X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/a067481f5e9782cf82ebc6c1487d58083ec2fbd5..9bbf0601aa9af6089fdcfc0fd09bfdf64382705c:/src/surf/surf.c diff --git a/src/surf/surf.c b/src/surf/surf.c index 5618234139..157cbe2639 100644 --- a/src/surf/surf.c +++ b/src/surf/surf.c @@ -4,14 +4,17 @@ /* This program is free software; you can redistribute it and/or modify it * under the terms of the license (GNU LGPL) which comes with this package. */ -#include - #include "surf_private.h" #include "xbt/module.h" #include "mc/mc.h" +#include "simix/smx_host_private.h" #include "surf/surf_resource.h" #include "xbt/xbt_os_thread.h" +#include "simgrid/sg_config.h" + +#include +XBT_LOG_NEW_CATEGORY(surf, "All SURF categories"); XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_kernel, surf, "Logging specific to SURF (kernel)"); @@ -107,7 +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; @@ -134,7 +141,7 @@ s_surf_model_description_t surf_network_model_description[] = { #ifdef HAVE_NS3 {"NS3", "Network pseudo-model using the NS3 tcp model instead of an analytic model", - surf_network_model_init_NS3}, + surf_network_model_init_NS3}, #endif {"Reno", "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).", @@ -187,11 +194,20 @@ 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 -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 */ @@ -288,30 +304,97 @@ double generic_maxmin_share_resources(xbt_swag_t running_actions, return min; } -XBT_LOG_EXTERNAL_CATEGORY(surf_cpu); -XBT_LOG_EXTERNAL_CATEGORY(surf_kernel); -XBT_LOG_EXTERNAL_CATEGORY(surf_lagrange); -XBT_LOG_EXTERNAL_CATEGORY(surf_lagrange_dichotomy); -XBT_LOG_EXTERNAL_CATEGORY(surf_maxmin); -XBT_LOG_EXTERNAL_CATEGORY(surf_network); -XBT_LOG_EXTERNAL_CATEGORY(surf_trace); -XBT_LOG_EXTERNAL_CATEGORY(surf_parse); -XBT_LOG_EXTERNAL_CATEGORY(surf_timer); -XBT_LOG_EXTERNAL_CATEGORY(surf_workstation); -XBT_LOG_EXTERNAL_CATEGORY(surf_config); -XBT_LOG_EXTERNAL_CATEGORY(surf_route); +double generic_share_resources_lazy(double now, surf_model_t model) +{ + surf_action_lmm_t action = NULL; + double min = -1; + double value; -#ifdef HAVE_GTNETS -XBT_LOG_EXTERNAL_CATEGORY(surf_network_gtnets); -#endif + XBT_DEBUG + ("Before share resources, the size of modified actions set is %d", + xbt_swag_size(model->model_private->modified_set)); + + lmm_solve(model->model_private->maxmin_system); + + XBT_DEBUG + ("After share resources, The size of modified actions set is %d", + xbt_swag_size(model->model_private->modified_set)); + + while((action = xbt_swag_extract(model->model_private->modified_set))) { + int max_dur_flag = 0; + if (action->generic_action.state_set != + model->states.running_action_set) + continue; + + /* bogus priority, skip it */ + if (action->generic_action.priority <= 0) + continue; + + generic_update_action_remaining_lazy(action,now); + + min = -1; + value = lmm_variable_getvalue(action->variable); + if (value > 0) { + if (action->generic_action.remains > 0) { + value = action->generic_action.remains / value; + min = now + value; + } else { + value = 0.0; + min = now; + } + } + + if ((action->generic_action.max_duration != NO_MAX_DURATION) + && (min == -1 + || action->generic_action.start + + action->generic_action.max_duration < min)) { + min = action->generic_action.start + + action->generic_action.max_duration; + max_dur_flag = 1; + } + + XBT_DEBUG("Action(%p) Start %lf Finish %lf Max_duration %lf", action, + action->generic_action.start, now + value, + action->generic_action.max_duration); + + if (min != -1) { + surf_action_lmm_heap_remove(model->model_private->action_heap,action); + surf_action_lmm_heap_insert(model->model_private->action_heap,action, min, max_dur_flag ? MAX_DURATION : NORMAL); + XBT_DEBUG("Insert at heap action(%p) min %lf now %lf", action, min, + now); + } else DIE_IMPOSSIBLE; + } + + //hereafter must have already the min value for this resource model + if (xbt_heap_size(model->model_private->action_heap) > 0) + min = xbt_heap_maxkey(model->model_private->action_heap) - now; + else + min = -1; + + XBT_DEBUG("The minimum with the HEAP %lf", min); + + return min; +} static XBT_INLINE void routing_asr_host_free(void *p) { - network_element_t elm = (network_element_t)p; + sg_routing_edge_t elm = p; free(elm->name); xbt_free(elm); } +static XBT_INLINE void routing_asr_prop_free(void *p) +{ + xbt_dict_t elm = p; + xbt_dict_free(&elm); +} + +void sg_version(int *ver_major,int *ver_minor,int *ver_patch) { + *ver_major = SIMGRID_VERSION_MAJOR; + *ver_minor = SIMGRID_VERSION_MINOR; + *ver_patch = SIMGRID_VERSION_PATCH; +} + void surf_init(int *argc, char **argv) { XBT_DEBUG("Create all Libs"); @@ -320,47 +403,35 @@ void surf_init(int *argc, char **argv) as_router_lib = xbt_lib_new(); storage_lib = xbt_lib_new(); storage_type_lib = xbt_lib_new(); + watched_hosts_lib = xbt_dict_new(); - XBT_DEBUG("ADD ROUTING LEVEL"); + XBT_DEBUG("Add routing levels"); ROUTING_HOST_LEVEL = xbt_lib_add_level(host_lib,routing_asr_host_free); ROUTING_ASR_LEVEL = xbt_lib_add_level(as_router_lib,routing_asr_host_free); - ROUTING_STORAGE_LEVEL = xbt_lib_add_level(storage_lib,xbt_free); - ROUTING_STORAGE_HOST_LEVEL = xbt_lib_add_level(storage_lib,routing_storage_host_free); - ROUTING_STORAGE_TYPE_LEVEL = xbt_lib_add_level(storage_type_lib,routing_storage_type_free); + ROUTING_PROP_ASR_LEVEL = xbt_lib_add_level(as_router_lib,routing_asr_prop_free); - XBT_DEBUG("ADD SURF LEVELS"); + XBT_DEBUG("Add SURF levels"); SURF_CPU_LEVEL = xbt_lib_add_level(host_lib,surf_resource_free); SURF_WKS_LEVEL = xbt_lib_add_level(host_lib,surf_resource_free); SURF_LINK_LEVEL = xbt_lib_add_level(link_lib,surf_resource_free); - SURF_STORAGE_LEVEL = xbt_lib_add_level(storage_lib,surf_resource_free); - - /* Connect our log channels: that must be done manually under windows */ - XBT_LOG_CONNECT(surf_cpu, surf); - XBT_LOG_CONNECT(surf_kernel, surf); - XBT_LOG_CONNECT(surf_lagrange, surf); - XBT_LOG_CONNECT(surf_lagrange_dichotomy, surf_lagrange); - XBT_LOG_CONNECT(surf_maxmin, surf); - XBT_LOG_CONNECT(surf_network, surf); - XBT_LOG_CONNECT(surf_trace, surf); - XBT_LOG_CONNECT(surf_parse, surf); - XBT_LOG_CONNECT(surf_timer, surf); - XBT_LOG_CONNECT(surf_workstation, surf); - XBT_LOG_CONNECT(surf_config, surf); - XBT_LOG_CONNECT(surf_route, surf); - -#ifdef HAVE_GTNETS - XBT_LOG_CONNECT(surf_network_gtnets, surf); -#endif 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(); } @@ -399,11 +470,14 @@ 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) { @@ -430,6 +504,8 @@ void surf_exit(void) xbt_lib_free(&storage_lib); xbt_lib_free(&storage_type_lib); + xbt_dict_free(&watched_hosts_lib); + tmgr_finalize(); surf_parse_lex_destroy(); surf_parse_free_callbacks(); @@ -455,17 +531,22 @@ 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) { + min = -1.0; /* duration */ double next_event_date = -1.0; double model_next_action_end = -1.0; @@ -482,33 +563,23 @@ 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; - - if (surf_get_nthreads() > 1) { - /* parallel version */ -#ifdef CONTEXT_THREADS - xbt_parmap_apply(surf_parmap, (void_f_pvoid_t) surf_share_resources, model_list); -#endif - } - else { - /* sequential version */ - xbt_dynar_foreach(model_list, iter, model) { - surf_share_resources(model); - } + /* sequential version */ + 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]; } } - XBT_DEBUG("Min for resources (remember that NS3 dont update that value) : %f", min); + XBT_DEBUG("Min for resources (remember that NS3 don't update that value) : %f", min); XBT_DEBUG("Looking for next trace event"); @@ -524,23 +595,23 @@ 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) { - XBT_DEBUG("no next TRACE event. Stop searching for it"); - break; + 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, @@ -557,7 +628,7 @@ 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); @@ -565,25 +636,17 @@ double surf_solve(double max_date) * This may cause an infinite loop if one cpu has a trace with periodicity = 0 and the other a trace with periodicity > 0. * The options are: all traces with same periodicity(0 or >0) or we need to change the way how the events are managed */ if (min == -1.0) { - XBT_DEBUG("No next event at all. Bail out now."); + XBT_DEBUG("No next event at all. Bail out now."); return -1.0; } XBT_DEBUG("Duration set to %f", min); 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); - } + /* FIXME: model_list or model_list_invoke? revisit here later */ + /* sequential version */ + xbt_dynar_foreach(model_list, iter, model) { + surf_update_actions_state(model); } #ifdef HAVE_TRACING @@ -604,7 +667,7 @@ 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); } @@ -613,43 +676,60 @@ 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); } -/** - * \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. + * + * 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. * - * A value of 1 means no parallelism. + * 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 +static void remove_watched_host(void *key) +{ + xbt_dict_remove(watched_hosts_lib, *(char**)key); +} - 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 +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); } - - surf_nthreads = nthreads; + xbt_dynar_map(hosts, remove_watched_host); + xbt_dynar_free(&hosts); }