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     comp = 0.0;
  37
  38     prev_load_broadcast = -1;   // force sending of load on first send()
  39     expected_load = load;
  40     total_load_running += load;
  41     total_load_init += load;
  42
  43     ctrl_close_pending = data_close_pending = neigh.size();
  44     close_received = false;
  45     finalizing = false;
  46
  47     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  48     if (!LOG_ISENABLED(logp))
  49         return;
  50     std::ostringstream oss;
  51     oss << neigh.size() << " neighbor";
  52     if (!neigh.empty()) {
  53         oss << ESSE(neigh.size()) << ": ";
  54         std::transform(neigh.begin(), neigh.end() - 1,
  55                        std::ostream_iterator<const char*>(oss, ", "),
  56                        std::tr1::mem_fn(&neighbor::get_name));
  57         oss << neigh.back().get_name();
  58     }
  59     LOG1(logp, "Got %s.", oss.str().c_str());
  60     print_loads(false, logp);
  61 }
  62
  63 process::~process()
  64 {
  65     total_load_exit += load;
  66 }
  67
  68 int process::run()
  69 {
  70     double next_iter_after_date = 0.0;
  71     INFO1("Initial load: %g", load);
  72     VERB0("Starting...");
  73     comp_iter = lb_iter = 0;
  74     while (true) {
  75         if (opt::min_iter_duration) {
  76             double now = MSG_get_clock();
  77             if (now < next_iter_after_date){
  78                 double delay = next_iter_after_date - now;
  79                 DEBUG1("sleeping for %g s", delay);
  80                 MSG_process_sleep(delay);
  81             }
  82             next_iter_after_date = MSG_get_clock() + opt::min_iter_duration;
  83         }
  84
  85         double ld = lb_load();
  86         if (ld > 0.0) {
  87             ++lb_iter;
  88
  89             if (opt::log_rate && lb_iter % opt::log_rate == 0) {
  90                 if (opt::bookkeeping)
  91                     INFO4("(%u:%u) current load: %g ; expected: %g",
  92                           lb_iter, comp_iter, load, expected_load);
  93                 else
  94                     INFO2("(%u) current load: %g",
  95                           lb_iter, load);
  96             }
  97
  98             ld -= load_balance(ld);
  99
 100             print_loads(true, xbt_log_priority_debug);
 101         }
 102         lb_load() = ld;
 103
 104         // send load information, and load (data) if any
 105         send();
 106         if (load > 0.0) {
 107             ++comp_iter;
 108             compute();
 109         }
 110
 111         if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 112             VERB2("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 113             break;
 114         }
 115         if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 116             VERB2("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 117             break;
 118         }
 119         if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 120             VERB2("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 121             break;
 122         }
 123
 124         // block on receiving unless there is something to compute or
 125         // to send
 126         double timeout;
 127         if (load != 0 || lb_load() != prev_load_broadcast)
 128             timeout = 0.0;
 129         else if (opt::min_iter_duration)
 130             timeout = opt::min_iter_duration;
 131         else
 132             timeout = 1.0;
 133         receive(timeout);
 134
 135         // one of our neighbor is finalizing
 136         if (opt::exit_on_close && close_received) {
 137             VERB0("Close received");
 138             break;
 139         }
 140
 141         // have no load and cannot receive anything
 142         if (load == 0.0 && !may_receive()) {
 143             VERB0("I'm a poor lonesome process, and I have no load...");
 144             break;
 145         }
 146
 147         // fixme: this check should be implemented with a distributed
 148         // algorithm, and not a shared global variable!
 149         // fixme: should this chunk be moved before call to receive() ?
 150         if (100.0 * total_load_running / total_load_init <=
 151             opt::load_ratio_threshold) {
 152             VERB0("No more load to balance in system.");
 153             break;
 154         } else {
 155             DEBUG1("still %g load to balance, continuing...", total_load_running);
 156         }
 157     }
 158     VERB0("Going to finalize...");
 159     finalize();
 160
 161     /* Open Questions :
 162      * - definition of load on heterogeneous hosts ?
 163      * - how to detect convergence ?
 164      * - how to manage link failures ?
 165      */
 166
 167     VERB0("Done.");
 168     if (opt::bookkeeping) {
 169         INFO4("Final load after %d:%d iterations: %g ; expected: %g",
 170               lb_iter, comp_iter, load, expected_load);
 171     } else {
 172         INFO2("Final load after %d iterations: %g",
 173               lb_iter, load);
 174         if (lb_iter != comp_iter)
 175             WARN2("lb_iter (%d) and comp_iter (%d) differ!",
 176                   lb_iter, comp_iter);
 177     }
 178     return 0;
 179 }
 180
 181 double process::sum_of_to_send() const
 182 {
 183     using std::tr1::bind;
 184     using std::tr1::placeholders::_1;
 185     using std::tr1::placeholders::_2;
 186
 187     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 188                            bind(std::plus<double>(),
 189                                 _1, bind(&neighbor::get_to_send, _2)));
 190 }
 191
 192 double process::load_balance(double /*my_load*/)
 193 {
 194     if (lb_iter == 1)           // warn only once
 195         WARN0("process::load_balance() is a no-op!");
 196     return 0.0;
 197 }
 198
 199 void process::compute()
 200 {
 201     if (load > 0.0) {
 202         double flops = opt::comp_cost(load);
 203         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 204         DEBUG2("compute %g flop%s", flops, ESSE(flops));
 205         MSG_task_execute(task);
 206         comp += flops;
 207         MSG_task_destroy(task);
 208     } else {
 209         DEBUG0("nothing to compute !");
 210     }
 211 }
 212
 213 void process::send1_no_bookkeeping(neighbor& nb)
 214 {
 215     if (load != prev_load_broadcast)
 216         comm.send(nb.get_ctrl_mbox(), new message(message::INFO, load));
 217     double load_to_send = nb.get_to_send();
 218     if (load_to_send > 0.0) {
 219         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 220         nb.set_to_send(0.0);
 221     }
 222 }
 223
 224 void process::send1_bookkeeping(neighbor& nb)
 225 {
 226     if (expected_load != prev_load_broadcast)
 227         comm.send(nb.get_ctrl_mbox(),
 228                   new message(message::INFO, expected_load));
 229     double load_to_send;
 230     double new_debt;
 231     double debt_to_send = nb.get_to_send();
 232     if (debt_to_send > 0.0) {
 233         comm.send(nb.get_ctrl_mbox(),
 234                   new message(message::CREDIT, debt_to_send));
 235         nb.set_to_send(0.0);
 236         new_debt = nb.get_debt() + debt_to_send;
 237     } else {
 238         new_debt = nb.get_debt();
 239     }
 240     if (load <= new_debt) {
 241         load_to_send = load;
 242         nb.set_debt(new_debt - load_to_send);
 243         load = 0.0;
 244     } else {
 245         load_to_send = new_debt;
 246         nb.set_debt(0.0);
 247         load -= load_to_send;
 248     }
 249     if (load_to_send > 0.0)
 250         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 251 }
 252
 253 void process::send()
 254 {
 255     using std::tr1::bind;
 256     using std::tr1::placeholders::_1;
 257
 258     if (opt::bookkeeping) {
 259         std::for_each(neigh.begin(), neigh.end(),
 260                       bind(&process::send1_bookkeeping, this, _1));
 261         prev_load_broadcast = expected_load;
 262     } else {
 263         std::for_each(neigh.begin(), neigh.end(),
 264                       bind(&process::send1_no_bookkeeping, this, _1));
 265         prev_load_broadcast = load;
 266     }
 267     comm.flush(false);
 268 }
 269
 270 void process::receive(double timeout)
 271 {
 272     message* msg;
 273     m_host_t from;
 274
 275     DEBUG2("%sblocking receive (%g)", "\0non-" + !timeout, timeout);
 276     while (may_receive() && comm.recv(msg, from, timeout)) {
 277         switch (msg->get_type()) {
 278         case message::INFO: {
 279             neighbor* n = rev_neigh[from];
 280             n->set_load(msg->get_amount());
 281             break;
 282         }
 283         case message::CREDIT:
 284             expected_load += msg->get_amount();
 285             break;
 286         case message::LOAD: {
 287             double ld = msg->get_amount();
 288             load += ld;
 289             if (finalizing)
 290                 total_load_running -= ld;
 291             break;
 292         }
 293         case message::CTRL_CLOSE:
 294             ctrl_close_pending--;
 295             close_received = true;
 296             break;
 297         case message::DATA_CLOSE:
 298             data_close_pending--;
 299             close_received = true;
 300             break;
 301         }
 302         delete msg;
 303         timeout = 0.0;          // only wait on first recv
 304     }
 305     comm.flush(false);
 306 }
 307
 308 void process::finalize1(neighbor& nb)
 309 {
 310     comm.send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 311     comm.send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 312 }
 313
 314 void process::finalize()
 315 {
 316     using std::tr1::bind;
 317     using std::tr1::placeholders::_1;
 318
 319     finalizing = true;
 320     total_load_running -= load;
 321
 322     DEBUG2("send CLOSE to %lu neighbor%s",
 323            (unsigned long )neigh.size(), ESSE(neigh.size()));
 324     std::for_each(neigh.begin(), neigh.end(),
 325                   bind(&process::finalize1, this, _1));
 326
 327     DEBUG2("wait for CLOSE from %lu neighbor%s",
 328            (unsigned long )neigh.size(), ESSE(neigh.size()));
 329     while (may_receive()) {
 330         comm.flush(false);
 331         receive(-1.0);
 332     }
 333
 334     comm.flush(true);
 335 }
 336
 337 #define print_loads_generic(vec, verbose, logp, cat)                    \
 338     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 339         using std::tr1::bind;                                           \
 340         using std::tr1::placeholders::_1;                               \
 341         XCLOG0(cat, logp, "Neighbor loads:");                           \
 342         std::for_each(vec.begin(), vec.end(),                           \
 343                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 344     } else ((void)0)
 345
 346 void process::print_loads(bool verbose,
 347                           e_xbt_log_priority_t logp,
 348                           xbt_log_category_t cat) const
 349 {
 350     print_loads_generic(neigh, verbose, logp, cat);
 351 }
 352
 353 void process::print_loads_p(bool verbose,
 354                             e_xbt_log_priority_t logp,
 355                             xbt_log_category_t cat) const
 356 {
 357     print_loads_generic(pneigh, verbose, logp, cat);
 358 }
 359
 360 #undef print_loads_generic
 361
 362 // Local variables:
 363 // mode: c++
 364 // End: