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