X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/1b6d36b7082f7cc657eb2473d1ac27d250954688..4e78565ea6b354a0e6250a87b483f909665a0ac3:/src/surf/surf.c diff --git a/src/surf/surf.c b/src/surf/surf.c index 5fb7317eaa..d2ded06483 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 @@ -109,7 +110,11 @@ int __surf_is_absolute_file_path(const char *file_path) double NOW = 0; -xbt_dynar_t model_list = NULL; +/* model_list_invoke contains only surf_workstation and surf_vm_workstation. + * The callback functions of cpu_model and network_model will be called from + * those of these workstation models. */ +xbt_dynar_t model_list = NULL; /* for destroying all models correctly */ +xbt_dynar_t model_list_invoke = NULL; /* for invoking callbacks */ tmgr_history_t history = NULL; lmm_system_t maxmin_system = NULL; xbt_dynar_t surf_path = NULL; @@ -189,6 +194,16 @@ s_surf_model_description_t surf_storage_model_description[] = { {NULL, NULL, NULL} /* this array must be NULL terminated */ }; +/* ********************************************************************* */ +/* TUTORIAL: New model */ +s_surf_model_description_t surf_new_model_description[] = { + {"default", + "Tutorial model.", + surf_new_model_init_default}, + {NULL, NULL, NULL} /* this array must be NULL terminated */ +}; +/* ********************************************************************* */ + #ifdef CONTEXT_THREADS static xbt_parmap_t surf_parmap = NULL; /* parallel map on models */ #endif @@ -404,12 +419,20 @@ void surf_init(int *argc, char **argv) xbt_init(argc, argv); if (!model_list) model_list = xbt_dynar_new(sizeof(surf_model_private_t), NULL); + if (!model_list_invoke) + model_list_invoke = 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(); } @@ -448,11 +471,14 @@ 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(); + model->model_private->finalize(model); xbt_dynar_free(&model_list); + + xbt_dynar_free(&model_list_invoke); + routing_exit(); if (maxmin_system) { @@ -506,13 +532,17 @@ 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); + } } } + + /* FIXME: see what is check_update_action_state(). if necessary, use model_list_invoke. */ xbt_dynar_foreach(model_list, iter, model) - model->model_private->update_actions_state(NOW, 0.0); + model->model_private->update_actions_state(model, NOW, 0.0); } double surf_solve(double max_date) @@ -533,7 +563,7 @@ double surf_solve(double max_date) XBT_DEBUG("Looking for next action end for all models except NS3"); if (surf_mins == NULL) { - surf_mins = xbt_new(double, xbt_dynar_length(model_list)); + surf_mins = xbt_new(double, xbt_dynar_length(model_list_invoke)); } surf_min_index = 0; @@ -541,20 +571,20 @@ double surf_solve(double max_date) if (surf_get_nthreads() > 1) { /* parallel version */ #ifdef CONTEXT_THREADS - xbt_parmap_apply(surf_parmap, (void_f_pvoid_t) surf_share_resources, model_list); + xbt_parmap_apply(surf_parmap, (void_f_pvoid_t) surf_share_resources, model_list_invoke); #else xbt_die("Asked to run in parallel, but no thread at hand..."); #endif } else { /* sequential version */ - xbt_dynar_foreach(model_list, iter, model) { + xbt_dynar_foreach(model_list_invoke, iter, model) { surf_share_resources(model); } } unsigned i; - for (i = 0; i < xbt_dynar_length(model_list); i++) { + for (i = 0; i < xbt_dynar_length(model_list_invoke); i++) { if ((min < 0.0 || surf_mins[i] < min) && surf_mins[i] >= 0.0) { min = surf_mins[i]; @@ -577,23 +607,23 @@ 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); + model_next_action_end = surf_network_model->model_private->share_resources(surf_network_model, 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, @@ -610,7 +640,7 @@ 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); @@ -633,6 +663,7 @@ double surf_solve(double max_date) #endif } else { + /* FIXME: model_list or model_list_invoke? revisit here later */ /* sequential version */ xbt_dynar_foreach(model_list, iter, model) { surf_update_actions_state(model); @@ -657,7 +688,7 @@ static void surf_share_resources(surf_model_t model) int i = __sync_fetch_and_add(&surf_min_index, 1); if (strcmp(model->name,"network NS3")) { XBT_DEBUG("Running for Resource [%s]", model->name); - next_action_end = model->model_private->share_resources(NOW); + next_action_end = model->model_private->share_resources(model, NOW); XBT_DEBUG("Resource [%s] : next action end = %f", model->name, next_action_end); } @@ -666,7 +697,7 @@ static void surf_share_resources(surf_model_t model) static void surf_update_actions_state(surf_model_t model) { - model->model_private->update_actions_state(NOW, min); + model->model_private->update_actions_state(model, NOW, min); } /** @@ -707,11 +738,45 @@ void surf_set_nthreads(int nthreads) { 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) @@ -719,9 +784,11 @@ 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); }