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     prev_load_broadcast = -1;   // force sending of load on first send()
  37     expected_load = load;
  38     total_load_running += load;
  39     total_load_init += load;
  40
  41     ctrl_close_pending = data_close_pending = neigh.size();
  42     close_received = false;
  43     finalizing = false;
  44
  45     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  46     if (!LOG_ISENABLED(logp))
  47         return;
  48     std::ostringstream oss;
  49     oss << neigh.size() << " neighbor";
  50     if (!neigh.empty()) {
  51         oss << ESSE(neigh.size()) << ": ";
  52         std::transform(neigh.begin(), neigh.end() - 1,
  53                        std::ostream_iterator<const char*>(oss, ", "),
  54                        std::tr1::mem_fn(&neighbor::get_name));
  55         oss << neigh.back().get_name();
  56     }
  57     LOG1(logp, "Got %s.", oss.str().c_str());
  58     print_loads(false, logp);
  59 }
  60
  61 process::~process()
  62 {
  63     total_load_exit += load;
  64 }
  65
  66 int process::run()
  67 {
  68     INFO1("Initial load: %g", load);
  69     VERB0("Starting...");
  70     comp_iter = lb_iter = 0;
  71     while (true) {
  72         if (load > 0.0) {
  73             ++comp_iter;
  74             if (opt::log_rate && comp_iter % opt::log_rate == 0) {
  75                 if (opt::bookkeeping)
  76                     INFO4("(%u:%u) current load: %g ; expected: %g",
  77                           comp_iter, lb_iter, load, expected_load);
  78                 else
  79                     INFO2("(%u) current load: %g",
  80                           comp_iter, load);
  81             }
  82
  83             if (opt::bookkeeping)
  84                 expected_load -= load_balance(expected_load);
  85             else
  86                 load -= load_balance(load);
  87
  88             print_loads(true, xbt_log_priority_debug);
  89
  90             send();
  91             compute();
  92
  93         } else {
  94             // send load information, and load when bookkeeping
  95             send();
  96         }
  97
  98         if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter)
  99             break;
 100         if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter)
 101             break;
 102
 103         // block on receiving unless there is something to compute or
 104         // to send
 105         bool wait = (load == 0 &&
 106                      ((opt::bookkeeping ? expected_load : load)
 107                       == prev_load_broadcast));
 108         receive(wait);
 109
 110         // one of our neighbor is finalizing
 111         if (opt::exit_on_close && close_received)
 112             break;
 113
 114         // have no load and cannot receive anything
 115         if (load == 0.0 && !may_receive())
 116             break;
 117
 118         // fixme: this check should be implemented with a distributed
 119         // algorithm, and not a shared global variable!
 120         // fixme: should this chunk be moved before call to receive() ?
 121         if (100.0 * total_load_running / total_load_init <=
 122             opt::load_ratio_threshold) {
 123             VERB0("No more load to balance in system, stopping.");
 124             break;
 125         }
 126
 127     }
 128     VERB0("Going to finalize...");
 129     finalize();
 130
 131     /* Open Questions :
 132      * - definition of load on heterogeneous hosts ?
 133      * - how to detect convergence ?
 134      * - how to manage link failures ?
 135      */
 136
 137     VERB0("Done.");
 138     INFO3("Final load after %d iteration%s: %g",
 139           comp_iter, ESSE(comp_iter), load);
 140     if (opt::bookkeeping)
 141         INFO1("Expected load: %g", expected_load);
 142     return 0;
 143 }
 144
 145 double process::sum_of_to_send() const
 146 {
 147     using std::tr1::bind;
 148     using std::tr1::placeholders::_1;
 149     using std::tr1::placeholders::_2;
 150
 151     return std::accumulate(neigh.begin(), neigh.end(), 0.0,
 152                            bind(std::plus<double>(),
 153                                 _1, bind(&neighbor::get_to_send, _2)));
 154 }
 155
 156 double process::load_balance(double /*my_load*/)
 157 {
 158     if (lb_iter == 1)           // warn only once
 159         WARN0("process::load_balance is a no-op!");
 160     return 0.0;
 161 }
 162
 163 void process::compute()
 164 {
 165     if (load > 0.0) {
 166         double duration = opt::comp_cost(load);
 167         m_task_t task = MSG_task_create("computation", duration, 0.0, NULL);
 168         DEBUG2("compute %g flop%s", duration, ESSE(duration));
 169         MSG_task_execute(task);
 170         MSG_task_destroy(task);
 171     } else {
 172         DEBUG0("nothing to compute !");
 173     }
 174 }
 175
 176 void process::send1_no_bookkeeping(neighbor& nb)
 177 {
 178     if (load != prev_load_broadcast)
 179         comm.send(nb.get_ctrl_mbox(), new message(message::INFO, load));
 180     double load_to_send = nb.get_to_send();
 181     if (load_to_send > 0.0) {
 182         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 183         nb.set_to_send(0.0);
 184     }
 185 }
 186
 187 void process::send1_bookkeeping(neighbor& nb)
 188 {
 189     if (expected_load != prev_load_broadcast)
 190         comm.send(nb.get_ctrl_mbox(),
 191                   new message(message::INFO, expected_load));
 192     double load_to_send;
 193     double new_debt;
 194     double debt_to_send = nb.get_to_send();
 195     if (debt_to_send > 0.0) {
 196         comm.send(nb.get_ctrl_mbox(),
 197                   new message(message::CREDIT, debt_to_send));
 198         nb.set_to_send(0.0);
 199         new_debt = nb.get_debt() + debt_to_send;
 200     } else {
 201         new_debt = nb.get_debt();
 202     }
 203     if (load <= new_debt) {
 204         load_to_send = load;
 205         nb.set_debt(new_debt - load_to_send);
 206         load = 0.0;
 207     } else {
 208         load_to_send = new_debt;
 209         nb.set_debt(0.0);
 210         load -= load_to_send;
 211     }
 212     if (load_to_send > 0.0)
 213         comm.send(nb.get_data_mbox(), new message(message::LOAD, load_to_send));
 214 }
 215
 216 void process::send()
 217 {
 218     using std::tr1::bind;
 219     using std::tr1::placeholders::_1;
 220
 221     if (opt::bookkeeping) {
 222         std::for_each(neigh.begin(), neigh.end(),
 223                       bind(&process::send1_bookkeeping, this, _1));
 224         prev_load_broadcast = expected_load;
 225     } else {
 226         std::for_each(neigh.begin(), neigh.end(),
 227                       bind(&process::send1_no_bookkeeping, this, _1));
 228         prev_load_broadcast = load;
 229     }
 230     comm.flush(false);
 231 }
 232
 233 void process::receive(bool wait)
 234 {
 235     message* msg;
 236     m_host_t from;
 237
 238     while (may_receive() && comm.recv(msg, from, wait)) {
 239         switch (msg->get_type()) {
 240         case message::INFO: {
 241             neighbor* n = rev_neigh[from];
 242             n->set_load(msg->get_amount());
 243             break;
 244         }
 245         case message::CREDIT:
 246             expected_load += msg->get_amount();
 247             break;
 248         case message::LOAD: {
 249             double ld = msg->get_amount();
 250             load += ld;
 251             if (finalizing)
 252                 total_load_running -= ld;
 253             break;
 254         }
 255         case message::CTRL_CLOSE:
 256             ctrl_close_pending--;
 257             close_received = true;
 258             break;
 259         case message::DATA_CLOSE:
 260             data_close_pending--;
 261             close_received = true;
 262             break;
 263         }
 264         delete msg;
 265         wait = false;           // only wait on first recv
 266     }
 267     comm.flush(false);
 268 }
 269
 270 void process::finalize1(neighbor& nb)
 271 {
 272     comm.send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 273     comm.send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 274 }
 275
 276 void process::finalize()
 277 {
 278     using std::tr1::bind;
 279     using std::tr1::placeholders::_1;
 280
 281     finalizing = true;
 282     total_load_running -= load;
 283
 284     DEBUG2("send CLOSE to %lu neighbor%s",
 285            (unsigned long )neigh.size(), ESSE(neigh.size()));
 286     std::for_each(neigh.begin(), neigh.end(),
 287                   bind(&process::finalize1, this, _1));
 288
 289     DEBUG2("wait for CLOSE from %lu neighbor%s",
 290            (unsigned long )neigh.size(), ESSE(neigh.size()));
 291     while (may_receive()) {
 292         comm.flush(false);
 293         receive(true);
 294     }
 295
 296     comm.flush(true);
 297 }
 298
 299 #define print_loads_generic(vec, verbose, logp, cat)                    \
 300     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 301         using std::tr1::bind;                                           \
 302         using std::tr1::placeholders::_1;                               \
 303         XCLOG0(cat, logp, "Neighbor loads:");                           \
 304         std::for_each(vec.begin(), vec.end(),                           \
 305                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 306     } else ((void)0)
 307
 308 void process::print_loads(bool verbose,
 309                           e_xbt_log_priority_t logp,
 310                           xbt_log_category_t cat) const
 311 {
 312     print_loads_generic(neigh, verbose, logp, cat);
 313 }
 314
 315 void process::print_loads_p(bool verbose,
 316                             e_xbt_log_priority_t logp,
 317                             xbt_log_category_t cat) const
 318 {
 319     print_loads_generic(pneigh, verbose, logp, cat);
 320 }
 321
 322 #undef print_loads_generic
 323
 324 // Local variables:
 325 // mode: c++
 326 // End: