1 /* Copyright (c) 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011. The SimGrid Team.
2 * All rights reserved. */
4 /* This program is free software; you can redistribute it and/or modify it
5 * under the terms of the license (GNU LGPL) which comes with this package. */
7 #include "surf_private.h"
8 #include "xbt/module.h"
10 #include "simix/smx_host_private.h"
11 #include "surf/surf_resource.h"
12 #include "xbt/xbt_os_thread.h"
13 #include "simgrid/sg_config.h"
17 XBT_LOG_NEW_CATEGORY(surf, "All SURF categories");
18 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_kernel, surf,
19 "Logging specific to SURF (kernel)");
21 /* Additional declarations for Windows portability. */
29 static const char *disk_drives_letter_table[MAX_DRIVE] = {
57 #endif /* #ifdef _XBT_WIN32 */
60 * Returns the initial path. On Windows the initial path is
61 * the current directory for the current process in the other
62 * case the function returns "./" that represents the current
63 * directory on Unix/Linux platforms.
66 const char *__surf_get_initial_path(void)
71 char current_directory[MAX_PATH + 1] = { 0 };
72 unsigned int len = GetCurrentDirectory(MAX_PATH + 1, current_directory);
78 strncpy(root, current_directory, 3);
80 for (i = 0; i < MAX_DRIVE; i++) {
81 if (toupper(root[0]) == disk_drives_letter_table[i][0])
82 return disk_drives_letter_table[i];
91 /* The __surf_is_absolute_file_path() returns 1 if
92 * file_path is a absolute file path, in the other
93 * case the function returns 0.
95 int __surf_is_absolute_file_path(const char *file_path)
98 WIN32_FIND_DATA wfd = { 0 };
99 HANDLE hFile = FindFirstFile(file_path, &wfd);
101 if (INVALID_HANDLE_VALUE == hFile)
107 return (file_path[0] == '/');
113 xbt_dynar_t model_list = NULL;
114 tmgr_history_t history = NULL;
115 lmm_system_t maxmin_system = NULL;
116 xbt_dynar_t surf_path = NULL;
118 /* Don't forget to update the option description in smx_config when you change this */
119 s_surf_model_description_t surf_network_model_description[] = {
121 "Realistic network analytic model (slow-start modeled by multiplying latency by 10.4, bandwidth by .92; bottleneck sharing uses a payload of S=8775 for evaluating RTT). ",
122 surf_network_model_init_LegrandVelho},
124 "Simplistic network model where all communication take a constant time (one second). This model provides the lowest realism, but is (marginally) faster.",
125 surf_network_model_init_Constant},
127 "Realistic network model specifically tailored for HPC settings (accurate modeling of slow start with correction factors on three intervals: < 1KiB, < 64 KiB, >= 64 KiB)",
128 surf_network_model_init_SMPI},
130 "Legacy network analytic model (Very similar to LV08, but without corrective factors. The timings of small messages are thus poorly modeled).",
131 surf_network_model_init_CM02},
134 "Network pseudo-model using the GTNets simulator instead of an analytic model",
135 surf_network_model_init_GTNETS},
139 "Network pseudo-model using the NS3 tcp model instead of an analytic model",
140 surf_network_model_init_NS3},
143 "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
144 surf_network_model_init_Reno},
146 "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
147 surf_network_model_init_Reno2},
149 "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
150 surf_network_model_init_Vegas},
151 {NULL, NULL, NULL} /* this array must be NULL terminated */
154 s_surf_model_description_t surf_cpu_model_description[] = {
156 "Simplistic CPU model (time=size/power).",
157 surf_cpu_model_init_Cas01},
158 {NULL, NULL, NULL} /* this array must be NULL terminated */
161 s_surf_model_description_t surf_workstation_model_description[] = {
163 "Default workstation model. Currently, CPU:Cas01 and network:LV08 (with cross traffic enabled)",
164 surf_workstation_model_init_current_default},
166 "Workstation model that is automatically chosen if you change the network and CPU models",
167 surf_workstation_model_init_compound},
168 {"ptask_L07", "Workstation model somehow similar to Cas01+CM02 but allowing parallel tasks",
169 surf_workstation_model_init_ptask_L07},
170 {NULL, NULL, NULL} /* this array must be NULL terminated */
173 s_surf_model_description_t surf_optimization_mode_description[] = {
175 "Lazy action management (partial invalidation in lmm + heap in action remaining).",
178 "Trace integration. Highly optimized mode when using availability traces (only available for the Cas01 CPU model for now).",
181 "Full update of remaining and variables. Slow but may be useful when debugging.",
183 {NULL, NULL, NULL} /* this array must be NULL terminated */
186 s_surf_model_description_t surf_storage_model_description[] = {
188 "Simplistic storage model.",
189 surf_storage_model_init_default},
190 {NULL, NULL, NULL} /* this array must be NULL terminated */
193 /* ********************************************************************* */
194 /* TUTORIAL: New model */
195 s_surf_model_description_t surf_new_model_description[] = {
198 surf_new_model_init_default},
199 {NULL, NULL, NULL} /* this array must be NULL terminated */
201 /* ********************************************************************* */
203 #ifdef CONTEXT_THREADS
204 static xbt_parmap_t surf_parmap = NULL; /* parallel map on models */
207 static double *surf_mins = NULL; /* return value of share_resources for each model */
208 static int surf_min_index; /* current index in surf_mins */
209 static double min; /* duration determined by surf_solve */
211 static void surf_share_resources(surf_model_t model);
212 static void surf_update_actions_state(surf_model_t model);
214 /** Displays the long description of all registered models, and quit */
215 void model_help(const char *category, s_surf_model_description_t * table)
218 printf("Long description of the %s models accepted by this simulator:\n",
220 for (i = 0; table[i].name; i++)
221 printf(" %s: %s\n", table[i].name, table[i].description);
224 int find_model_description(s_surf_model_description_t * table,
228 char *name_list = NULL;
230 for (i = 0; table[i].name; i++)
231 if (!strcmp(name, table[i].name)) {
234 name_list = strdup(table[0].name);
235 for (i = 1; table[i].name; i++) {
237 xbt_realloc(name_list,
238 strlen(name_list) + strlen(table[i].name) + 3);
239 strcat(name_list, ", ");
240 strcat(name_list, table[i].name);
242 xbt_die("Model '%s' is invalid! Valid models are: %s.", name, name_list);
246 double generic_maxmin_share_resources(xbt_swag_t running_actions,
249 void (*solve) (lmm_system_t))
251 surf_action_t action = NULL;
254 #define VARIABLE(action) (*((lmm_variable_t*)(((char *) (action)) + (offset))))
258 xbt_swag_foreach(action, running_actions) {
259 value = lmm_variable_getvalue(VARIABLE(action));
260 if ((value > 0) || (action->max_duration >= 0))
268 if (action->remains > 0)
269 min = action->remains / value;
272 if ((action->max_duration >= 0) && (action->max_duration < min))
273 min = action->max_duration;
275 min = action->max_duration;
278 for (action = xbt_swag_getNext(action, running_actions->offset);
280 action = xbt_swag_getNext(action, running_actions->offset)) {
281 value = lmm_variable_getvalue(VARIABLE(action));
283 if (action->remains > 0)
284 value = action->remains / value;
289 XBT_DEBUG("Updating min (value) with %p: %f", action, min);
292 if ((action->max_duration >= 0) && (action->max_duration < min)) {
293 min = action->max_duration;
294 XBT_DEBUG("Updating min (duration) with %p: %f", action, min);
297 XBT_DEBUG("min value : %f", min);
303 double generic_share_resources_lazy(double now, surf_model_t model)
305 surf_action_lmm_t action = NULL;
310 ("Before share resources, the size of modified actions set is %d",
311 xbt_swag_size(model->model_private->modified_set));
313 lmm_solve(model->model_private->maxmin_system);
316 ("After share resources, The size of modified actions set is %d",
317 xbt_swag_size(model->model_private->modified_set));
319 while((action = xbt_swag_extract(model->model_private->modified_set))) {
320 int max_dur_flag = 0;
322 if (action->generic_action.state_set !=
323 model->states.running_action_set)
326 /* bogus priority, skip it */
327 if (action->generic_action.priority <= 0)
330 generic_update_action_remaining_lazy(action,now);
333 value = lmm_variable_getvalue(action->variable);
335 if (action->generic_action.remains > 0) {
336 value = action->generic_action.remains / value;
344 if ((action->generic_action.max_duration != NO_MAX_DURATION)
346 || action->generic_action.start +
347 action->generic_action.max_duration < min)) {
348 min = action->generic_action.start +
349 action->generic_action.max_duration;
353 XBT_DEBUG("Action(%p) Start %lf Finish %lf Max_duration %lf", action,
354 action->generic_action.start, now + value,
355 action->generic_action.max_duration);
358 surf_action_lmm_heap_remove(model->model_private->action_heap,action);
359 surf_action_lmm_heap_insert(model->model_private->action_heap,action, min, max_dur_flag ? MAX_DURATION : NORMAL);
360 XBT_DEBUG("Insert at heap action(%p) min %lf now %lf", action, min,
362 } else DIE_IMPOSSIBLE;
365 //hereafter must have already the min value for this resource model
366 if (xbt_heap_size(model->model_private->action_heap) > 0)
367 min = xbt_heap_maxkey(model->model_private->action_heap) - now;
371 XBT_DEBUG("The minimum with the HEAP %lf", min);
375 static XBT_INLINE void routing_asr_host_free(void *p)
377 sg_routing_edge_t elm = p;
382 static XBT_INLINE void routing_asr_prop_free(void *p)
388 void sg_version(int *ver_major,int *ver_minor,int *ver_patch) {
389 *ver_major = SIMGRID_VERSION_MAJOR;
390 *ver_minor = SIMGRID_VERSION_MINOR;
391 *ver_patch = SIMGRID_VERSION_PATCH;
394 void surf_init(int *argc, char **argv)
396 XBT_DEBUG("Create all Libs");
397 host_lib = xbt_lib_new();
398 link_lib = xbt_lib_new();
399 as_router_lib = xbt_lib_new();
400 storage_lib = xbt_lib_new();
401 storage_type_lib = xbt_lib_new();
402 watched_hosts_lib = xbt_dict_new();
404 XBT_DEBUG("Add routing levels");
405 ROUTING_HOST_LEVEL = xbt_lib_add_level(host_lib,routing_asr_host_free);
406 ROUTING_ASR_LEVEL = xbt_lib_add_level(as_router_lib,routing_asr_host_free);
407 ROUTING_PROP_ASR_LEVEL = xbt_lib_add_level(as_router_lib,routing_asr_prop_free);
409 XBT_DEBUG("Add SURF levels");
410 SURF_CPU_LEVEL = xbt_lib_add_level(host_lib,surf_resource_free);
411 SURF_WKS_LEVEL = xbt_lib_add_level(host_lib,surf_resource_free);
412 SURF_LINK_LEVEL = xbt_lib_add_level(link_lib,surf_resource_free);
414 xbt_init(argc, argv);
416 model_list = xbt_dynar_new(sizeof(surf_model_private_t), NULL);
418 history = tmgr_history_new();
421 TRACE_add_start_function(TRACE_surf_alloc);
422 TRACE_add_end_function(TRACE_surf_release);
425 sg_config_init(argc, argv);
433 # define FILE_DELIM "\\"
435 # define FILE_DELIM "/" /* FIXME: move to better location */
438 FILE *surf_fopen(const char *name, const char *mode)
441 char *path_elm = NULL;
447 if (__surf_is_absolute_file_path(name)) /* don't mess with absolute file names */
448 return fopen(name, mode);
450 /* search relative files in the path */
451 xbt_dynar_foreach(surf_path, cpt, path_elm) {
452 buff = bprintf("%s" FILE_DELIM "%s", path_elm, name);
453 file = fopen(buff, mode);
465 surf_model_t model = NULL;
467 sg_config_finalize();
469 xbt_dynar_foreach(model_list, iter, model)
470 model->model_private->finalize();
471 xbt_dynar_free(&model_list);
475 lmm_system_free(maxmin_system);
476 maxmin_system = NULL;
479 tmgr_history_free(history);
484 #ifdef CONTEXT_THREADS
485 xbt_parmap_destroy(surf_parmap);
490 xbt_dynar_free(&surf_path);
492 xbt_lib_free(&host_lib);
493 xbt_lib_free(&link_lib);
494 xbt_lib_free(&as_router_lib);
495 xbt_lib_free(&storage_lib);
496 xbt_lib_free(&storage_type_lib);
498 xbt_dict_free(&watched_hosts_lib);
501 surf_parse_lex_destroy();
502 surf_parse_free_callbacks();
504 NOW = 0; /* Just in case the user plans to restart the simulation afterward */
507 void surf_presolve(void)
509 double next_event_date = -1.0;
510 tmgr_trace_event_t event = NULL;
512 surf_resource_t resource = NULL;
513 surf_model_t model = NULL;
517 ("First Run! Let's \"purge\" events and put models in the right state");
518 while ((next_event_date = tmgr_history_next_date(history)) != -1.0) {
519 if (next_event_date > NOW)
522 tmgr_history_get_next_event_leq(history, next_event_date,
524 (void **) &resource))) {
526 resource->model->model_private->update_resource_state(resource,
532 xbt_dynar_foreach(model_list, iter, model)
533 model->model_private->update_actions_state(NOW, 0.0);
536 double surf_solve(double max_date)
538 min = -1.0; /* duration */
539 double next_event_date = -1.0;
540 double model_next_action_end = -1.0;
542 surf_resource_t resource = NULL;
543 surf_model_t model = NULL;
544 tmgr_trace_event_t event = NULL;
547 if (max_date != -1.0 && max_date != NOW) {
548 min = max_date - NOW;
551 XBT_DEBUG("Looking for next action end for all models except NS3");
553 if (surf_mins == NULL) {
554 surf_mins = xbt_new(double, xbt_dynar_length(model_list));
558 /* sequential version */
559 xbt_dynar_foreach(model_list, iter, model) {
560 surf_share_resources(model);
564 for (i = 0; i < xbt_dynar_length(model_list); i++) {
565 if ((min < 0.0 || surf_mins[i] < min)
566 && surf_mins[i] >= 0.0) {
571 XBT_DEBUG("Min for resources (remember that NS3 don't update that value) : %f", min);
573 XBT_DEBUG("Looking for next trace event");
576 XBT_DEBUG("Next TRACE event : %f", next_event_date);
578 next_event_date = tmgr_history_next_date(history);
580 if(surf_network_model->name && !strcmp(surf_network_model->name,"network NS3")){
581 if(next_event_date!=-1.0 && min!=-1.0) {
582 min = MIN(next_event_date - NOW, min);
584 min = MAX(next_event_date - NOW, min);
587 XBT_DEBUG("Run for network at most %f", min);
588 // run until min or next flow
589 model_next_action_end = surf_network_model->model_private->share_resources(min);
591 XBT_DEBUG("Min for network : %f", model_next_action_end);
592 if(model_next_action_end>=0.0)
593 min = model_next_action_end;
596 if (next_event_date < 0.0) {
597 XBT_DEBUG("no next TRACE event. Stop searching for it");
601 if ((min == -1.0) || (next_event_date > NOW + min)) break;
603 XBT_DEBUG("Updating models (min = %g, NOW = %g, next_event_date = %g)",min, NOW, next_event_date);
605 tmgr_history_get_next_event_leq(history, next_event_date,
607 (void **) &resource))) {
608 if (resource->model->model_private->resource_used(resource)) {
609 min = next_event_date - NOW;
611 ("This event will modify model state. Next event set to %f",
614 /* update state of model_obj according to new value. Does not touch lmm.
615 It will be modified if needed when updating actions */
616 XBT_DEBUG("Calling update_resource_state for resource %s with min %lf",
617 resource->model->name, min);
618 resource->model->model_private->update_resource_state(resource,
624 /* FIXME: Moved this test to here to avoid stopping simulation if there are actions running on cpus and all cpus are with availability = 0.
625 * This may cause an infinite loop if one cpu has a trace with periodicity = 0 and the other a trace with periodicity > 0.
626 * The options are: all traces with same periodicity(0 or >0) or we need to change the way how the events are managed */
628 XBT_DEBUG("No next event at all. Bail out now.");
632 XBT_DEBUG("Duration set to %f", min);
636 /* sequential version */
637 xbt_dynar_foreach(model_list, iter, model) {
638 surf_update_actions_state(model);
642 TRACE_paje_dump_buffer (0);
648 XBT_INLINE double surf_get_clock(void)
653 static void surf_share_resources(surf_model_t model)
655 double next_action_end = -1.0;
656 int i = __sync_fetch_and_add(&surf_min_index, 1);
657 if (strcmp(model->name,"network NS3")) {
658 XBT_DEBUG("Running for Resource [%s]", model->name);
659 next_action_end = model->model_private->share_resources(NOW);
660 XBT_DEBUG("Resource [%s] : next action end = %f",
661 model->name, next_action_end);
663 surf_mins[i] = next_action_end;
666 static void surf_update_actions_state(surf_model_t model)
668 model->model_private->update_actions_state(NOW, min);
672 /* This function is a pimple that we ought to fix. But it won't be easy.
674 * The surf_solve() function does properly return the set of actions that changed.
675 * Instead, each model change a global data, and then the caller of surf_solve must
676 * pick into these sets of action_failed and action_done.
678 * This was not clean but ok as long as we didn't had to restart the processes when the resource comes back up.
679 * We worked by putting sentinel actions on every resources we are interested in,
680 * so that surf informs us if/when the corresponding resource fails.
682 * But this does not work to get Simix informed of when a resource comes back up, and this is where this pimple comes.
683 * We have a set of resources that are currently down and for which simix needs to know when it comes back up.
684 * And the current function is called *at every simulation step* to sweep over that set, searching for a resource
685 * that was turned back up in the meanwhile. This is UGLY and slow.
687 * The proper solution would be to not rely on globals for the action_failed and action_done swags.
688 * They must be passed as parameter by the caller (the handling of these actions in simix may let you
689 * think that these two sets can be merged, but their handling in SimDag induce the contrary unless this
690 * simdag code can check by itself whether the action is done of failed -- seems very doable, but yet more
693 * Once surf_solve() is passed the set of actions that changed, you want to add a new set of resources back up
694 * as parameter to this function. You also want to add a boolean field "restart_watched" to each resource, and
695 * make sure that whenever a resource with this field enabled comes back up, it's added to that set so that Simix
696 * sees it and react accordingly. This would kill that need for surf to call simix.
700 static void remove_watched_host(void *key)
702 xbt_dict_remove(watched_hosts_lib, *(char**)key);
705 void surf_watched_hosts(void)
709 xbt_dict_cursor_t cursor;
710 xbt_dynar_t hosts = xbt_dynar_new(sizeof(char*), NULL);
712 XBT_DEBUG("Check for host SURF_RESOURCE_ON on watched_hosts_lib");
713 xbt_dict_foreach(watched_hosts_lib,cursor,key,host)
715 if(SIMIX_host_get_state(host) == SURF_RESOURCE_ON){
716 XBT_INFO("Restart processes on host: %s",SIMIX_host_get_name(host));
717 SIMIX_host_autorestart(host);
718 xbt_dynar_push_as(hosts, char*, key);
721 XBT_DEBUG("See SURF_RESOURCE_OFF on host: %s",key);
723 xbt_dynar_map(hosts, remove_watched_host);
724 xbt_dynar_free(&hosts);