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         double ld = lb_load();
  84         if (ld > 0.0) {
  85             ++lb_iter;
  86
  87             if (opt::log_rate && lb_iter % opt::log_rate == 0) {
  88                 if (opt::bookkeeping)
  89                     INFO4("(%u:%u) current load: %g ; expected: %g",
  90                           lb_iter, comp_iter, load, expected_load);
  91                 else
  92                     INFO2("(%u) current load: %g",
  93                           lb_iter, load);
  94             }
  95
  96             ld -= load_balance(ld);
  97
  98             print_loads(true, xbt_log_priority_debug);
  99         }
 100         lb_load() = ld;
 101
 102         // send load information, and load (data) if any
 103         send();
 104         if (load > 0.0) {
 105             ++comp_iter;
 106             compute();
 107         }
 108
 109         if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 110             VERB2("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 111             break;
 112         }
 113         if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 114             VERB2("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_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 && lb_load() == prev_load_broadcast);
 125         receive(wait);
 126
 127         // one of our neighbor is finalizing
 128         if (opt::exit_on_close && close_received) {
 129             VERB0("Close received");
 130             break;
 131         }
 132
 133         // have no load and cannot receive anything
 134         if (load == 0.0 && !may_receive()) {
 135             VERB0("I'm a poor lonesome process, and I have no load...");
 136             break;
 137         }
 138
 139         // fixme: this check should be implemented with a distributed
 140         // algorithm, and not a shared global variable!
 141         // fixme: should this chunk be moved before call to receive() ?
 142         if (100.0 * total_load_running / total_load_init <=
 143             opt::load_ratio_threshold) {
 144             VERB0("No more load to balance in system.");
 145             break;
 146         }
 147     }
 148     VERB0("Going to finalize...");
 149     finalize();
 150
 151     /* Open Questions :
 152      * - definition of load on heterogeneous hosts ?
 153      * - how to detect convergence ?
 154      * - how to manage link failures ?
 155      */
 156
 157     VERB0("Done.");
 158     if (opt::bookkeeping) {
 159         INFO4("Final load after %d:%d iterations: %g ; expected: %g",
 160               lb_iter, comp_iter, load, expected_load);
 161     } else {
 162         INFO2("Final load after %d iterations: %g",
 163               lb_iter, load);
 164         if (lb_iter != comp_iter)
 165             WARN2("lb_iter (%d) and comp_iter (%d) differ!",
 166                   lb_iter, comp_iter);
 167     }
 168     return 0;
 169 }
 170
 171 double process::sum_of_to_send() const
 172 {
 173     using std::tr1::bind;
 174     using std::tr1::placeholders::_1;
 175     using std::tr1::placeholders::_2;
 176
 177     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 178                            bind(std::plus<double>(),
 179                                 _1, bind(&neighbor::get_to_send, _2)));
 180 }
 181
 182 double process::load_balance(double /*my_load*/)
 183 {
 184     if (lb_iter == 1)           // warn only once
 185         WARN0("process::load_balance() is a no-op!");
 186     return 0.0;
 187 }
 188
 189 void process::compute()
 190 {
 191     if (load > 0.0) {
 192         double duration = opt::comp_cost(load);
 193         m_task_t task = MSG_task_create("computation", duration, 0.0, NULL);
 194         DEBUG2("compute %g flop%s", duration, ESSE(duration));
 195         MSG_task_execute(task);
 196         MSG_task_destroy(task);
 197     } else {
 198         DEBUG0("nothing to compute !");
 199     }
 200 }
 201
 202 void process::send1_no_bookkeeping(neighbor& nb)
 203 {
 204     if (load != prev_load_broadcast)
 205         comm.send(nb.get_ctrl_mbox(), new message(message::INFO, load));
 206     double load_to_send = nb.get_to_send();
 207     if (load_to_send > 0.0) {
 208         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 209         nb.set_to_send(0.0);
 210     }
 211 }
 212
 213 void process::send1_bookkeeping(neighbor& nb)
 214 {
 215     if (expected_load != prev_load_broadcast)
 216         comm.send(nb.get_ctrl_mbox(),
 217                   new message(message::INFO, expected_load));
 218     double load_to_send;
 219     double new_debt;
 220     double debt_to_send = nb.get_to_send();
 221     if (debt_to_send > 0.0) {
 222         comm.send(nb.get_ctrl_mbox(),
 223                   new message(message::CREDIT, debt_to_send));
 224         nb.set_to_send(0.0);
 225         new_debt = nb.get_debt() + debt_to_send;
 226     } else {
 227         new_debt = nb.get_debt();
 228     }
 229     if (load <= new_debt) {
 230         load_to_send = load;
 231         nb.set_debt(new_debt - load_to_send);
 232         load = 0.0;
 233     } else {
 234         load_to_send = new_debt;
 235         nb.set_debt(0.0);
 236         load -= load_to_send;
 237     }
 238     if (load_to_send > 0.0)
 239         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 240 }
 241
 242 void process::send()
 243 {
 244     using std::tr1::bind;
 245     using std::tr1::placeholders::_1;
 246
 247     if (opt::bookkeeping) {
 248         std::for_each(neigh.begin(), neigh.end(),
 249                       bind(&process::send1_bookkeeping, this, _1));
 250         prev_load_broadcast = expected_load;
 251     } else {
 252         std::for_each(neigh.begin(), neigh.end(),
 253                       bind(&process::send1_no_bookkeeping, this, _1));
 254         prev_load_broadcast = load;
 255     }
 256     comm.flush(false);
 257 }
 258
 259 void process::receive(bool wait)
 260 {
 261     message* msg;
 262     m_host_t from;
 263
 264     while (may_receive() && comm.recv(msg, from, wait)) {
 265         switch (msg->get_type()) {
 266         case message::INFO: {
 267             neighbor* n = rev_neigh[from];
 268             n->set_load(msg->get_amount());
 269             break;
 270         }
 271         case message::CREDIT:
 272             expected_load += msg->get_amount();
 273             break;
 274         case message::LOAD: {
 275             double ld = msg->get_amount();
 276             load += ld;
 277             if (finalizing)
 278                 total_load_running -= ld;
 279             break;
 280         }
 281         case message::CTRL_CLOSE:
 282             ctrl_close_pending--;
 283             close_received = true;
 284             break;
 285         case message::DATA_CLOSE:
 286             data_close_pending--;
 287             close_received = true;
 288             break;
 289         }
 290         delete msg;
 291         wait = false;           // only wait on first recv
 292     }
 293     comm.flush(false);
 294 }
 295
 296 void process::finalize1(neighbor& nb)
 297 {
 298     comm.send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 299     comm.send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 300 }
 301
 302 void process::finalize()
 303 {
 304     using std::tr1::bind;
 305     using std::tr1::placeholders::_1;
 306
 307     finalizing = true;
 308     total_load_running -= load;
 309
 310     DEBUG2("send CLOSE to %lu neighbor%s",
 311            (unsigned long )neigh.size(), ESSE(neigh.size()));
 312     std::for_each(neigh.begin(), neigh.end(),
 313                   bind(&process::finalize1, this, _1));
 314
 315     DEBUG2("wait for CLOSE from %lu neighbor%s",
 316            (unsigned long )neigh.size(), ESSE(neigh.size()));
 317     while (may_receive()) {
 318         comm.flush(false);
 319         receive(true);
 320     }
 321
 322     comm.flush(true);
 323 }
 324
 325 #define print_loads_generic(vec, verbose, logp, cat)                    \
 326     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 327         using std::tr1::bind;                                           \
 328         using std::tr1::placeholders::_1;                               \
 329         XCLOG0(cat, logp, "Neighbor loads:");                           \
 330         std::for_each(vec.begin(), vec.end(),                           \
 331                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 332     } else ((void)0)
 333
 334 void process::print_loads(bool verbose,
 335                           e_xbt_log_priority_t logp,
 336                           xbt_log_category_t cat) const
 337 {
 338     print_loads_generic(neigh, verbose, logp, cat);
 339 }
 340
 341 void process::print_loads_p(bool verbose,
 342                             e_xbt_log_priority_t logp,
 343                             xbt_log_category_t cat) const
 344 {
 345     print_loads_generic(pneigh, verbose, logp, cat);
 346 }
 347
 348 #undef print_loads_generic
 349
 350 // Local variables:
 351 // mode: c++
 352 // End: