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 namespace {
  22
  23     void sleep_until_date(double& date, double duration = 0.0)
  24     {
  25         double sleep_duration = date - MSG_get_clock();
  26         if (sleep_duration > 0.0)
  27             MSG_process_sleep(sleep_duration);
  28         date = MSG_get_clock() + duration;
  29     }
  30
  31 }
  32
  33 process::process(int argc, char* argv[])
  34 {
  35     if (argc < 2 || !(std::istringstream(argv[1]) >> real_load))
  36         throw std::invalid_argument("bad or missing initial load parameter");
  37
  38     neigh.assign(argv + 2, argv + argc);
  39
  40     pneigh.reserve(neigh.size());
  41     for (unsigned i = 0 ; i < neigh.size() ; i++) {
  42         neighbor* ptr = &neigh[i];
  43         m_host_t host = MSG_get_host_by_name(ptr->get_name());
  44         pneigh.push_back(ptr);
  45         rev_neigh.insert(std::make_pair(host, ptr));
  46     }
  47
  48     comp = 0.0;
  49
  50     prev_load_broadcast = -1;   // force sending of load on first send_all()
  51     expected_load = real_load;
  52     total_load_running += real_load;
  53     total_load_init += real_load;
  54
  55     ctrl_close_pending = data_close_pending = neigh.size();
  56     close_received = false;
  57     finalizing = false;
  58
  59     comp_iter = lb_iter = 0;
  60
  61     compute_thread = new_msg_thread("compute",
  62                                     std::tr1::bind(&process::compute_loop,
  63                                                    this));
  64
  65     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
  66     if (!LOG_ISENABLED(logp))
  67         return;
  68     std::ostringstream oss;
  69     oss << neigh.size() << " neighbor";
  70     if (!neigh.empty()) {
  71         oss << ESSE(neigh.size()) << ": ";
  72         std::transform(neigh.begin(), neigh.end() - 1,
  73                        std::ostream_iterator<const char*>(oss, ", "),
  74                        std::tr1::mem_fn(&neighbor::get_name));
  75         oss << neigh.back().get_name();
  76     }
  77     XBT_LOG(logp, "Got %s.", oss.str().c_str());
  78     print_loads(false, logp);
  79 }
  80
  81 process::~process()
  82 {
  83     delete compute_thread;
  84     total_load_exit += real_load;
  85     if (opt::bookkeeping) {
  86         XBT_INFO("Final load after %d:%d iterations: %g ; expected: %g",
  87                  lb_iter, comp_iter, real_load, expected_load);
  88     } else {
  89         XBT_INFO("Final load after %d:%d iterations: %g",
  90                  lb_iter, comp_iter, real_load);
  91     }
  92     XBT_VERB("Total computation for this process: %g", comp);
  93 }
  94
  95 int process::run()
  96 {
  97     XBT_INFO("Initial load: %g", real_load);
  98     XBT_VERB("Starting...");
  99     compute_thread->start();
 100     load_balance_loop();
 101     compute_thread->wait();
 102     XBT_VERB("Done.");
 103     return 0;
 104 }
 105
 106 void process::load_balance_loop()
 107 {
 108     using std::tr1::bind;
 109     using std::tr1::placeholders::_1;
 110
 111     double next_iter_after_date = MSG_get_clock() + opt::min_lb_iter_duration;
 112     while (still_running()) {
 113         ++lb_iter;
 114
 115         if (opt::log_rate && lb_iter % opt::log_rate == 0) {
 116             if (opt::bookkeeping)
 117                 XBT_INFO("(%u:%u) current load: %g ; expected: %g",
 118                          lb_iter, comp_iter, real_load, expected_load);
 119             else
 120                 XBT_INFO("(%u:%u) current load: %g",
 121                          lb_iter, comp_iter, real_load);
 122         }
 123
 124         if (get_load() > 0.0)
 125             load_balance();
 126
 127         print_loads(true, xbt_log_priority_debug);
 128
 129         // send
 130         std::for_each(neigh.begin(), neigh.end(),
 131                       bind(&process::ctrl_send, this, _1));
 132
 133         sleep_until_date(next_iter_after_date, opt::min_lb_iter_duration);
 134         ctrl_receive(0.0);
 135
 136         comm.ctrl_flush(false);
 137     }
 138
 139     XBT_VERB("Going to finalize for %s...", __func__);
 140     XBT_DEBUG("send CTRL_CLOSE to %zu neighbor%s",
 141               neigh.size(), ESSE(neigh.size()));
 142     std::for_each(neigh.begin(), neigh.end(),
 143                   bind(&process::ctrl_close, this, _1));
 144     while (ctrl_close_pending) {
 145         comm.ctrl_flush(false);
 146         XBT_DEBUG("waiting for %d CTRL CLOSE", ctrl_close_pending);
 147         ctrl_receive(-1.0);
 148     }
 149     comm.ctrl_flush(true);
 150 }
 151
 152 void process::compute_loop()
 153 {
 154     using std::tr1::bind;
 155     using std::tr1::placeholders::_1;
 156
 157     double next_iter_after_date = MSG_get_clock() + opt::min_comp_iter_duration;
 158     while (still_running()) {
 159         // receive
 160         if (real_load > 0.0)
 161             data_receive(0.0);
 162         else
 163             data_receive(opt::min_comp_iter_duration);
 164
 165         comm.data_flush(false);
 166
 167         if (real_load == 0.0)
 168             continue;
 169
 170         // send
 171         std::for_each(neigh.begin(), neigh.end(),
 172                       bind(&process::data_send, this, _1));
 173
 174         // compute
 175         ++comp_iter;
 176         double flops = opt::comp_cost(real_load);
 177         m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
 178         TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
 179         XBT_DEBUG("compute %g flop%s", flops, ESSE(flops));
 180         MSG_task_execute(task);
 181         comp += flops;
 182         MSG_task_destroy(task);
 183
 184         sleep_until_date(next_iter_after_date, opt::min_comp_iter_duration);
 185     }
 186
 187     XBT_VERB("Going to finalize for %s...", __func__);
 188     XBT_DEBUG("send DATA_CLOSE to %zu neighbor%s",
 189               neigh.size(), ESSE(neigh.size()));
 190     std::for_each(neigh.begin(), neigh.end(),
 191                   bind(&process::data_close, this, _1));
 192     while (data_close_pending) {
 193         comm.data_flush(false);
 194         XBT_DEBUG("waiting for %d DATA CLOSE", data_close_pending);
 195         data_receive(-1.0);
 196     }
 197     comm.data_flush(true);
 198 }
 199
 200 bool process::still_running()
 201 {
 202     static bool last_status = true;
 203
 204     if (!last_status) {
 205         /* nop */
 206
 207     } else if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
 208         XBT_VERB("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
 209         last_status = false;
 210
 211     } else if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
 212         XBT_VERB("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
 213         last_status = false;
 214
 215     } else if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
 216         XBT_VERB("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
 217         last_status = false;
 218
 219     } else if (opt::exit_on_close && close_received) {
 220         XBT_VERB("Close received");
 221         last_status = false;
 222
 223     } else if (real_load == 0.0 && !data_close_pending) {
 224         XBT_VERB("I'm a poor lonesome process, and I have no load...");
 225         last_status = false;
 226
 227     } else if (100.0 * total_load_running / total_load_init <=
 228                opt::load_ratio_threshold) {
 229         // fixme: this check should be implemented with a distributed
 230         // algorithm, and not a shared global variable!
 231         // fixme: should this chunk be moved before call to receive() ?
 232         XBT_VERB("No more load to balance in system.");
 233         last_status = false;
 234     }
 235
 236     return last_status;
 237 }
 238
 239 void process::load_balance()
 240 {
 241     if (lb_iter == 1)           // warn only once
 242         XBT_WARN("process::load_balance() is a no-op!");
 243 }
 244
 245 void process::send(neighbor& nb, double amount)
 246 {
 247     set_load(get_load() - amount);
 248     nb.set_to_send(nb.get_to_send() + amount);
 249     nb.set_load(nb.get_load() + amount);
 250 }
 251
 252 #define print_loads_generic(vec, verbose, logp, cat)                    \
 253     if (_XBT_LOG_ISENABLEDV((*cat), logp)) {                            \
 254         using std::tr1::bind;                                           \
 255         using std::tr1::placeholders::_1;                               \
 256         XBT_XCLOG(cat, logp, "Neighbor loads:");                        \
 257         std::for_each(vec.begin(), vec.end(),                           \
 258                       bind(&neighbor::print, _1, verbose, logp, cat));  \
 259     } else ((void)0)
 260
 261 void process::print_loads(bool verbose,
 262                           e_xbt_log_priority_t logp,
 263                           xbt_log_category_t cat) const
 264 {
 265     print_loads_generic(neigh, verbose, logp, cat);
 266 }
 267
 268 void process::print_loads_p(bool verbose,
 269                             e_xbt_log_priority_t logp,
 270                             xbt_log_category_t cat) const
 271 {
 272     print_loads_generic(pneigh, verbose, logp, cat);
 273 }
 274
 275 #undef print_loads_generic
 276
 277 void process::ctrl_send(neighbor& nb)
 278 {
 279     double info_to_send = get_load();
 280     if (info_to_send != prev_load_broadcast)
 281         comm.ctrl_send(nb.get_ctrl_mbox(),
 282                        new message(message::INFO, info_to_send));
 283     if (opt::bookkeeping) {
 284         double debt_to_send = nb.get_to_send();
 285         if (debt_to_send > 0.0) {
 286             nb.set_to_send(0.0);
 287             nb.set_debt(nb.get_debt() + debt_to_send);
 288             comm.ctrl_send(nb.get_ctrl_mbox(),
 289                            new message(message::CREDIT, debt_to_send));
 290         }
 291     }
 292 }
 293
 294 void process::data_send(neighbor& nb)
 295 {
 296     double load_to_send;
 297     if (opt::bookkeeping) {
 298         if (real_load <= nb.get_debt()) {
 299             load_to_send = real_load;
 300             nb.set_debt(nb.get_debt() - load_to_send);
 301             real_load = 0.0;
 302         } else {
 303             load_to_send = nb.get_debt();
 304             nb.set_debt(0.0);
 305             real_load -= load_to_send;
 306         }
 307     } else {
 308         load_to_send = nb.get_to_send();
 309         nb.set_to_send(0.0);
 310     }
 311     if (load_to_send > 0.0)
 312         comm.data_send(nb.get_data_mbox(),
 313                        new message(message::LOAD, load_to_send));
 314 }
 315
 316 void process::ctrl_close(neighbor& nb)
 317 {
 318     comm.ctrl_send(nb.get_ctrl_mbox(), new message(message::CTRL_CLOSE, 0.0));
 319 }
 320
 321 void process::data_close(neighbor& nb)
 322 {
 323     comm.data_send(nb.get_data_mbox(), new message(message::DATA_CLOSE, 0.0));
 324 }
 325
 326 void process::ctrl_receive(double timeout)
 327 {
 328     message* msg;
 329     m_host_t from;
 330
 331     XBT_DEBUG("%sblocking receive on ctrl (%g)", "\0non-" + !timeout, timeout);
 332     while (ctrl_close_pending && comm.ctrl_recv(msg, from, timeout)) {
 333         handle_message(msg, from);
 334         timeout = 0.0;
 335     }
 336 }
 337
 338 void process::data_receive(double timeout)
 339 {
 340     message* msg;
 341     m_host_t from;
 342
 343     XBT_DEBUG("%sblocking receive on data (%g)", "\0non-" + !timeout, timeout);
 344     while (data_close_pending && comm.data_recv(msg, from, timeout)) {
 345         handle_message(msg, from);
 346         timeout = 0.0;
 347     }
 348 }
 349
 350 void process::handle_message(message* msg, m_host_t from)
 351 {
 352     switch (msg->get_type()) {
 353     case message::INFO: {
 354         neighbor* n = rev_neigh[from];
 355         n->set_load(msg->get_amount());
 356         break;
 357     }
 358     case message::CREDIT:
 359         expected_load += msg->get_amount();
 360         break;
 361     case message::LOAD: {
 362         double ld = msg->get_amount();
 363         real_load += ld;
 364         if (finalizing)
 365             total_load_running -= ld;
 366         break;
 367     }
 368     case message::CTRL_CLOSE:
 369         ctrl_close_pending--;
 370         close_received = true;
 371         break;
 372     case message::DATA_CLOSE:
 373         data_close_pending--;
 374         close_received = true;
 375         break;
 376     }
 377     delete msg;
 378 }
 379
 380 // Local variables:
 381 // mode: c++
 382 // End: