X-Git-Url: https://bilbo.iut-bm.univ-fcomte.fr/and/gitweb/loba.git/blobdiff_plain/46ef54f8ce05841e476c3cd9d425a9c680a84882..3634a92de5281a6b19ca03b49b21dad853f386b3:/main.cpp?ds=inline diff --git a/main.cpp b/main.cpp index 3ee3ae1..e148574 100644 --- a/main.cpp +++ b/main.cpp @@ -1,15 +1,12 @@ -#include +#include +#include +#include +#include // strchr #include #include +#include #include #include -#include "deployment.h" -#include "hostdata.h" -#include "misc.h" -#include "options.h" -#include "process.h" -#include "timer.h" -#include "version.h" // Creates log categories XBT_LOG_NEW_CATEGORY(simu, "Root of simulation messages"); @@ -18,61 +15,197 @@ XBT_LOG_NEW_SUBCATEGORY(depl, main, "Messages from auto deployment"); XBT_LOG_NEW_SUBCATEGORY(comm, simu, "Messages from asynchronous pipes"); XBT_LOG_NEW_SUBCATEGORY(proc, simu, "Messages from base process class"); XBT_LOG_NEW_SUBCATEGORY(loba, simu, "Messages from load-balancer"); +XBT_LOG_NEW_SUBCATEGORY(thrd, simu, "Messages from thread wrapper class"); XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(main); -// Failure exit status -enum { - EXIT_NO_FAILURE = 0x00, // no error - EXIT_FAILURE_ARGS = 0x01, // bad arguments - EXIT_FAILURE_INIT = 0x02, // failed to initialize simulator - EXIT_FAILURE_SIMU = 0x04, // simulation failed - EXIT_FAILURE_CLEAN = 0x08, // error at cleanup -}; +#include "deployment.h" +#include "hostdata.h" +#include "misc.h" +#include "options.h" +#include "process.h" +#include "simgrid_features.h" +#include "statistics.h" +#include "synchro.h" +#include "timer.h" +#include "tracing.h" +#include "version.h" + +#define DATA_DESCR_WIDTH 39 + +namespace { + // Failure exit status + enum { + EXIT_NO_FAILURE = 0x00, // no error + EXIT_FAILURE_ARGS = 0x01, // bad arguments + EXIT_FAILURE_INIT = 0x02, // failed to initialize simulator + EXIT_FAILURE_SIMU = 0x04, // simulation failed + EXIT_FAILURE_CLEAN = 0x08, // error at cleanup + EXIT_FAILURE_INTR = 0x10, // interrupted by user + EXIT_FAILURE_LOAD = 0x20, // lost load on exit + EXIT_FAILURE_OTHER = 0x40, // other error + }; + + // Cannot be globally initialized... + mutex_t* proc_mutex; + condition_t* proc_cond; + unsigned proc_counter = 0; + + statistics loads; + statistics comps; + statistics comp_iterations; + statistics all_comp_iterations; + statistics iter_deviation; + statistics data_send_amount; + statistics data_recv_amount; + statistics data_send_count; + statistics data_recv_count; + statistics ctrl_send_amount; + statistics ctrl_recv_amount; + statistics ctrl_send_count; + statistics ctrl_recv_count; + statistics idle_duration; + statistics convergence; -int simulation_main(int argc, char* argv[]) +} + +static int simulation_main(int argc, char* argv[]) { int result; process* proc; - try { - proc = opt::loba_algorithms.new_instance(opt::loba_algo, argc, argv); - result = proc->run(); - delete proc; - } - catch (std::invalid_argument& e) { - THROW1(arg_error, 0, "%s", e.what()); - } + + proc = opt::loba_algorithms.new_instance(opt::loba_algo, argc, argv); + + proc_mutex->acquire(); + ++proc_counter; + proc_mutex->release(); + + result = proc->run(); + + proc_mutex->acquire(); + loads.push(proc->get_real_load()); + comps.push(proc->get_comp_amount()); + comp_iterations.push(proc->get_comp_iter()); + all_comp_iterations.push(proc->get_all_comp_iter()); + iter_deviation.push(proc->get_iter_deviation()); + data_send_amount.push(proc->get_data_send_amount()); + data_recv_amount.push(proc->get_data_recv_amount()); + data_send_count.push(proc->get_data_send_count()); + data_recv_count.push(proc->get_data_recv_count()); + ctrl_send_amount.push(proc->get_ctrl_send_amount()); + ctrl_recv_amount.push(proc->get_ctrl_recv_amount()); + ctrl_send_count.push(proc->get_ctrl_send_count()); + ctrl_recv_count.push(proc->get_ctrl_recv_count()); + idle_duration.push(proc->get_idle_duration()); + double c = proc->get_convergence(); + if (c != -1.0) + convergence.push(c); + + // Synchronization barrier... + // The goal is to circumvent a limitation in SimGrid (at least + // in version 3.5): a process must be alive when another one + // destroys a communication they had together. + + --proc_counter; + proc_cond->broadcast(); + while (proc_counter > 0) + proc_cond->wait(*proc_mutex); + proc_mutex->release(); + + delete proc; + return result; } -void check_for_lost_load() +static bool check_for_lost_load() { - const double threshold = 1e-4; + bool res = true; double total_init = process::get_total_load_init(); double total_exit = process::get_total_load_exit(); double lost = total_init - total_exit; - double lost_ratio = 100 * lost / total_init; - if (lost_ratio < -threshold) { - CRITICAL2("Gained load at exit! %g (%g%%) <============", - lost, lost_ratio); - } else if (lost_ratio > threshold) { - CRITICAL2("Lost load at exit! %g (%g%%) <============", + double lost_ratio = 100.0 * lost / total_init; + if (lost_ratio < -opt::load_ratio_threshold) { + XBT_ERROR("Gained load at exit! %g (%g%%) <============", + -lost, -lost_ratio); + res = false; + } else if (lost_ratio > opt::load_ratio_threshold) { + XBT_ERROR("Lost load at exit! %g (%g%%) <============", lost, lost_ratio); + res = false; + } else + XBT_VERB("Total load at exit looks good: %g (%g%%)", lost, lost_ratio); + + double total_running = process::get_total_load_running(); + double running_ratio = 100.0 * total_running / total_init; + if (running_ratio < -opt::load_ratio_threshold) { + XBT_ERROR("Negative running load at exit! %g (%g%%) <============", + total_running, running_ratio); + res = false; + } else if (running_ratio > opt::load_ratio_threshold) { + XBT_ERROR("Remaining running load at exit! %g (%g%%) <============", + total_running, running_ratio); + res = false; + } else + XBT_VERB("Running load at exit looks good: %g (%g%%)", + total_running, running_ratio); + return res; +} + +static void check_file_access(const std::string& name) +{ + if (access(name.c_str(), R_OK) != 0) { + std::cerr << "ERROR: cannot access to file \"" + << name << "\" for reading: " << strerror(errno) << "\n"; + exit(EXIT_FAILURE_ARGS); + } +} + +static void signal_handler(int /*sig*/) +{ + if (!opt::exit_request) { + XBT_CRITICAL(">>>>>>>>>>" + " caught CTRL-C: global exit requested " + "<<<<<<<<<<"); + opt::exit_request = 1; } else { - DEBUG2("Total load at exit looks good: %g (%g%%)", lost, lost_ratio); + XBT_CRITICAL(">>>>>>>>>>" + " caught CTRL-C for the 2nd time: exit immediately " + "<<<<<<<<<<"); + exit(EXIT_FAILURE_INTR); } +} +static void install_signal_handler() +{ + struct sigaction action; + action.sa_handler = signal_handler; + sigemptyset(&action.sa_mask); + action.sa_flags = SA_RESTART; + if (sigaction(SIGINT, &action, NULL) == -1) { + std::cerr << "ERROR: sigaction: " << strerror(errno) << "\n"; + exit(EXIT_FAILURE_OTHER); + } } +#define PR_VALUE(descr, format, ...) \ + XBT_INFO("| %.*s: " format, DATA_DESCR_WIDTH, \ + descr " ................................................", \ + __VA_ARGS__) + +#define PR_STATS(descr, st) \ + XBT_INFO("| %.*s: %g / %g / %g", DATA_DESCR_WIDTH, \ + descr " (sum/avg/dev) ..................................", \ + st.get_sum(), st.get_mean(), st.get_stddev()) + int main(int argc, char* argv[]) { - // Note: variables used after THROW must be declared as volatile. - volatile int exit_status = 0; // global exit status - volatile double simulated_time = -1.0; - timestamp simulation_time; - xbt_ex_t ex; - MSG_error_t res; + int exit_status = 0; // global exit status + double simulated_time = -1.0; + timestamp elapsed_time(timestamp::wallclock_time); + timestamp simulation_time(timestamp::cpu_time); + msg_error_t res; + elapsed_time.start(); simulation_time.start(); // Set default logging parameters @@ -87,12 +220,11 @@ int main(int argc, char* argv[]) } // Initialize some MSG internal data. - // Note: MSG_global_init() may throw an exception, but it seems - // impossible to catch it correctly :-( - MSG_global_init(&argc, argv); + MSG_init(&argc, argv); + install_signal_handler(); // Parse global parameters - int parse_res = opt::parse_args(&argc, argv); + bool parse_res = opt::parse_args(&argc, argv); if (!parse_res || opt::version_requested || opt::help_requested) { if (opt::version_requested) @@ -102,80 +234,134 @@ int main(int argc, char* argv[]) "Compiled on " << version::date << "\n\n"; if (!parse_res || opt::help_requested) opt::usage(); - MSG_clean(); + res = MSG_clean(); exit(parse_res ? EXIT_NO_FAILURE : EXIT_FAILURE_ARGS); } - INFO3("%s v%s (%s)", opt::program_name.c_str(), version::num.c_str(), + XBT_INFO("%s v%s (%s)", opt::program_name.c_str(), version::num.c_str(), version::date.c_str()); opt::print(); - TRY { - exit_status = EXIT_FAILURE_INIT; // ===== - - // Register the default function of an agent - // MSG_function_register("simulation_main", simulation_main); - MSG_function_register_default(simulation_main); - - // Create the platform and the application. - MSG_create_environment(opt::platform_file.c_str()); - hostdata::create(); - if (opt::auto_depl::enabled) { - if (!opt::auto_depl::nhosts) - opt::auto_depl::nhosts = hostdata::size(); - if (opt::auto_depl::nhosts > hostdata::size()) { - WARN2("%u hosts is too much: limiting to %u", - opt::auto_depl::nhosts, (unsigned )hostdata::size()); - opt::auto_depl::nhosts = hostdata::size(); - } - if (!opt::auto_depl::load) - opt::auto_depl::load = opt::auto_depl::nhosts; - MY_launch_application(); // it is already opt::* aware... - } else { - MSG_launch_application(opt::deployment_file.c_str()); + // Register the default function of an agent + // MSG_function_register("simulation_main", simulation_main); + MSG_function_register_default(simulation_main); + + // Create the platform and the application. + XBT_DEBUG("Loading platform file..."); + check_file_access(opt::platform_file); + MSG_create_environment(opt::platform_file.c_str()); + XBT_DEBUG("Creating hostdata..."); + hostdata::create(); + XBT_INFO("Loaded description of %zd hosts.", hostdata::size()); + XBT_DEBUG("Deploying processes..."); + if (opt::auto_depl::enabled) { + if (!opt::auto_depl::nhosts) + opt::auto_depl::nhosts = hostdata::size(); + if (opt::auto_depl::nhosts > hostdata::size()) { + XBT_WARN("%u hosts is too much: limiting to %zu", + opt::auto_depl::nhosts, hostdata::size()); + opt::auto_depl::nhosts = hostdata::size(); + } + if (opt::auto_depl::load == 0.0) { + XBT_WARN("Initial load is zero! " + "Falling back on old behaviour (load = nhosts)."); + opt::auto_depl::load = opt::auto_depl::nhosts; + } else if (opt::auto_depl::load < 0.0) + opt::auto_depl::load = + -opt::auto_depl::load * opt::auto_depl::nhosts; + double iload = std::trunc(opt::auto_depl::load); + if (opt::integer_transfer && opt::auto_depl::load != iload) { + XBT_WARN("Total load %g is not an integer. Truncate it.", + opt::auto_depl::load); + opt::auto_depl::load = iload; } + MY_launch_application(); // it is already opt::* aware... + } else { + check_file_access(opt::deployment_file); + MSG_launch_application(opt::deployment_file.c_str()); + } - exit_status = EXIT_FAILURE_SIMU; // ===== + // Register tracing categories + TRACE_category_with_color(TRACE_CAT_COMP, TRACE_COLOR_COMP); + TRACE_category_with_color(TRACE_CAT_CTRL, TRACE_COLOR_CTRL); + TRACE_category_with_color(TRACE_CAT_DATA, TRACE_COLOR_DATA); - // Launch the MSG simulation. - INFO1("Starting simulation at %f...", MSG_get_clock()); - res = MSG_main(); - simulated_time = MSG_get_clock(); - INFO1("Simulation ended at %f.", simulated_time); - check_for_lost_load(); - if (res != MSG_OK) - THROW1(0, 0, "MSG_main() failed with status %#x", res); + proc_mutex = new mutex_t(); + proc_cond = new condition_t(); + process::set_proc_mutex(proc_mutex); - exit_status = EXIT_NO_FAILURE; // ===== - } - CATCH (ex) { - int len = strlen(ex.msg); - if (len > 0 && ex.msg[len - 1] == '\n') - ex.msg[len - 1] = '\0'; // strip the ending '\n' - ERROR1("%s", ex.msg); - DEBUG3("Error from %s() in %s:%d", ex.func, ex.file, ex.line); - xbt_ex_free(ex); + // Launch the MSG simulation. + XBT_INFO("Starting simulation at %f...", MSG_get_clock()); + res = MSG_main(); + simulated_time = MSG_get_clock(); + XBT_INFO("Simulation ended at %f.", simulated_time); + + process::set_proc_mutex(NULL); + delete proc_cond; + delete proc_mutex; + + if (res == MSG_OK) { + exit_status = EXIT_NO_FAILURE; + } else { + XBT_ERROR("MSG_main() failed with status %#x", res); + exit_status = EXIT_FAILURE_SIMU; } // Clean the MSG simulation. hostdata::destroy(); res = MSG_clean(); if (res != MSG_OK) { - ERROR1("MSG_clean() failed with status %#x", res); + XBT_ERROR("MSG_clean() failed with status %#x", res); exit_status |= EXIT_FAILURE_CLEAN; } // Report final simulation status. if (simulated_time >= 0.0) { simulation_time.stop(); - INFO0(",----[ Results ]"); - INFO1("| Total simulated time...: %g", simulated_time); - INFO1("| Total simulation time..: %g", simulation_time.duration()); - INFO0("`----"); + elapsed_time.stop(); + if (!check_for_lost_load()) + exit_status |= EXIT_FAILURE_LOAD; + + XBT_INFO(",----[ Results ]"); + PR_STATS("Load", loads); + PR_STATS("Computation", comps); + PR_STATS("Comp. iterations", comp_iterations); + PR_STATS("X-Comp. iterations", all_comp_iterations); + PR_STATS("Data send amount", data_send_amount); + PR_STATS("Data recv amount", data_recv_amount); + PR_STATS("Data send count", data_send_count); + PR_STATS("Data recv count", data_recv_count); + PR_STATS("Ctrl send amount", ctrl_send_amount); + PR_STATS("Ctrl recv amount", ctrl_recv_amount); + PR_STATS("Ctrl send count", ctrl_send_count); + PR_STATS("Ctrl recv count", ctrl_recv_count); + PR_VALUE("Total simulated time", "%g", simulated_time); + PR_VALUE("Total simulation time", "%g", simulation_time.duration()); + PR_VALUE("Elapsed (wall clock) time", "%g", elapsed_time.duration()); + XBT_INFO("`----"); + + double load_imbalance = 100.0 * loads.get_stddev() / loads.get_mean(); + double transfer_amount = + data_send_amount.get_sum() / opt::comm_cost(loads.get_sum()); + + XBT_INFO(",----[ Useful metrics ]"); + PR_VALUE("Final load imbalance", "%g %s", load_imbalance, + "percent of the load average"); + PR_VALUE("Data transfer amount", "%g %s", transfer_amount, + "times the total amount of data"); + PR_VALUE("Number of hosts that converged", "%u / %u", + convergence.get_count(), loads.get_count()); + PR_VALUE("Times of convergence (min/max/avg/dev)", "%g / %g / %g / %g", + convergence.get_min(), convergence.get_max(), + convergence.get_mean(), convergence.get_stddev()); + PR_STATS("Idle duration", idle_duration); + PR_STATS("Supernumer. comp. iter.", iter_deviation); + XBT_INFO("`----"); + } if (exit_status) - ERROR1("Simulation failed (%#x).", exit_status); + XBT_ERROR("Simulation failed (%#x).", exit_status); else - INFO0("Simulation succeeded."); + XBT_INFO("Simulation succeeded."); return exit_status; }