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         bool wait = (load == 0 && lb_load() == prev_load_broadcast);
 127         receive(wait);
 128
 129         // one of our neighbor is finalizing
 130         if (opt::exit_on_close && close_received) {
 131             VERB0("Close received");
 132             break;
 133         }
 134
 135         // have no load and cannot receive anything
 136         if (load == 0.0 && !may_receive()) {
 137             VERB0("I'm a poor lonesome process, and I have no load...");
 138             break;
 139         }
 140
 141         // fixme: this check should be implemented with a distributed
 142         // algorithm, and not a shared global variable!
 143         // fixme: should this chunk be moved before call to receive() ?
 144         if (100.0 * total_load_running / total_load_init <=
 145             opt::load_ratio_threshold) {
 146             VERB0("No more load to balance in system.");
 147             break;
 148         } else {
 149             DEBUG1("still %g load to balance, continuing...", total_load_running);
 150         }
 151     }
 152     VERB0("Going to finalize...");
 153     finalize();
 154
 155     /* Open Questions :
 156      * - definition of load on heterogeneous hosts ?
 157      * - how to detect convergence ?
 158      * - how to manage link failures ?
 159      */
 160
 161     VERB0("Done.");
 162     if (opt::bookkeeping) {
 163         INFO4("Final load after %d:%d iterations: %g ; expected: %g",
 164               lb_iter, comp_iter, load, expected_load);
 165     } else {
 166         INFO2("Final load after %d iterations: %g",
 167               lb_iter, load);
 168         if (lb_iter != comp_iter)
 169             WARN2("lb_iter (%d) and comp_iter (%d) differ!",
 170                   lb_iter, comp_iter);
 171     }
 172     return 0;
 173 }
 174
 175 double process::sum_of_to_send() const
 176 {
 177     using std::tr1::bind;
 178     using std::tr1::placeholders::_1;
 179     using std::tr1::placeholders::_2;
 180
 181     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 182                            bind(std::plus<double>(),
 183                                 _1, bind(&neighbor::get_to_send, _2)));
 184 }
 185
 186 double process::load_balance(double /*my_load*/)
 187 {
 188     if (lb_iter == 1)           // warn only once
 189         WARN0("process::load_balance() is a no-op!");
 190     return 0.0;
 191 }
 192
 193 void process::compute()
 194 {
 195     if (load > 0.0) {
 196         double flops = opt::comp_cost(load);
 197         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 198         DEBUG2("compute %g flop%s", flops, ESSE(flops));
 199         MSG_task_execute(task);
 200         comp += flops;
 201         MSG_task_destroy(task);
 202     } else {
 203         DEBUG0("nothing to compute !");
 204     }
 205 }
 206
 207 void process::send1_no_bookkeeping(neighbor& nb)
 208 {
 209     if (load != prev_load_broadcast)
 210         comm.send(nb.get_ctrl_mbox(), new message(message::INFO, load));
 211     double load_to_send = nb.get_to_send();
 212     if (load_to_send > 0.0) {
 213         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 214         nb.set_to_send(0.0);
 215     }
 216 }
 217
 218 void process::send1_bookkeeping(neighbor& nb)
 219 {
 220     if (expected_load != prev_load_broadcast)
 221         comm.send(nb.get_ctrl_mbox(),
 222                   new message(message::INFO, expected_load));
 223     double load_to_send;
 224     double new_debt;
 225     double debt_to_send = nb.get_to_send();
 226     if (debt_to_send > 0.0) {
 227         comm.send(nb.get_ctrl_mbox(),
 228                   new message(message::CREDIT, debt_to_send));
 229         nb.set_to_send(0.0);
 230         new_debt = nb.get_debt() + debt_to_send;
 231     } else {
 232         new_debt = nb.get_debt();
 233     }
 234     if (load <= new_debt) {
 235         load_to_send = load;
 236         nb.set_debt(new_debt - load_to_send);
 237         load = 0.0;
 238     } else {
 239         load_to_send = new_debt;
 240         nb.set_debt(0.0);
 241         load -= load_to_send;
 242     }
 243     if (load_to_send > 0.0)
 244         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 245 }
 246
 247 void process::send()
 248 {
 249     using std::tr1::bind;
 250     using std::tr1::placeholders::_1;
 251
 252     if (opt::bookkeeping) {
 253         std::for_each(neigh.begin(), neigh.end(),
 254                       bind(&process::send1_bookkeeping, this, _1));
 255         prev_load_broadcast = expected_load;
 256     } else {
 257         std::for_each(neigh.begin(), neigh.end(),
 258                       bind(&process::send1_no_bookkeeping, this, _1));
 259         prev_load_broadcast = load;
 260     }
 261     comm.flush(false);
 262 }
 263
 264 void process::receive(bool wait)
 265 {
 266     message* msg;
 267     m_host_t from;
 268
 269     DEBUG1("%sblocking receive", "\0non-" + !wait);
 270     while (may_receive() && comm.recv(msg, from, wait)) {
 271         switch (msg->get_type()) {
 272         case message::INFO: {
 273             neighbor* n = rev_neigh[from];
 274             n->set_load(msg->get_amount());
 275             break;
 276         }
 277         case message::CREDIT:
 278             expected_load += msg->get_amount();
 279             break;
 280         case message::LOAD: {
 281             double ld = msg->get_amount();
 282             load += ld;
 283             if (finalizing)
 284                 total_load_running -= ld;
 285             break;
 286         }
 287         case message::CTRL_CLOSE:
 288             ctrl_close_pending--;
 289             close_received = true;
 290             break;
 291         case message::DATA_CLOSE:
 292             data_close_pending--;
 293             close_received = true;
 294             break;
 295         }
 296         delete msg;
 297         wait = false;           // only wait on first recv
 298     }
 299     comm.flush(false);
 300 }
 301
 302 void process::finalize1(neighbor& nb)
 303 {
 304     comm.send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 305     comm.send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 306 }
 307
 308 void process::finalize()
 309 {
 310     using std::tr1::bind;
 311     using std::tr1::placeholders::_1;
 312
 313     finalizing = true;
 314     total_load_running -= load;
 315
 316     DEBUG2("send CLOSE to %lu neighbor%s",
 317            (unsigned long )neigh.size(), ESSE(neigh.size()));
 318     std::for_each(neigh.begin(), neigh.end(),
 319                   bind(&process::finalize1, this, _1));
 320
 321     DEBUG2("wait for CLOSE from %lu neighbor%s",
 322            (unsigned long )neigh.size(), ESSE(neigh.size()));
 323     while (may_receive()) {
 324         comm.flush(false);
 325         receive(true);
 326     }
 327
 328     comm.flush(true);
 329 }
 330
 331 #define print_loads_generic(vec, verbose, logp, cat)                    \
 332     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 333         using std::tr1::bind;                                           \
 334         using std::tr1::placeholders::_1;                               \
 335         XCLOG0(cat, logp, "Neighbor loads:");                           \
 336         std::for_each(vec.begin(), vec.end(),                           \
 337                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 338     } else ((void)0)
 339
 340 void process::print_loads(bool verbose,
 341                           e_xbt_log_priority_t logp,
 342                           xbt_log_category_t cat) const
 343 {
 344     print_loads_generic(neigh, verbose, logp, cat);
 345 }
 346
 347 void process::print_loads_p(bool verbose,
 348                             e_xbt_log_priority_t logp,
 349                             xbt_log_category_t cat) const
 350 {
 351     print_loads_generic(pneigh, verbose, logp, cat);
 352 }
 353
 354 #undef print_loads_generic
 355
 356 // Local variables:
 357 // mode: c++
 358 // End: