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