X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/9003b522ca66f1a3723c63ce1c4753ad7aea9b70..61a993d9efe970edcbb1cb3d947553b2f188b327:/src/surf/surf.c diff --git a/src/surf/surf.c b/src/surf/surf.c index 67cd8a4b34..0da0209825 100644 --- a/src/surf/surf.c +++ b/src/surf/surf.c @@ -10,6 +10,7 @@ #include "simix/smx_host_private.h" #include "surf/surf_resource.h" #include "xbt/xbt_os_thread.h" +#include "simgrid/sg_config.h" #include @@ -391,8 +392,6 @@ void sg_version(int *ver_major,int *ver_minor,int *ver_patch) { *ver_patch = SIMGRID_VERSION_PATCH; } -xbt_dynar_t sg_cmdline = NULL; - void surf_init(int *argc, char **argv) { XBT_DEBUG("Create all Libs"); @@ -413,20 +412,21 @@ void surf_init(int *argc, char **argv) SURF_WKS_LEVEL = xbt_lib_add_level(host_lib,surf_resource_free); SURF_LINK_LEVEL = xbt_lib_add_level(link_lib,surf_resource_free); - sg_cmdline = xbt_dynar_new(sizeof(char*),NULL); - int i; - for (i=0;i<*argc;i++) { - xbt_dynar_push(sg_cmdline,&(argv[i])); - } xbt_init(argc, argv); if (!model_list) model_list = xbt_dynar_new(sizeof(surf_model_private_t), NULL); if (!history) history = tmgr_history_new(); - surf_config_init(argc, argv); +#ifdef HAVE_TRACING + TRACE_add_start_function(TRACE_surf_alloc); + TRACE_add_end_function(TRACE_surf_release); +#endif + + sg_config_init(argc, argv); + surf_action_init(); - if (MC_IS_ENABLED) + if (MC_is_active()) MC_memory_init(); } @@ -465,7 +465,7 @@ void surf_exit(void) unsigned int iter; surf_model_t model = NULL; - surf_config_finalize(); + sg_config_finalize(); xbt_dynar_foreach(model_list, iter, model) model->model_private->finalize(); @@ -489,7 +489,6 @@ void surf_exit(void) #endif xbt_dynar_free(&surf_path); - xbt_dynar_free(&sg_cmdline); xbt_lib_free(&host_lib); xbt_lib_free(&link_lib); @@ -524,9 +523,11 @@ void surf_presolve(void) tmgr_history_get_next_event_leq(history, next_event_date, &value, (void **) &resource))) { - resource->model->model_private->update_resource_state(resource, - event, value, - NOW); + if (value >= 0){ + resource->model->model_private->update_resource_state(resource, + event, value, + NOW); + } } } xbt_dynar_foreach(model_list, iter, model) @@ -555,20 +556,9 @@ double surf_solve(double max_date) } surf_min_index = 0; - - if (surf_get_nthreads() > 1) { - /* parallel version */ -#ifdef CONTEXT_THREADS - xbt_parmap_apply(surf_parmap, (void_f_pvoid_t) surf_share_resources, model_list); -#else - xbt_die("Asked to run in parallel, but no thread at hand..."); -#endif - } - else { - /* sequential version */ - xbt_dynar_foreach(model_list, iter, model) { - surf_share_resources(model); - } + /* sequential version */ + xbt_dynar_foreach(model_list, iter, model) { + surf_share_resources(model); } unsigned i; @@ -595,23 +585,23 @@ double surf_solve(double max_date) min = MAX(next_event_date - NOW, min); } - XBT_DEBUG("Run for NS3 at most %f", min); + XBT_DEBUG("Run for network at most %f", min); // run until min or next flow model_next_action_end = surf_network_model->model_private->share_resources(min); - XBT_DEBUG("Min for NS3 : %f", model_next_action_end); + XBT_DEBUG("Min for network : %f", model_next_action_end); if(model_next_action_end>=0.0) min = model_next_action_end; } - if (next_event_date == -1.0) { + if (next_event_date < 0.0) { XBT_DEBUG("no next TRACE event. Stop searching for it"); break; } - if ((min != -1.0) && (next_event_date > NOW + min)) break; + if ((min == -1.0) || (next_event_date > NOW + min)) break; - XBT_DEBUG("Updating models"); + XBT_DEBUG("Updating models (min = %g, NOW = %g, next_event_date = %g)",min, NOW, next_event_date); while ((event = tmgr_history_get_next_event_leq(history, next_event_date, &value, @@ -628,7 +618,7 @@ double surf_solve(double max_date) resource->model->name, min); resource->model->model_private->update_resource_state(resource, event, value, - NOW + min); + next_event_date); } } while (1); @@ -644,17 +634,9 @@ double surf_solve(double max_date) NOW = NOW + min; - if (surf_get_nthreads() > 1) { - /* parallel version */ -#ifdef CONTEXT_THREADS - xbt_parmap_apply(surf_parmap, (void_f_pvoid_t) surf_update_actions_state, model_list); -#endif - } - else { - /* sequential version */ - xbt_dynar_foreach(model_list, iter, model) { - surf_update_actions_state(model); - } + /* sequential version */ + xbt_dynar_foreach(model_list, iter, model) { + surf_update_actions_state(model); } #ifdef HAVE_TRACING @@ -687,42 +669,38 @@ static void surf_update_actions_state(surf_model_t model) model->model_private->update_actions_state(NOW, min); } -/** - * \brief Returns the number of parallel threads used to update the models. - * \return the number of threads (1 means no parallelism) - */ -int surf_get_nthreads(void) { - return surf_nthreads; -} -/** - * \brief Sets the number of parallel threads used to update the models. +/* This function is a pimple that we ought to fix. But it won't be easy. + * + * The surf_solve() function does properly return the set of actions that changed. + * Instead, each model change a global data, and then the caller of surf_solve must + * pick into these sets of action_failed and action_done. + * + * This was not clean but ok as long as we didn't had to restart the processes when the resource comes back up. + * We worked by putting sentinel actions on every resources we are interested in, + * so that surf informs us if/when the corresponding resource fails. * - * A value of 1 means no parallelism. + * But this does not work to get Simix informed of when a resource comes back up, and this is where this pimple comes. + * We have a set of resources that are currently down and for which simix needs to know when it comes back up. + * And the current function is called *at every simulation step* to sweep over that set, searching for a resource + * that was turned back up in the meanwhile. This is UGLY and slow. + * + * The proper solution would be to not rely on globals for the action_failed and action_done swags. + * They must be passed as parameter by the caller (the handling of these actions in simix may let you + * think that these two sets can be merged, but their handling in SimDag induce the contrary unless this + * simdag code can check by itself whether the action is done of failed -- seems very doable, but yet more + * cleanup to do). + * + * Once surf_solve() is passed the set of actions that changed, you want to add a new set of resources back up + * as parameter to this function. You also want to add a boolean field "restart_watched" to each resource, and + * make sure that whenever a resource with this field enabled comes back up, it's added to that set so that Simix + * sees it and react accordingly. This would kill that need for surf to call simix. * - * \param nb_threads the number of threads to use */ -void surf_set_nthreads(int nthreads) { - if (nthreads<=0) { - nthreads = xbt_os_get_numcores(); - XBT_INFO("Auto-setting surf/nthreads to %d",nthreads); - } - -#ifdef CONTEXT_THREADS - xbt_parmap_destroy(surf_parmap); - surf_parmap = NULL; -#endif - - if (nthreads > 1) { -#ifdef CONTEXT_THREADS - surf_parmap = xbt_parmap_new(nthreads, XBT_PARMAP_DEFAULT); -#else - THROWF(arg_error, 0, "Cannot activate parallel threads in Surf: your architecture does not support threads"); -#endif - } - - surf_nthreads = nthreads; +static void remove_watched_host(void *key) +{ + xbt_dict_remove(watched_hosts_lib, *(char**)key); } void surf_watched_hosts(void) @@ -730,6 +708,7 @@ void surf_watched_hosts(void) char *key; void *host; xbt_dict_cursor_t cursor; + xbt_dynar_t hosts = xbt_dynar_new(sizeof(char*), NULL); XBT_DEBUG("Check for host SURF_RESOURCE_ON on watched_hosts_lib"); xbt_dict_foreach(watched_hosts_lib,cursor,key,host) @@ -737,9 +716,11 @@ void surf_watched_hosts(void) if(SIMIX_host_get_state(host) == SURF_RESOURCE_ON){ XBT_INFO("Restart processes on host: %s",SIMIX_host_get_name(host)); SIMIX_host_autorestart(host); - xbt_dict_remove(watched_hosts_lib,key); + xbt_dynar_push_as(hosts, char*, key); } else XBT_DEBUG("See SURF_RESOURCE_OFF on host: %s",key); } + xbt_dynar_map(hosts, remove_watched_host); + xbt_dynar_free(&hosts); }