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