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     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 += real_load;
  66     if (opt::bookkeeping) {
  67         INFO4("Final load after %d:%d iterations: %g ; expected: %g",
  68               lb_iter, comp_iter, real_load, expected_load);
  69     } else {
  70         INFO2("Final load after %d iterations: %g",
  71               lb_iter, real_load);
  72         if (lb_iter != comp_iter)
  73             WARN2("lb_iter (%d) and comp_iter (%d) differ!",
  74                   lb_iter, comp_iter);
  75     }
  76     VERB1("Total computation for this process: %g", comp);
  77 }
  78
  79 int process::run()
  80 {
  81     double next_iter_after_date = 0.0;
  82     INFO1("Initial load: %g", real_load);
  83     VERB0("Starting...");
  84     comp_iter = lb_iter = 0;
  85     while (true) {
  86         if (get_load() > 0.0) {
  87             double now = MSG_get_clock();
  88             if (now < next_iter_after_date)
  89                 MSG_process_sleep(next_iter_after_date - now);
  90             next_iter_after_date = MSG_get_clock() + opt::min_iter_duration;
  91
  92             ++lb_iter;
  93
  94             if (opt::log_rate && lb_iter % opt::log_rate == 0) {
  95                 if (opt::bookkeeping)
  96                     INFO4("(%u:%u) current load: %g ; expected: %g",
  97                           lb_iter, comp_iter, real_load, expected_load);
  98                 else
  99                     INFO2("(%u) current load: %g",
 100                           lb_iter, real_load);
 101             }
 102
 103             load_balance();
 104
 105             print_loads(true, xbt_log_priority_debug);
 106         }
 107
 108         // send load information, and load (data) if any
 109         send_all();
 110         if (real_load > 0.0) {
 111             ++comp_iter;
 112             compute();
 113         }
 114
 115         if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 116             VERB2("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 117             break;
 118         }
 119         if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 120             VERB2("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 121             break;
 122         }
 123         if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 124             VERB2("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 125             break;
 126         }
 127
 128         // block on receiving unless there is something to compute or
 129         // to send
 130         double timeout;
 131         if (real_load != 0 || get_load() != prev_load_broadcast)
 132             timeout = 0.0;
 133         else if (opt::min_iter_duration)
 134             timeout = opt::min_iter_duration;
 135         else
 136             timeout = 1.0;
 137         receive(timeout);
 138
 139         // one of our neighbor is finalizing
 140         if (opt::exit_on_close && close_received) {
 141             VERB0("Close received");
 142             break;
 143         }
 144
 145         // have no load and cannot receive anything
 146         if (real_load == 0.0 && !may_receive()) {
 147             VERB0("I'm a poor lonesome process, and I have no load...");
 148             break;
 149         }
 150
 151         // fixme: this check should be implemented with a distributed
 152         // algorithm, and not a shared global variable!
 153         // fixme: should this chunk be moved before call to receive() ?
 154         if (100.0 * total_load_running / total_load_init <=
 155             opt::load_ratio_threshold) {
 156             VERB0("No more load to balance in system.");
 157             break;
 158         } else {
 159             DEBUG1("still %g load to balance, continuing...", total_load_running);
 160         }
 161     }
 162     VERB0("Going to finalize...");
 163     finalize();
 164
 165     /* Open Questions :
 166      * - definition of load on heterogeneous hosts ?
 167      * - how to detect convergence ?
 168      * - how to manage link failures ?
 169      */
 170
 171     VERB0("Done.");
 172     return 0;
 173 }
 174
 175 void process::load_balance()
 176 {
 177     if (lb_iter == 1)           // warn only once
 178         WARN0("process::load_balance() is a no-op!");
 179 }
 180
 181 void process::compute()
 182 {
 183     if (real_load > 0.0) {
 184         double flops = opt::comp_cost(real_load);
 185         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 186         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 187         DEBUG2("compute %g flop%s", flops, ESSE(flops));
 188         MSG_task_execute(task);
 189         comp += flops;
 190         MSG_task_destroy(task);
 191     } else {
 192         DEBUG0("nothing to compute !");
 193     }
 194 }
 195
 196 void process::send(neighbor& nb, double amount)
 197 {
 198     set_load(get_load() - amount);
 199     nb.set_to_send(nb.get_to_send() + amount);
 200     nb.set_load(nb.get_load() + amount); // fixme: make this optional?
 201 }
 202
 203 void process::send1_no_bookkeeping(neighbor& nb)
 204 {
 205     if (real_load != prev_load_broadcast)
 206         comm.send(nb.get_ctrl_mbox(), new message(message::INFO, real_load));
 207     double load_to_send = nb.get_to_send();
 208     if (load_to_send > 0.0) {
 209         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 210         nb.set_to_send(0.0);
 211     }
 212 }
 213
 214 void process::send1_bookkeeping(neighbor& nb)
 215 {
 216     if (expected_load != prev_load_broadcast)
 217         comm.send(nb.get_ctrl_mbox(),
 218                   new message(message::INFO, expected_load));
 219     double load_to_send;
 220     double new_debt;
 221     double debt_to_send = nb.get_to_send();
 222     if (debt_to_send > 0.0) {
 223         comm.send(nb.get_ctrl_mbox(),
 224                   new message(message::CREDIT, debt_to_send));
 225         nb.set_to_send(0.0);
 226         new_debt = nb.get_debt() + debt_to_send;
 227     } else {
 228         new_debt = nb.get_debt();
 229     }
 230     if (real_load <= new_debt) {
 231         load_to_send = real_load;
 232         nb.set_debt(new_debt - load_to_send);
 233         real_load = 0.0;
 234     } else {
 235         load_to_send = new_debt;
 236         nb.set_debt(0.0);
 237         real_load -= load_to_send;
 238     }
 239     if (load_to_send > 0.0)
 240         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 241 }
 242
 243 void process::send_all()
 244 {
 245     using std::tr1::bind;
 246     using std::tr1::placeholders::_1;
 247
 248     if (opt::bookkeeping) {
 249         std::for_each(neigh.begin(), neigh.end(),
 250                       bind(&process::send1_bookkeeping, this, _1));
 251         prev_load_broadcast = expected_load;
 252     } else {
 253         std::for_each(neigh.begin(), neigh.end(),
 254                       bind(&process::send1_no_bookkeeping, this, _1));
 255         prev_load_broadcast = real_load;
 256     }
 257     comm.flush(false);
 258 }
 259
 260 void process::receive(double timeout)
 261 {
 262     message* msg;
 263     m_host_t from;
 264
 265     DEBUG2("%sblocking receive (%g)", "\0non-" + !timeout, timeout);
 266     while (may_receive() && comm.recv(msg, from, timeout)) {
 267         switch (msg->get_type()) {
 268         case message::INFO: {
 269             neighbor* n = rev_neigh[from];
 270             n->set_load(msg->get_amount());
 271             break;
 272         }
 273         case message::CREDIT:
 274             expected_load += msg->get_amount();
 275             break;
 276         case message::LOAD: {
 277             double ld = msg->get_amount();
 278             real_load += ld;
 279             if (finalizing)
 280                 total_load_running -= ld;
 281             break;
 282         }
 283         case message::CTRL_CLOSE:
 284             ctrl_close_pending--;
 285             close_received = true;
 286             break;
 287         case message::DATA_CLOSE:
 288             data_close_pending--;
 289             close_received = true;
 290             break;
 291         }
 292         delete msg;
 293         timeout = 0.0;          // only wait on first recv
 294     }
 295     comm.flush(false);
 296 }
 297
 298 void process::finalize1(neighbor& nb)
 299 {
 300     comm.send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 301     comm.send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 302 }
 303
 304 void process::finalize()
 305 {
 306     using std::tr1::bind;
 307     using std::tr1::placeholders::_1;
 308
 309     finalizing = true;
 310     total_load_running -= real_load;
 311
 312     DEBUG2("send CLOSE to %lu neighbor%s",
 313            (unsigned long )neigh.size(), ESSE(neigh.size()));
 314     std::for_each(neigh.begin(), neigh.end(),
 315                   bind(&process::finalize1, this, _1));
 316
 317     DEBUG2("wait for CLOSE from %lu neighbor%s",
 318            (unsigned long )neigh.size(), ESSE(neigh.size()));
 319     while (may_receive()) {
 320         comm.flush(false);
 321         receive(-1.0);
 322     }
 323
 324     comm.flush(true);
 325 }
 326
 327 #define print_loads_generic(vec, verbose, logp, cat)                    \
 328     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 329         using std::tr1::bind;                                           \
 330         using std::tr1::placeholders::_1;                               \
 331         XCLOG0(cat, logp, "Neighbor loads:");                           \
 332         std::for_each(vec.begin(), vec.end(),                           \
 333                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 334     } else ((void)0)
 335
 336 void process::print_loads(bool verbose,
 337                           e_xbt_log_priority_t logp,
 338                           xbt_log_category_t cat) const
 339 {
 340     print_loads_generic(neigh, verbose, logp, cat);
 341 }
 342
 343 void process::print_loads_p(bool verbose,
 344                             e_xbt_log_priority_t logp,
 345                             xbt_log_category_t cat) const
 346 {
 347     print_loads_generic(pneigh, verbose, logp, cat);
 348 }
 349
 350 #undef print_loads_generic
 351
 352 // Local variables:
 353 // mode: c++
 354 // End: