process.cpp

   1 #include <algorithm>
   2 #include <tr1/functional>
   3 #include <iterator>
   4 #include <stdexcept>
   5 #include <sstream>
   6 #include <xbt/log.h>
   7 #include <xbt/time.h>
   8
   9 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
  10
  11 #include "misc.h"
  12 #include "options.h"
  13 #include "tracing.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]) >> real_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     comp = 0.0;
  37
  38     prev_load_broadcast = -1;   // force sending of load on first send_all()
  39     expected_load = real_load;
  40     total_load_running += real_load;
  41     total_load_init += real_load;
  42
  43     ctrl_close_pending = data_close_pending = neigh.size();
  44     close_received = false;
  45     finalizing = false;
  46
  47     comp_iter = lb_iter = 0;
  48
  49     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  50     if (!LOG_ISENABLED(logp))
  51         return;
  52     std::ostringstream oss;
  53     oss << neigh.size() << " neighbor";
  54     if (!neigh.empty()) {
  55         oss << ESSE(neigh.size()) << ": ";
  56         std::transform(neigh.begin(), neigh.end() - 1,
  57                        std::ostream_iterator<const char*>(oss, ", "),
  58                        std::tr1::mem_fn(&neighbor::get_name));
  59         oss << neigh.back().get_name();
  60     }
  61     LOG1(logp, "Got %s.", oss.str().c_str());
  62     print_loads(false, logp);
  63 }
  64
  65 process::~process()
  66 {
  67     total_load_exit += real_load;
  68     if (opt::bookkeeping) {
  69         INFO4("Final load after %d:%d iterations: %g ; expected: %g",
  70               lb_iter, comp_iter, real_load, expected_load);
  71     } else {
  72         INFO2("Final load after %d iterations: %g",
  73               lb_iter, real_load);
  74         if (lb_iter != comp_iter)
  75             WARN2("lb_iter (%d) and comp_iter (%d) differ!",
  76                   lb_iter, comp_iter);
  77     }
  78     VERB1("Total computation for this process: %g", comp);
  79 }
  80
  81 int process::run()
  82 {
  83     double next_iter_after_date = 0.0;
  84     INFO1("Initial load: %g", real_load);
  85     VERB0("Starting...");
  86     while (true) {
  87         if (get_load() > 0.0) {
  88             double now = MSG_get_clock();
  89             if (now < next_iter_after_date)
  90                 MSG_process_sleep(next_iter_after_date - now);
  91             next_iter_after_date = MSG_get_clock() + opt::min_iter_duration;
  92
  93             ++lb_iter;
  94
  95             if (opt::log_rate && lb_iter % opt::log_rate == 0) {
  96                 if (opt::bookkeeping)
  97                     INFO4("(%u:%u) current load: %g ; expected: %g",
  98                           lb_iter, comp_iter, real_load, expected_load);
  99                 else
 100                     INFO2("(%u) current load: %g",
 101                           lb_iter, real_load);
 102             }
 103
 104             load_balance();
 105
 106             print_loads(true, xbt_log_priority_debug);
 107         }
 108
 109         // send load information, and load (data) if any
 110         send_all();
 111         if (real_load > 0.0) {
 112             ++comp_iter;
 113             compute();
 114         }
 115
 116         if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 117             VERB2("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 118             break;
 119         }
 120         if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 121             VERB2("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 122             break;
 123         }
 124         if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 125             VERB2("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 126             break;
 127         }
 128
 129         // block on receiving unless there is something to compute or
 130         // to send
 131         double timeout;
 132         if (real_load != 0 || get_load() != prev_load_broadcast)
 133             timeout = 0.0;
 134         else if (opt::min_iter_duration)
 135             timeout = opt::min_iter_duration;
 136         else
 137             timeout = 1.0;
 138         receive(timeout);
 139
 140         // one of our neighbor is finalizing
 141         if (opt::exit_on_close && close_received) {
 142             VERB0("Close received");
 143             break;
 144         }
 145
 146         // have no load and cannot receive anything
 147         if (real_load == 0.0 && !may_receive()) {
 148             VERB0("I'm a poor lonesome process, and I have no load...");
 149             break;
 150         }
 151
 152         // fixme: this check should be implemented with a distributed
 153         // algorithm, and not a shared global variable!
 154         // fixme: should this chunk be moved before call to receive() ?
 155         if (100.0 * total_load_running / total_load_init <=
 156             opt::load_ratio_threshold) {
 157             VERB0("No more load to balance in system.");
 158             break;
 159         } else {
 160             DEBUG1("still %g load to balance, continuing...", total_load_running);
 161         }
 162     }
 163     VERB0("Going to finalize...");
 164     finalize();
 165
 166     /* Open Questions :
 167      * - definition of load on heterogeneous hosts ?
 168      * - how to detect convergence ?
 169      * - how to manage link failures ?
 170      */
 171
 172     VERB0("Done.");
 173     return 0;
 174 }
 175
 176 void process::load_balance()
 177 {
 178     if (lb_iter == 1)           // warn only once
 179         WARN0("process::load_balance() is a no-op!");
 180 }
 181
 182 void process::compute()
 183 {
 184     if (real_load > 0.0) {
 185         double flops = opt::comp_cost(real_load);
 186         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 187         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 188         DEBUG2("compute %g flop%s", flops, ESSE(flops));
 189         MSG_task_execute(task);
 190         comp += flops;
 191         MSG_task_destroy(task);
 192     } else {
 193         DEBUG0("nothing to compute !");
 194     }
 195 }
 196
 197 void process::send(neighbor& nb, double amount)
 198 {
 199     set_load(get_load() - amount);
 200     nb.set_to_send(nb.get_to_send() + amount);
 201     nb.set_load(nb.get_load() + amount); // fixme: make this optional?
 202 }
 203
 204 void process::send1_no_bookkeeping(neighbor& nb)
 205 {
 206     if (real_load != prev_load_broadcast)
 207         comm.send(nb.get_ctrl_mbox(), new message(message::INFO, real_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 (real_load <= new_debt) {
 232         load_to_send = real_load;
 233         nb.set_debt(new_debt - load_to_send);
 234         real_load = 0.0;
 235     } else {
 236         load_to_send = new_debt;
 237         nb.set_debt(0.0);
 238         real_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_all()
 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 = real_load;
 257     }
 258     comm.flush(false);
 259 }
 260
 261 void process::receive(double timeout)
 262 {
 263     message* msg;
 264     m_host_t from;
 265
 266     DEBUG2("%sblocking receive (%g)", "\0non-" + !timeout, timeout);
 267     while (may_receive() && comm.recv(msg, from, timeout)) {
 268         switch (msg->get_type()) {
 269         case message::INFO: {
 270             neighbor* n = rev_neigh[from];
 271             n->set_load(msg->get_amount());
 272             break;
 273         }
 274         case message::CREDIT:
 275             expected_load += msg->get_amount();
 276             break;
 277         case message::LOAD: {
 278             double ld = msg->get_amount();
 279             real_load += ld;
 280             if (finalizing)
 281                 total_load_running -= ld;
 282             break;
 283         }
 284         case message::CTRL_CLOSE:
 285             ctrl_close_pending--;
 286             close_received = true;
 287             break;
 288         case message::DATA_CLOSE:
 289             data_close_pending--;
 290             close_received = true;
 291             break;
 292         }
 293         delete msg;
 294         timeout = 0.0;          // only wait on first recv
 295     }
 296     comm.flush(false);
 297 }
 298
 299 void process::finalize1(neighbor& nb)
 300 {
 301     comm.send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 302     comm.send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 303 }
 304
 305 void process::finalize()
 306 {
 307     using std::tr1::bind;
 308     using std::tr1::placeholders::_1;
 309
 310     finalizing = true;
 311     total_load_running -= real_load;
 312
 313     DEBUG2("send CLOSE to %lu neighbor%s",
 314            (unsigned long )neigh.size(), ESSE(neigh.size()));
 315     std::for_each(neigh.begin(), neigh.end(),
 316                   bind(&process::finalize1, this, _1));
 317
 318     while (may_receive()) {
 319         comm.flush(false);
 320         DEBUG2("waiting for %d CTRL and %d DATA CLOSE",
 321                ctrl_close_pending, data_close_pending);
 322         receive(-1.0);
 323     }
 324
 325     comm.flush(true);
 326 }
 327
 328 #define print_loads_generic(vec, verbose, logp, cat)                    \
 329     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 330         using std::tr1::bind;                                           \
 331         using std::tr1::placeholders::_1;                               \
 332         XCLOG0(cat, logp, "Neighbor loads:");                           \
 333         std::for_each(vec.begin(), vec.end(),                           \
 334                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 335     } else ((void)0)
 336
 337 void process::print_loads(bool verbose,
 338                           e_xbt_log_priority_t logp,
 339                           xbt_log_category_t cat) const
 340 {
 341     print_loads_generic(neigh, verbose, logp, cat);
 342 }
 343
 344 void process::print_loads_p(bool verbose,
 345                             e_xbt_log_priority_t logp,
 346                             xbt_log_category_t cat) const
 347 {
 348     print_loads_generic(pneigh, verbose, logp, cat);
 349 }
 350
 351 #undef print_loads_generic
 352
 353 // Local variables:
 354 // mode: c++
 355 // End: