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             break;
 111         if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter)
 112             break;
 113         if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 114             VERB2("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 115             break;
 116         }
 117
 118         // block on receiving unless there is something to compute or
 119         // to send
 120         bool wait = (load == 0 &&
 121                      ((opt::bookkeeping ? expected_load : load)
 122                       == prev_load_broadcast));
 123         receive(wait);
 124
 125         // one of our neighbor is finalizing
 126         if (opt::exit_on_close && close_received)
 127             break;
 128
 129         // have no load and cannot receive anything
 130         if (load == 0.0 && !may_receive())
 131             break;
 132
 133         // fixme: this check should be implemented with a distributed
 134         // algorithm, and not a shared global variable!
 135         // fixme: should this chunk be moved before call to receive() ?
 136         if (100.0 * total_load_running / total_load_init <=
 137             opt::load_ratio_threshold) {
 138             VERB0("No more load to balance in system, stopping.");
 139             break;
 140         }
 141
 142     }
 143     VERB0("Going to finalize...");
 144     finalize();
 145
 146     /* Open Questions :
 147      * - definition of load on heterogeneous hosts ?
 148      * - how to detect convergence ?
 149      * - how to manage link failures ?
 150      */
 151
 152     VERB0("Done.");
 153     INFO3("Final load after %d iteration%s: %g",
 154           comp_iter, ESSE(comp_iter), load);
 155     if (opt::bookkeeping)
 156         INFO1("Expected load: %g", expected_load);
 157     return 0;
 158 }
 159
 160 double process::sum_of_to_send() const
 161 {
 162     using std::tr1::bind;
 163     using std::tr1::placeholders::_1;
 164     using std::tr1::placeholders::_2;
 165
 166     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 167                            bind(std::plus<double>(),
 168                                 _1, bind(&neighbor::get_to_send, _2)));
 169 }
 170
 171 double process::load_balance(double /*my_load*/)
 172 {
 173     if (lb_iter == 1)           // warn only once
 174         WARN0("process::load_balance() is a no-op!");
 175     return 0.0;
 176 }
 177
 178 void process::compute()
 179 {
 180     if (load > 0.0) {
 181         double duration = opt::comp_cost(load);
 182         m_task_t task = MSG_task_create("computation", duration, 0.0, NULL);
 183         DEBUG2("compute %g flop%s", duration, ESSE(duration));
 184         MSG_task_execute(task);
 185         MSG_task_destroy(task);
 186     } else {
 187         DEBUG0("nothing to compute !");
 188     }
 189 }
 190
 191 void process::send1_no_bookkeeping(neighbor& nb)
 192 {
 193     if (load != prev_load_broadcast)
 194         comm.send(nb.get_ctrl_mbox(), new message(message::INFO, load));
 195     double load_to_send = nb.get_to_send();
 196     if (load_to_send > 0.0) {
 197         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 198         nb.set_to_send(0.0);
 199     }
 200 }
 201
 202 void process::send1_bookkeeping(neighbor& nb)
 203 {
 204     if (expected_load != prev_load_broadcast)
 205         comm.send(nb.get_ctrl_mbox(),
 206                   new message(message::INFO, expected_load));
 207     double load_to_send;
 208     double new_debt;
 209     double debt_to_send = nb.get_to_send();
 210     if (debt_to_send > 0.0) {
 211         comm.send(nb.get_ctrl_mbox(),
 212                   new message(message::CREDIT, debt_to_send));
 213         nb.set_to_send(0.0);
 214         new_debt = nb.get_debt() + debt_to_send;
 215     } else {
 216         new_debt = nb.get_debt();
 217     }
 218     if (load <= new_debt) {
 219         load_to_send = load;
 220         nb.set_debt(new_debt - load_to_send);
 221         load = 0.0;
 222     } else {
 223         load_to_send = new_debt;
 224         nb.set_debt(0.0);
 225         load -= load_to_send;
 226     }
 227     if (load_to_send > 0.0)
 228         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 229 }
 230
 231 void process::send()
 232 {
 233     using std::tr1::bind;
 234     using std::tr1::placeholders::_1;
 235
 236     if (opt::bookkeeping) {
 237         std::for_each(neigh.begin(), neigh.end(),
 238                       bind(&process::send1_bookkeeping, this, _1));
 239         prev_load_broadcast = expected_load;
 240     } else {
 241         std::for_each(neigh.begin(), neigh.end(),
 242                       bind(&process::send1_no_bookkeeping, this, _1));
 243         prev_load_broadcast = load;
 244     }
 245     comm.flush(false);
 246 }
 247
 248 void process::receive(bool wait)
 249 {
 250     message* msg;
 251     m_host_t from;
 252
 253     while (may_receive() && comm.recv(msg, from, wait)) {
 254         switch (msg->get_type()) {
 255         case message::INFO: {
 256             neighbor* n = rev_neigh[from];
 257             n->set_load(msg->get_amount());
 258             break;
 259         }
 260         case message::CREDIT:
 261             expected_load += msg->get_amount();
 262             break;
 263         case message::LOAD: {
 264             double ld = msg->get_amount();
 265             load += ld;
 266             if (finalizing)
 267                 total_load_running -= ld;
 268             break;
 269         }
 270         case message::CTRL_CLOSE:
 271             ctrl_close_pending--;
 272             close_received = true;
 273             break;
 274         case message::DATA_CLOSE:
 275             data_close_pending--;
 276             close_received = true;
 277             break;
 278         }
 279         delete msg;
 280         wait = false;           // only wait on first recv
 281     }
 282     comm.flush(false);
 283 }
 284
 285 void process::finalize1(neighbor& nb)
 286 {
 287     comm.send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 288     comm.send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 289 }
 290
 291 void process::finalize()
 292 {
 293     using std::tr1::bind;
 294     using std::tr1::placeholders::_1;
 295
 296     finalizing = true;
 297     total_load_running -= load;
 298
 299     DEBUG2("send CLOSE to %lu neighbor%s",
 300            (unsigned long )neigh.size(), ESSE(neigh.size()));
 301     std::for_each(neigh.begin(), neigh.end(),
 302                   bind(&process::finalize1, this, _1));
 303
 304     DEBUG2("wait for CLOSE from %lu neighbor%s",
 305            (unsigned long )neigh.size(), ESSE(neigh.size()));
 306     while (may_receive()) {
 307         comm.flush(false);
 308         receive(true);
 309     }
 310
 311     comm.flush(true);
 312 }
 313
 314 #define print_loads_generic(vec, verbose, logp, cat)                    \
 315     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 316         using std::tr1::bind;                                           \
 317         using std::tr1::placeholders::_1;                               \
 318         XCLOG0(cat, logp, "Neighbor loads:");                           \
 319         std::for_each(vec.begin(), vec.end(),                           \
 320                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 321     } else ((void)0)
 322
 323 void process::print_loads(bool verbose,
 324                           e_xbt_log_priority_t logp,
 325                           xbt_log_category_t cat) const
 326 {
 327     print_loads_generic(neigh, verbose, logp, cat);
 328 }
 329
 330 void process::print_loads_p(bool verbose,
 331                             e_xbt_log_priority_t logp,
 332                             xbt_log_category_t cat) const
 333 {
 334     print_loads_generic(pneigh, verbose, logp, cat);
 335 }
 336
 337 #undef print_loads_generic
 338
 339 // Local variables:
 340 // mode: c++
 341 // End: