235         last_status = false;
 236
 237     } else if (opt::exit_on_close && close_received) {
 238         XBT_VERB("Close received");
 239         last_status = false;
 240
 241     } else if (real_load == 0.0 && !data_close_pending) {
 242         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 243         last_status = false;
 244
 245     } else if (100.0 * total_load_running / total_load_init <=
 246                opt::load_ratio_threshold) {
 247         // fixme: this check should be implemented with a distributed
 248         // algorithm, and not a shared global variable!
 249         XBT_VERB("No more load to balance in system.");
 250         last_status = false;
 251     }
 252
 253     return last_status;
 254 }
 255
 256 double process::get_sum_of_to_send() const
 257 {
 258     using std::tr1::bind;
 259     using std::tr1::placeholders::_1;
 260     using std::tr1::placeholders::_2;
 261
 262     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 263                            bind(std::plus<double>(),
 264                                 _1, bind(&neighbor::get_to_send, _2)));
 265 }
 266
 267 void process::load_balance()
 268 {
 269     if (lb_iter == 1)           // warn only once
 270         XBT_WARN("process::load_balance() is a no-op!");
 271 }
 272
 273 void process::send(neighbor& nb, double amount)
 274 {
 275     set_load(get_load() - amount);
 276     nb.set_to_send(nb.get_to_send() + amount);
 277     nb.set_load(nb.get_load() + amount);
 278 }
 279
 280 void process::ctrl_send(neighbor& nb)
 281 {
 282     double info_to_send = get_load();
 283     if (info_to_send != prev_load_broadcast)
 284         comm.ctrl_send(nb.get_ctrl_mbox(),
 285                        new message(message::INFO, info_to_send));
 286     if (opt::bookkeeping) {
 287         double debt_to_send = nb.get_to_send();
 288         if (debt_to_send > 0.0) {
 289             nb.set_to_send(0.0);
 290             nb.set_debt(nb.get_debt() + debt_to_send);
 291             comm.ctrl_send(nb.get_ctrl_mbox(),
 292                            new message(message::CREDIT, debt_to_send));
 293         }
 294     }
 295 }
 296
 297 void process::data_send(neighbor& nb)
 298 {
 299     double load_to_send;
 300     if (opt::bookkeeping) {
 301         if (real_load <= nb.get_debt()) {
 302             load_to_send = real_load;
 303             nb.set_debt(nb.get_debt() - load_to_send);
 304             real_load = 0.0;
 305         } else {
 306             load_to_send = nb.get_debt();
 307             nb.set_debt(0.0);
 308             real_load -= load_to_send;
 309         }
 310     } else {
 311         load_to_send = nb.get_to_send();
 312         nb.set_to_send(0.0);
 313         // do not update real_load here
 314     }
 315     if (load_to_send > 0.0)
 316         comm.data_send(nb.get_data_mbox(),
 317                        new message(message::LOAD, load_to_send));
 318 }
 319
 320 void process::ctrl_close(neighbor& nb)
 321 {
 322     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 323 }
 324
 325 void process::data_close(neighbor& nb)
 326 {
 327     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 328 }
 329
 330 void process::ctrl_receive(double timeout)
 331 {
 332     message* msg;
 333     m_host_t from;
 334
 335     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 336     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 337         handle_message(msg, from);
 338         timeout = 0.0;
 339     }
 340 }
 341
 342 void process::data_receive(double timeout)
 343 {
 344     message* msg;
 345     m_host_t from;
 346
 347     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 348     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 349         handle_message(msg, from);
 350         timeout = 0.0;
 351     }
 352 }
 353
 354 void process::handle_message(message* msg, m_host_t from)
 355 {
 356     switch (msg->get_type()) {
 357     case message::INFO: {
 358         neighbor* n = rev_neigh[from];
 359         n->set_load(msg->get_amount());
 360         break;
 361     }
 362     case message::CREDIT:
 363         expected_load += msg->get_amount();
 364         break;
 365     case message::LOAD: {
 366         double ld = msg->get_amount();
 367         real_load += ld;
 368         if (finalizing)
 369             total_load_running -= ld;
 370         break;
 371     }
 372     case message::CTRL_CLOSE:
 373         ctrl_close_pending--;
 374         close_received = true;
 375         break;
 376     case message::DATA_CLOSE:
 377         data_close_pending--;
 378         close_received = true;
 379         break;
 380     }
 381     delete msg;
 382 }
 383
 384 #define print_loads_generic(vec, verbose, logp, cat)                    \
 385     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 386         using std::tr1::bind;                                           \
 387         using std::tr1::placeholders::_1;                               \
 388         XBT_XCLOG(cat, logp, "My load: %g (real); %g (expected).  "     \
 389                   "Neighbor loads:", real_load, expected_load);         \
 390         std::for_each(vec.begin(), vec.end(),                           \
 391                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 392     } else ((void)0)
 393
 394 void process::print_loads(bool verbose,
 395                           e_xbt_log_priority_t logp,
 396                           xbt_log_category_t cat) const
 397 {
 398     print_loads_generic(neigh, verbose, logp, cat);
 399 }
 400
 401 void process::print_loads_p(bool verbose,
 402                             e_xbt_log_priority_t logp,
 403                             xbt_log_category_t cat) const
 404 {
 405     print_loads_generic(pneigh, verbose, logp, cat);
 406 }
 407
 408 #undef print_loads_generic
 409
 410 // Local variables:
 411 // mode: c++
 412 // End: