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