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
  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 process::process(int argc, char* argv[])
  22 {
  23     if (argc < 2 || !(std::istringstream(argv[1]) >> load))
  24         throw std::invalid_argument("bad or missing initial load parameter");
  25
  26     neigh.assign(argv + 2, argv + argc);
  27
  28     pneigh.reserve(neigh.size());
  29     for (unsigned i = 0 ; i < neigh.size() ; i++) {
  30         neighbor* ptr = &neigh[i];
  31         m_host_t host = MSG_get_host_by_name(ptr->get_name());
  32         pneigh.push_back(ptr);
  33         rev_neigh.insert(std::make_pair(host, ptr));
  34     }
  35
  36     prev_load_broadcast = -1;   // force sending of load on first send()
  37     expected_load = load;
  38     total_load_running += load;
  39     total_load_init += load;
  40
  41     ctrl_close_pending = data_close_pending = neigh.size();
  42     close_received = false;
  43     finalizing = false;
  44
  45     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  46     if (!LOG_ISENABLED(logp))
  47         return;
  48     std::ostringstream oss;
  49     oss << neigh.size() << " neighbor";
  50     if (!neigh.empty()) {
  51         oss << ESSE(neigh.size()) << ": ";
  52         std::transform(neigh.begin(), neigh.end() - 1,
  53                        std::ostream_iterator<const char*>(oss, ", "),
  54                        std::tr1::mem_fn(&neighbor::get_name));
  55         oss << neigh.back().get_name();
  56     }
  57     LOG1(logp, "Got %s.", oss.str().c_str());
  58     print_loads(false, logp);
  59 }
  60
  61 process::~process()
  62 {
  63     total_load_exit += load;
  64 }
  65
  66 int process::run()
  67 {
  68     double next_iter_after_date = 0.0;
  69     INFO1("Initial load: %g", load);
  70     VERB0("Starting...");
  71     comp_iter = lb_iter = 0;
  72     while (true) {
  73         if (opt::min_iter_duration) {
  74             double now = MSG_get_clock();
  75             if (now < next_iter_after_date){
  76                 double delay = next_iter_after_date - now;
  77                 DEBUG1("sleeping for %g s", delay);
  78                 MSG_process_sleep(next_iter_after_date - now);
  79             }
  80             next_iter_after_date = MSG_get_clock() + opt::min_iter_duration;
  81         }
  82
  83         if (load > 0.0) {
  84             ++comp_iter;
  85             if (opt::log_rate && comp_iter % opt::log_rate == 0) {
  86                 if (opt::bookkeeping)
  87                     INFO4("(%u:%u) current load: %g ; expected: %g",
  88                           comp_iter, lb_iter, load, expected_load);
  89                 else
  90                     INFO2("(%u) current load: %g",
  91                           comp_iter, load);
  92             }
  93
  94             if (opt::bookkeeping)
  95                 expected_load -= load_balance(expected_load);
  96             else
  97                 load -= load_balance(load);
  98
  99             print_loads(true, xbt_log_priority_debug);
 100
 101             send();
 102             compute();
 103
 104         } else {
 105             // send load information, and load when bookkeeping
 106             send();
 107         }
 108
 109         if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 110             VERB2("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 111             break;
 112         }
 113         if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 114             VERB2("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 115             break;
 116         }
 117         if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 118             VERB2("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 119             break;
 120         }
 121
 122         // block on receiving unless there is something to compute or
 123         // to send
 124         bool wait = (load == 0 &&
 125                      ((opt::bookkeeping ? expected_load : load)
 126                       == prev_load_broadcast));
 127         receive(wait);
 128
 129         // one of our neighbor is finalizing
 130         if (opt::exit_on_close && close_received) {
 131             VERB0("Close received");
 132             break;
 133         }
 134
 135         // have no load and cannot receive anything
 136         if (load == 0.0 && !may_receive()) {
 137             VERB0("I'm a poor lonesome process, and I have no load...");
 138             break;
 139         }
 140
 141         // fixme: this check should be implemented with a distributed
 142         // algorithm, and not a shared global variable!
 143         // fixme: should this chunk be moved before call to receive() ?
 144         if (100.0 * total_load_running / total_load_init <=
 145             opt::load_ratio_threshold) {
 146             VERB0("No more load to balance in system.");
 147             break;
 148         }
 149     }
 150     VERB0("Going to finalize...");
 151     finalize();
 152
 153     /* Open Questions :
 154      * - definition of load on heterogeneous hosts ?
 155      * - how to detect convergence ?
 156      * - how to manage link failures ?
 157      */
 158
 159     VERB0("Done.");
 160     if (opt::bookkeeping) {
 161         INFO4("Final load after %d:%d iterations: %g ; expected: %g",
 162               comp_iter, lb_iter, load, expected_load);
 163     } else {
 164         INFO2("Final load after %d iterations: %g",
 165               comp_iter, load);
 166         if (comp_iter != lb_iter)
 167             WARN2("comp_iter (%d) and lb_iter (%d) differ!",
 168                   comp_iter, lb_iter);
 169     }
 170     return 0;
 171 }
 172
 173 double process::sum_of_to_send() const
 174 {
 175     using std::tr1::bind;
 176     using std::tr1::placeholders::_1;
 177     using std::tr1::placeholders::_2;
 178
 179     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 180                            bind(std::plus<double>(),
 181                                 _1, bind(&neighbor::get_to_send, _2)));
 182 }
 183
 184 double process::load_balance(double /*my_load*/)
 185 {
 186     if (lb_iter == 1)           // warn only once
 187         WARN0("process::load_balance() is a no-op!");
 188     return 0.0;
 189 }
 190
 191 void process::compute()
 192 {
 193     if (load > 0.0) {
 194         double duration = opt::comp_cost(load);
 195         m_task_t task = MSG_task_create("computation", duration, 0.0, NULL);
 196         DEBUG2("compute %g flop%s", duration, ESSE(duration));
 197         MSG_task_execute(task);
 198         MSG_task_destroy(task);
 199     } else {
 200         DEBUG0("nothing to compute !");
 201     }
 202 }
 203
 204 void process::send1_no_bookkeeping(neighbor& nb)
 205 {
 206     if (load != prev_load_broadcast)
 207         comm.send(nb.get_ctrl_mbox(), new message(message::INFO, load));
 208     double load_to_send = nb.get_to_send();
 209     if (load_to_send > 0.0) {
 210         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 211         nb.set_to_send(0.0);
 212     }
 213 }
 214
 215 void process::send1_bookkeeping(neighbor& nb)
 216 {
 217     if (expected_load != prev_load_broadcast)
 218         comm.send(nb.get_ctrl_mbox(),
 219                   new message(message::INFO, expected_load));
 220     double load_to_send;
 221     double new_debt;
 222     double debt_to_send = nb.get_to_send();
 223     if (debt_to_send > 0.0) {
 224         comm.send(nb.get_ctrl_mbox(),
 225                   new message(message::CREDIT, debt_to_send));
 226         nb.set_to_send(0.0);
 227         new_debt = nb.get_debt() + debt_to_send;
 228     } else {
 229         new_debt = nb.get_debt();
 230     }
 231     if (load <= new_debt) {
 232         load_to_send = load;
 233         nb.set_debt(new_debt - load_to_send);
 234         load = 0.0;
 235     } else {
 236         load_to_send = new_debt;
 237         nb.set_debt(0.0);
 238         load -= load_to_send;
 239     }
 240     if (load_to_send > 0.0)
 241         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 242 }
 243
 244 void process::send()
 245 {
 246     using std::tr1::bind;
 247     using std::tr1::placeholders::_1;
 248
 249     if (opt::bookkeeping) {
 250         std::for_each(neigh.begin(), neigh.end(),
 251                       bind(&process::send1_bookkeeping, this, _1));
 252         prev_load_broadcast = expected_load;
 253     } else {
 254         std::for_each(neigh.begin(), neigh.end(),
 255                       bind(&process::send1_no_bookkeeping, this, _1));
 256         prev_load_broadcast = load;
 257     }
 258     comm.flush(false);
 259 }
 260
 261 void process::receive(bool wait)
 262 {
 263     message* msg;
 264     m_host_t from;
 265
 266     while (may_receive() && comm.recv(msg, from, wait)) {
 267         switch (msg->get_type()) {
 268         case message::INFO: {
 269             neighbor* n = rev_neigh[from];
 270             n->set_load(msg->get_amount());
 271             break;
 272         }
 273         case message::CREDIT:
 274             expected_load += msg->get_amount();
 275             break;
 276         case message::LOAD: {
 277             double ld = msg->get_amount();
 278             load += ld;
 279             if (finalizing)
 280                 total_load_running -= ld;
 281             break;
 282         }
 283         case message::CTRL_CLOSE:
 284             ctrl_close_pending--;
 285             close_received = true;
 286             break;
 287         case message::DATA_CLOSE:
 288             data_close_pending--;
 289             close_received = true;
 290             break;
 291         }
 292         delete msg;
 293         wait = false;           // only wait on first recv
 294     }
 295     comm.flush(false);
 296 }
 297
 298 void process::finalize1(neighbor& nb)
 299 {
 300     comm.send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 301     comm.send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 302 }
 303
 304 void process::finalize()
 305 {
 306     using std::tr1::bind;
 307     using std::tr1::placeholders::_1;
 308
 309     finalizing = true;
 310     total_load_running -= load;
 311
 312     DEBUG2("send CLOSE to %lu neighbor%s",
 313            (unsigned long )neigh.size(), ESSE(neigh.size()));
 314     std::for_each(neigh.begin(), neigh.end(),
 315                   bind(&process::finalize1, this, _1));
 316
 317     DEBUG2("wait for CLOSE from %lu neighbor%s",
 318            (unsigned long )neigh.size(), ESSE(neigh.size()));
 319     while (may_receive()) {
 320         comm.flush(false);
 321         receive(true);
 322     }
 323
 324     comm.flush(true);
 325 }
 326
 327 #define print_loads_generic(vec, verbose, logp, cat)                    \
 328     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 329         using std::tr1::bind;                                           \
 330         using std::tr1::placeholders::_1;                               \
 331         XCLOG0(cat, logp, "Neighbor loads:");                           \
 332         std::for_each(vec.begin(), vec.end(),                           \
 333                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 334     } else ((void)0)
 335
 336 void process::print_loads(bool verbose,
 337                           e_xbt_log_priority_t logp,
 338                           xbt_log_category_t cat) const
 339 {
 340     print_loads_generic(neigh, verbose, logp, cat);
 341 }
 342
 343 void process::print_loads_p(bool verbose,
 344                             e_xbt_log_priority_t logp,
 345                             xbt_log_category_t cat) const
 346 {
 347     print_loads_generic(pneigh, verbose, logp, cat);
 348 }
 349
 350 #undef print_loads_generic
 351
 352 // Local variables:
 353 // mode: c++
 354 // End: