X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/56882f01c7244f330aa344f6fc92a3c6abf5345b..05c1f36a7de4ea6a2f757b0e48de8231d3a49c30:/src/surf/surf.c diff --git a/src/surf/surf.c b/src/surf/surf.c index 410352937f..fbd51c2015 100644 --- a/src/surf/surf.c +++ b/src/surf/surf.c @@ -114,6 +114,8 @@ xbt_dynar_t model_list = NULL; tmgr_history_t history = NULL; lmm_system_t maxmin_system = NULL; xbt_dynar_t surf_path = NULL; +xbt_dynar_t host_that_restart = NULL; +xbt_dict_t watched_hosts_lib; /* Don't forget to update the option description in smx_config when you change this */ s_surf_model_description_t surf_network_model_description[] = { @@ -204,7 +206,6 @@ s_surf_model_description_t surf_new_model_description[] = { static xbt_parmap_t surf_parmap = NULL; /* parallel map on models */ #endif -static int surf_nthreads = 1; /* number of threads of the parmap (1 means no parallelism) */ static double *surf_mins = NULL; /* return value of share_resources for each model */ static int surf_min_index; /* current index in surf_mins */ static double min; /* duration determined by surf_solve */ @@ -487,7 +488,7 @@ void surf_exit(void) xbt_free(surf_mins); surf_mins = NULL; #endif - + if(host_that_restart)xbt_dynar_free(&host_that_restart); xbt_dynar_free(&surf_path); xbt_lib_free(&host_lib); @@ -545,6 +546,9 @@ double surf_solve(double max_date) tmgr_trace_event_t event = NULL; unsigned int iter; + if(!host_that_restart) + host_that_restart = xbt_dynar_new(sizeof(char*), NULL); + if (max_date != -1.0 && max_date != NOW) { min = max_date - NOW; } @@ -556,20 +560,9 @@ 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; @@ -617,7 +610,9 @@ double surf_solve(double max_date) tmgr_history_get_next_event_leq(history, next_event_date, &value, (void **) &resource))) { - if (resource->model->model_private->resource_used(resource)) { + if (resource->model->model_private->resource_used(resource) || + xbt_dict_get_or_null(watched_hosts_lib,resource->name) + ) { min = next_event_date - NOW; XBT_DEBUG ("This event will modify model state. Next event set to %f", @@ -626,7 +621,8 @@ double surf_solve(double max_date) /* update state of model_obj according to new value. Does not touch lmm. It will be modified if needed when updating actions */ XBT_DEBUG("Calling update_resource_state for resource %s with min %lf", - resource->model->name, min); + resource->name, min); + resource->model->model_private->update_resource_state(resource, event, value, next_event_date); @@ -645,17 +641,9 @@ 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 @@ -688,95 +676,3 @@ 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. - * - * A value of 1 means no parallelism. - * - * \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; -} - -/* 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. - * - * 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. - * - */ - -static void remove_watched_host(void *key) -{ - xbt_dict_remove(watched_hosts_lib, *(char**)key); -} - -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) - { - 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_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); -}