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             double ld = lb_load();
  95             ld -= load_balance(ld);
  96             lb_load() = ld;
  97
  98             print_loads(true, xbt_log_priority_debug);
  99
 100             send();
 101             compute();
 102
 103         } else {
 104             // send load information, and load when bookkeeping
 105             send();
 106         }
 107
 108         if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 109             VERB2("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 110             break;
 111         }
 112         if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 113             VERB2("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 114             break;
 115         }
 116         if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 117             VERB2("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 118             break;
 119         }
 120
 121         // block on receiving unless there is something to compute or
 122         // to send
 123         bool wait = (load == 0 && lb_load() == prev_load_broadcast);
 124         receive(wait);
 125
 126         // one of our neighbor is finalizing
 127         if (opt::exit_on_close && close_received) {
 128             VERB0("Close received");
 129             break;
 130         }
 131
 132         // have no load and cannot receive anything
 133         if (load == 0.0 && !may_receive()) {
 134             VERB0("I'm a poor lonesome process, and I have no load...");
 135             break;
 136         }
 137
 138         // fixme: this check should be implemented with a distributed
 139         // algorithm, and not a shared global variable!
 140         // fixme: should this chunk be moved before call to receive() ?
 141         if (100.0 * total_load_running / total_load_init <=
 142             opt::load_ratio_threshold) {
 143             VERB0("No more load to balance in system.");
 144             break;
 145         }
 146     }
 147     VERB0("Going to finalize...");
 148     finalize();
 149
 150     /* Open Questions :
 151      * - definition of load on heterogeneous hosts ?
 152      * - how to detect convergence ?
 153      * - how to manage link failures ?
 154      */
 155
 156     VERB0("Done.");
 157     if (opt::bookkeeping) {
 158         INFO4("Final load after %d:%d iterations: %g ; expected: %g",
 159               comp_iter, lb_iter, load, expected_load);
 160     } else {
 161         INFO2("Final load after %d iterations: %g",
 162               comp_iter, load);
 163         if (comp_iter != lb_iter)
 164             WARN2("comp_iter (%d) and lb_iter (%d) differ!",
 165                   comp_iter, lb_iter);
 166     }
 167     return 0;
 168 }
 169
 170 double process::sum_of_to_send() const
 171 {
 172     using std::tr1::bind;
 173     using std::tr1::placeholders::_1;
 174     using std::tr1::placeholders::_2;
 175
 176     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 177                            bind(std::plus<double>(),
 178                                 _1, bind(&neighbor::get_to_send, _2)));
 179 }
 180
 181 double process::load_balance(double /*my_load*/)
 182 {
 183     if (lb_iter == 1)           // warn only once
 184         WARN0("process::load_balance() is a no-op!");
 185     return 0.0;
 186 }
 187
 188 void process::compute()
 189 {
 190     if (load > 0.0) {
 191         double duration = opt::comp_cost(load);
 192         m_task_t task = MSG_task_create("computation", duration, 0.0, NULL);
 193         DEBUG2("compute %g flop%s", duration, ESSE(duration));
 194         MSG_task_execute(task);
 195         MSG_task_destroy(task);
 196     } else {
 197         DEBUG0("nothing to compute !");
 198     }
 199 }
 200
 201 void process::send1_no_bookkeeping(neighbor& nb)
 202 {
 203     if (load != prev_load_broadcast)
 204         comm.send(nb.get_ctrl_mbox(), new message(message::INFO, load));
 205     double load_to_send = nb.get_to_send();
 206     if (load_to_send > 0.0) {
 207         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 208         nb.set_to_send(0.0);
 209     }
 210 }
 211
 212 void process::send1_bookkeeping(neighbor& nb)
 213 {
 214     if (expected_load != prev_load_broadcast)
 215         comm.send(nb.get_ctrl_mbox(),
 216                   new message(message::INFO, expected_load));
 217     double load_to_send;
 218     double new_debt;
 219     double debt_to_send = nb.get_to_send();
 220     if (debt_to_send > 0.0) {
 221         comm.send(nb.get_ctrl_mbox(),
 222                   new message(message::CREDIT, debt_to_send));
 223         nb.set_to_send(0.0);
 224         new_debt = nb.get_debt() + debt_to_send;
 225     } else {
 226         new_debt = nb.get_debt();
 227     }
 228     if (load <= new_debt) {
 229         load_to_send = load;
 230         nb.set_debt(new_debt - load_to_send);
 231         load = 0.0;
 232     } else {
 233         load_to_send = new_debt;
 234         nb.set_debt(0.0);
 235         load -= load_to_send;
 236     }
 237     if (load_to_send > 0.0)
 238         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 239 }
 240
 241 void process::send()
 242 {
 243     using std::tr1::bind;
 244     using std::tr1::placeholders::_1;
 245
 246     if (opt::bookkeeping) {
 247         std::for_each(neigh.begin(), neigh.end(),
 248                       bind(&process::send1_bookkeeping, this, _1));
 249         prev_load_broadcast = expected_load;
 250     } else {
 251         std::for_each(neigh.begin(), neigh.end(),
 252                       bind(&process::send1_no_bookkeeping, this, _1));
 253         prev_load_broadcast = load;
 254     }
 255     comm.flush(false);
 256 }
 257
 258 void process::receive(bool wait)
 259 {
 260     message* msg;
 261     m_host_t from;
 262
 263     while (may_receive() && comm.recv(msg, from, wait)) {
 264         switch (msg->get_type()) {
 265         case message::INFO: {
 266             neighbor* n = rev_neigh[from];
 267             n->set_load(msg->get_amount());
 268             break;
 269         }
 270         case message::CREDIT:
 271             expected_load += msg->get_amount();
 272             break;
 273         case message::LOAD: {
 274             double ld = msg->get_amount();
 275             load += ld;
 276             if (finalizing)
 277                 total_load_running -= ld;
 278             break;
 279         }
 280         case message::CTRL_CLOSE:
 281             ctrl_close_pending--;
 282             close_received = true;
 283             break;
 284         case message::DATA_CLOSE:
 285             data_close_pending--;
 286             close_received = true;
 287             break;
 288         }
 289         delete msg;
 290         wait = false;           // only wait on first recv
 291     }
 292     comm.flush(false);
 293 }
 294
 295 void process::finalize1(neighbor& nb)
 296 {
 297     comm.send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 298     comm.send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 299 }
 300
 301 void process::finalize()
 302 {
 303     using std::tr1::bind;
 304     using std::tr1::placeholders::_1;
 305
 306     finalizing = true;
 307     total_load_running -= load;
 308
 309     DEBUG2("send CLOSE to %lu neighbor%s",
 310            (unsigned long )neigh.size(), ESSE(neigh.size()));
 311     std::for_each(neigh.begin(), neigh.end(),
 312                   bind(&process::finalize1, this, _1));
 313
 314     DEBUG2("wait for CLOSE from %lu neighbor%s",
 315            (unsigned long )neigh.size(), ESSE(neigh.size()));
 316     while (may_receive()) {
 317         comm.flush(false);
 318         receive(true);
 319     }
 320
 321     comm.flush(true);
 322 }
 323
 324 #define print_loads_generic(vec, verbose, logp, cat)                    \
 325     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 326         using std::tr1::bind;                                           \
 327         using std::tr1::placeholders::_1;                               \
 328         XCLOG0(cat, logp, "Neighbor loads:");                           \
 329         std::for_each(vec.begin(), vec.end(),                           \
 330                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 331     } else ((void)0)
 332
 333 void process::print_loads(bool verbose,
 334                           e_xbt_log_priority_t logp,
 335                           xbt_log_category_t cat) const
 336 {
 337     print_loads_generic(neigh, verbose, logp, cat);
 338 }
 339
 340 void process::print_loads_p(bool verbose,
 341                             e_xbt_log_priority_t logp,
 342                             xbt_log_category_t cat) const
 343 {
 344     print_loads_generic(pneigh, verbose, logp, cat);
 345 }
 346
 347 #undef print_loads_generic
 348
 349 // Local variables:
 350 // mode: c++
 351 // End: