src/surf/workstation.c

   1 /*      $Id$     */
   2
   3 /* Copyright (c) 2004 Arnaud Legrand. All rights reserved.                  */
   4
   5 /* This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the license (GNU LGPL) which comes with this package. */
   7
   8 #include "xbt/dict.h"
   9 #include "workstation_private.h"
  10 #include "cpu_private.h"
  11 #include "network_private.h"
  12
  13 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(workstation, surf,
  14                                 "Logging specific to the SURF workstation module");
  15
  16 surf_workstation_resource_t surf_workstation_resource = NULL;
  17
  18 xbt_dict_t workstation_set = NULL;
  19
  20 static workstation_CLM03_t workstation_new(const char *name,
  21                                      void *cpu, void *card)
  22 {
  23   workstation_CLM03_t workstation = xbt_new0(s_workstation_CLM03_t, 1);
  24
  25   workstation->resource = (surf_resource_t) surf_workstation_resource;
  26   workstation->name = xbt_strdup(name);
  27   workstation->cpu = cpu;
  28   workstation->network_card = card;
  29
  30   return workstation;
  31 }
  32
  33 static void workstation_free(void *workstation)
  34 {
  35   free(((workstation_CLM03_t)workstation)->name);
  36   free(workstation);
  37 }
  38
  39 static void create_workstations(void)
  40 {
  41    xbt_dict_cursor_t cursor = NULL;
  42   char *name = NULL;
  43   void *cpu = NULL;
  44   void *nw_card = NULL;
  45
  46   xbt_dict_foreach(cpu_set, cursor, name, cpu) {
  47     nw_card = NULL;
  48     xbt_dict_get(network_card_set, name, (void *) &nw_card);
  49     xbt_assert1(nw_card, "No corresponding card found for %s",name);
  50     xbt_dict_set(workstation_set, name,
  51                  workstation_new(name, cpu, nw_card), workstation_free);
  52   }
  53 }
  54
  55 static void *name_service(const char *name)
  56 {
  57   void *workstation = NULL;
  58
  59   xbt_dict_get(workstation_set, name, &workstation);
  60
  61   return workstation;
  62 }
  63
  64 static const char *get_resource_name(void *resource_id)
  65 {
  66   return ((workstation_CLM03_t) resource_id)->name;
  67 }
  68
  69 static int resource_used(void *resource_id)
  70 {
  71   xbt_assert0(0,
  72               "Workstation is a virtual resource. I should not be there!");
  73   return 0;
  74 }
  75
  76 static int parallel_action_free(surf_action_t action)
  77 {
  78   action->using--;
  79   if(!action->using) {
  80     xbt_swag_remove(action, action->state_set);
  81     if(((surf_action_parallel_task_CSL05_t)action)->variable)
  82       lmm_variable_free(maxmin_system, ((surf_action_parallel_task_CSL05_t)action)->variable);
  83     free(action);
  84     return 1;
  85   }
  86   return 0;
  87 }
  88
  89 static void parallel_action_use(surf_action_t action)
  90 {
  91   action->using++;
  92 }
  93
  94 static int action_free(surf_action_t action)
  95 {
  96   if(action->resource_type==(surf_resource_t)surf_network_resource)
  97     return surf_network_resource->common_public->action_free(action);
  98   else if(action->resource_type==(surf_resource_t)surf_cpu_resource)
  99     return surf_cpu_resource->common_public->action_free(action);
 100   else if(action->resource_type==(surf_resource_t)surf_workstation_resource)
 101     return parallel_action_free(action);
 102   else DIE_IMPOSSIBLE;
 103   return 0;
 104 }
 105
 106 static void action_use(surf_action_t action)
 107 {
 108   if(action->resource_type==(surf_resource_t)surf_network_resource)
 109     surf_network_resource->common_public->action_use(action);
 110   else if(action->resource_type==(surf_resource_t)surf_cpu_resource)
 111     surf_cpu_resource->common_public->action_use(action);
 112   else if(action->resource_type==(surf_resource_t)surf_workstation_resource)
 113     return parallel_action_use(action);
 114   else DIE_IMPOSSIBLE;
 115   return;
 116 }
 117
 118 static void action_cancel(surf_action_t action)
 119 {
 120   DIE_IMPOSSIBLE;
 121   return;
 122 }
 123
 124 static void action_recycle(surf_action_t action)
 125 {
 126   DIE_IMPOSSIBLE;
 127   return;
 128 }
 129
 130 static void action_change_state(surf_action_t action,
 131                                 e_surf_action_state_t state)
 132 {
 133   if(action->resource_type==(surf_resource_t)surf_network_resource)
 134     surf_network_resource->common_public->action_change_state(action,state);
 135   else if(action->resource_type==(surf_resource_t)surf_cpu_resource)
 136     surf_cpu_resource->common_public->action_change_state(action,state);
 137   else if(action->resource_type==(surf_resource_t)surf_workstation_resource)
 138     surf_action_change_state(action, state);
 139   else DIE_IMPOSSIBLE;
 140   return;
 141 }
 142
 143 static double share_resources(double now)
 144 {
 145   s_surf_action_parallel_task_CSL05_t action;
 146   return generic_maxmin_share_resources(surf_workstation_resource->common_public->
 147                                         states.running_action_set,
 148                                         xbt_swag_offset(action, variable));
 149 }
 150
 151 static void update_actions_state(double now, double delta)
 152 {
 153   surf_action_parallel_task_CSL05_t action = NULL;
 154   surf_action_parallel_task_CSL05_t next_action = NULL;
 155   xbt_swag_t running_actions =
 156       surf_workstation_resource->common_public->states.running_action_set;
 157   xbt_swag_t failed_actions =
 158       surf_workstation_resource->common_public->states.failed_action_set;
 159
 160   xbt_swag_foreach_safe(action, next_action, running_actions) {
 161     surf_double_update(&(action->generic_action.remains),
 162         lmm_variable_getvalue(action->variable) * delta);
 163     if (action->generic_action.max_duration != NO_MAX_DURATION)
 164       surf_double_update(&(action->generic_action.max_duration), delta);
 165     if ((action->generic_action.remains <= 0) &&
 166         (lmm_get_variable_weight(action->variable)>0)) {
 167       action->generic_action.finish = surf_get_clock();
 168       action_change_state((surf_action_t) action, SURF_ACTION_DONE);
 169     } else if ((action->generic_action.max_duration != NO_MAX_DURATION) &&
 170                (action->generic_action.max_duration <= 0)) {
 171       action->generic_action.finish = surf_get_clock();
 172       action_change_state((surf_action_t) action, SURF_ACTION_DONE);
 173     } else {                    /* Need to check that none of the resource has failed */
 174       lmm_constraint_t cnst = NULL;
 175       int i = 0;
 176       surf_resource_t resource = NULL;
 177
 178       while ((cnst =
 179               lmm_get_cnst_from_var(maxmin_system, action->variable,
 180                                     i++))) {
 181         resource = (surf_resource_t) lmm_constraint_id(cnst);
 182         if(resource== (surf_resource_t) surf_cpu_resource) {
 183           cpu_Cas01_t cpu = lmm_constraint_id(cnst);
 184           if (cpu->state_current == SURF_CPU_OFF) {
 185             action->generic_action.finish = surf_get_clock();
 186             action_change_state((surf_action_t) action, SURF_ACTION_FAILED);
 187             break;
 188           }
 189         } else if (resource== (surf_resource_t) surf_network_resource) {
 190           network_link_CM02_t nw_link = lmm_constraint_id(cnst);
 191
 192           if (nw_link->state_current == SURF_NETWORK_LINK_OFF) {
 193             action->generic_action.finish = surf_get_clock();
 194             action_change_state((surf_action_t) action, SURF_ACTION_FAILED);
 195             break;
 196           }
 197         }
 198       }
 199     }
 200   }
 201
 202   return;
 203 }
 204
 205 static void update_resource_state(void *id,
 206                                   tmgr_trace_event_t event_type,
 207                                   double value)
 208 {
 209   return;
 210 }
 211
 212 static surf_action_t execute(void *workstation, double size)
 213 {
 214   return surf_cpu_resource->extension_public->
 215       execute(((workstation_CLM03_t) workstation)->cpu, size);
 216 }
 217
 218 static surf_action_t action_sleep(void *workstation, double duration)
 219 {
 220   return surf_cpu_resource->extension_public->
 221       sleep(((workstation_CLM03_t) workstation)->cpu, duration);
 222 }
 223
 224 static void action_suspend(surf_action_t action)
 225 {
 226   if(action->resource_type==(surf_resource_t)surf_network_resource)
 227     surf_network_resource->common_public->suspend(action);
 228   else if(action->resource_type==(surf_resource_t)surf_cpu_resource)
 229     surf_cpu_resource->common_public->suspend(action);
 230   else DIE_IMPOSSIBLE;
 231 }
 232
 233 static void action_resume(surf_action_t action)
 234 {
 235   if(action->resource_type==(surf_resource_t)surf_network_resource)
 236     surf_network_resource->common_public->resume(action);
 237   else if(action->resource_type==(surf_resource_t)surf_cpu_resource)
 238     surf_cpu_resource->common_public->resume(action);
 239   else DIE_IMPOSSIBLE;
 240 }
 241
 242 static int action_is_suspended(surf_action_t action)
 243 {
 244   if(action->resource_type==(surf_resource_t)surf_network_resource)
 245     return surf_network_resource->common_public->is_suspended(action);
 246   if(action->resource_type==(surf_resource_t)surf_cpu_resource)
 247     return surf_cpu_resource->common_public->is_suspended(action);
 248   DIE_IMPOSSIBLE;
 249 }
 250
 251 static void action_set_max_duration(surf_action_t action, double duration)
 252 {
 253   if(action->resource_type==(surf_resource_t)surf_network_resource)
 254     surf_network_resource->common_public->set_max_duration(action,duration);
 255   else if(action->resource_type==(surf_resource_t)surf_cpu_resource)
 256     surf_cpu_resource->common_public->set_max_duration(action,duration);
 257   else  DIE_IMPOSSIBLE;
 258 }
 259
 260 static surf_action_t communicate(void *workstation_src,
 261                                  void *workstation_dst, double size,
 262                                  double rate)
 263 {
 264   return surf_network_resource->extension_public->
 265       communicate(((workstation_CLM03_t) workstation_src)->network_card,
 266                   ((workstation_CLM03_t) workstation_dst)->network_card, size, rate);
 267 }
 268
 269 static e_surf_cpu_state_t get_state(void *workstation)
 270 {
 271   return surf_cpu_resource->extension_public->
 272       get_state(((workstation_CLM03_t) workstation)->cpu);
 273 }
 274
 275 static surf_action_t execute_parallel_task (int workstation_nb,
 276                                             void **workstation_list,
 277                                             double *computation_amount,
 278                                             double *communication_amount,
 279                                             double amount,
 280                                             double rate)
 281 {
 282   surf_action_parallel_task_CSL05_t action = NULL;
 283   int i, j, k;
 284   xbt_dict_t network_link_set = xbt_dict_new();
 285   xbt_dict_cursor_t cursor = NULL;
 286   char *name = NULL;
 287   int nb_link = 0;
 288   int nb_host = 0;
 289   network_link_CM02_t link;
 290
 291   /* Compute the number of affected resources... */
 292   for(i=0; i< workstation_nb; i++) {
 293     for(j=0; j< workstation_nb; j++) {
 294       network_card_CM02_t card_src = ((workstation_CLM03_t*)workstation_list)[i]->network_card;
 295       network_card_CM02_t card_dst = ((workstation_CLM03_t*)workstation_list)[j]->network_card;
 296       int route_size = ROUTE_SIZE(card_src->id, card_dst->id);
 297       network_link_CM02_t *route = ROUTE(card_src->id, card_dst->id);
 298
 299       if(communication_amount[i*workstation_nb+j]>0)
 300         for(k=0; k< route_size; k++) {
 301           xbt_dict_set(network_link_set, route[k]->name, route[k], NULL);
 302         }
 303     }
 304   }
 305
 306   xbt_dict_foreach(network_link_set, cursor, name, link) {
 307     nb_link++;
 308   }
 309
 310   xbt_dict_free(&network_link_set);
 311
 312   for (i = 0; i<workstation_nb; i++)
 313     if(computation_amount[i]>0) nb_host++;
 314
 315   if(nb_link + workstation_nb == 0)
 316     return NULL;
 317
 318   action = xbt_new0(s_surf_action_parallel_task_CSL05_t, 1);
 319   action->generic_action.using = 1;
 320   action->generic_action.cost = amount;
 321   action->generic_action.remains = amount;
 322   action->generic_action.max_duration = NO_MAX_DURATION;
 323   action->generic_action.start = -1.0;
 324   action->generic_action.finish = -1.0;
 325   action->generic_action.resource_type =
 326       (surf_resource_t) surf_workstation_resource;
 327   action->suspended = 0;  /* Should be useless because of the
 328                              calloc but it seems to help valgrind... */
 329   action->generic_action.state_set =
 330       surf_workstation_resource->common_public->states.running_action_set;
 331
 332   xbt_swag_insert(action, action->generic_action.state_set);
 333   action->rate = rate;
 334
 335   if(action->rate>0)
 336     action->variable = lmm_variable_new(maxmin_system, action, 1.0, -1.0,
 337                                         nb_host + nb_link);
 338   else
 339     action->variable = lmm_variable_new(maxmin_system, action, 1.0, action->rate,
 340                                         nb_host + nb_link);
 341
 342   for (i = 0; i<workstation_nb; i++)
 343     if(computation_amount[i]>0)
 344       lmm_expand(maxmin_system, ((cpu_Cas01_t) ((workstation_CLM03_t) workstation_list[i])->cpu)->constraint,
 345                  action->variable, computation_amount[i]);
 346
 347   for (i=0; i<workstation_nb; i++) {
 348     for(j=0; j< workstation_nb; j++) {
 349       network_card_CM02_t card_src = ((workstation_CLM03_t*)workstation_list)[i]->network_card;
 350       network_card_CM02_t card_dst = ((workstation_CLM03_t*)workstation_list)[j]->network_card;
 351       int route_size = ROUTE_SIZE(card_src->id, card_dst->id);
 352       network_link_CM02_t *route = ROUTE(card_src->id, card_dst->id);
 353
 354       for(k=0; k< route_size; k++) {
 355         if(communication_amount[i*workstation_nb+j]>0) {
 356           lmm_expand_add(maxmin_system, route[k]->constraint,
 357                        action->variable, communication_amount[i*workstation_nb+j]);
 358         }
 359       }
 360     }
 361   }
 362
 363   return (surf_action_t) action;
 364 }
 365
 366 static void finalize(void)
 367 {
 368   xbt_dict_free(&workstation_set);
 369   xbt_swag_free(surf_workstation_resource->common_public->states.ready_action_set);
 370   xbt_swag_free(surf_workstation_resource->common_public->states.
 371                 running_action_set);
 372   xbt_swag_free(surf_workstation_resource->common_public->states.
 373                 failed_action_set);
 374   xbt_swag_free(surf_workstation_resource->common_public->states.done_action_set);
 375
 376   free(surf_workstation_resource->common_public);
 377   free(surf_workstation_resource->common_private);
 378   free(surf_workstation_resource->extension_public);
 379
 380   free(surf_workstation_resource);
 381   surf_workstation_resource = NULL;
 382 }
 383
 384 static void surf_workstation_resource_init_internal(void)
 385 {
 386   s_surf_action_t action;
 387
 388   surf_workstation_resource = xbt_new0(s_surf_workstation_resource_t, 1);
 389
 390   surf_workstation_resource->common_private =
 391       xbt_new0(s_surf_resource_private_t, 1);
 392   surf_workstation_resource->common_public =
 393       xbt_new0(s_surf_resource_public_t, 1);
 394 /*   surf_workstation_resource->extension_private = xbt_new0(s_surf_workstation_resource_extension_private_t,1); */
 395   surf_workstation_resource->extension_public =
 396       xbt_new0(s_surf_workstation_resource_extension_public_t, 1);
 397
 398   surf_workstation_resource->common_public->states.ready_action_set =
 399       xbt_swag_new(xbt_swag_offset(action, state_hookup));
 400   surf_workstation_resource->common_public->states.running_action_set =
 401       xbt_swag_new(xbt_swag_offset(action, state_hookup));
 402   surf_workstation_resource->common_public->states.failed_action_set =
 403       xbt_swag_new(xbt_swag_offset(action, state_hookup));
 404   surf_workstation_resource->common_public->states.done_action_set =
 405       xbt_swag_new(xbt_swag_offset(action, state_hookup));
 406
 407   surf_workstation_resource->common_public->name_service = name_service;
 408   surf_workstation_resource->common_public->get_resource_name =
 409       get_resource_name;
 410   surf_workstation_resource->common_public->action_get_state =
 411       surf_action_get_state;
 412   surf_workstation_resource->common_public->action_free = action_free;
 413   surf_workstation_resource->common_public->action_use = action_use;
 414   surf_workstation_resource->common_public->action_cancel = action_cancel;
 415   surf_workstation_resource->common_public->action_recycle =
 416       action_recycle;
 417   surf_workstation_resource->common_public->action_change_state =
 418       action_change_state;
 419   surf_workstation_resource->common_public->action_set_data = surf_action_set_data;
 420   surf_workstation_resource->common_public->name = "Workstation";
 421
 422   surf_workstation_resource->common_private->resource_used = resource_used;
 423   surf_workstation_resource->common_private->share_resources =
 424       share_resources;
 425   surf_workstation_resource->common_private->update_actions_state =
 426       update_actions_state;
 427   surf_workstation_resource->common_private->update_resource_state =
 428       update_resource_state;
 429   surf_workstation_resource->common_private->finalize = finalize;
 430
 431   surf_workstation_resource->common_public->suspend = action_suspend;
 432   surf_workstation_resource->common_public->resume = action_resume;
 433   surf_workstation_resource->common_public->is_suspended = action_is_suspended;
 434   surf_workstation_resource->common_public->set_max_duration = action_set_max_duration;
 435
 436   surf_workstation_resource->extension_public->execute = execute;
 437   surf_workstation_resource->extension_public->sleep = action_sleep;
 438   surf_workstation_resource->extension_public->get_state = get_state;
 439   surf_workstation_resource->extension_public->communicate = communicate;
 440   surf_workstation_resource->extension_public->execute_parallel_task =
 441     execute_parallel_task;
 442
 443   workstation_set = xbt_dict_new();
 444
 445   xbt_assert0(maxmin_system, "surf_init has to be called first!");
 446 }
 447
 448 /********************************************************************/
 449 /* The model used in MSG and presented at CCGrid03                  */
 450 /********************************************************************/
 451 /* @InProceedings{Casanova.CLM_03, */
 452 /*   author = {Henri Casanova and Arnaud Legrand and Loris Marchal}, */
 453 /*   title = {Scheduling Distributed Applications: the SimGrid Simulation Framework}, */
 454 /*   booktitle = {Proceedings of the third IEEE International Symposium on Cluster Computing and the Grid (CCGrid'03)}, */
 455 /*   publisher = {"IEEE Computer Society Press"}, */
 456 /*   month = {may}, */
 457 /*   year = {2003} */
 458 /* } */
 459 void surf_workstation_resource_init_CLM03(const char *filename)
 460 {
 461 /*   int i ; */
 462 /*   surf_resource_t resource =  NULL; */
 463
 464   surf_workstation_resource_init_internal();
 465   surf_cpu_resource_init_Cas01(filename);
 466   surf_network_resource_init_CM02(filename);
 467   create_workstations();
 468   xbt_dynar_push(resource_list, &surf_workstation_resource);
 469 /*   xbt_dynar_foreach(resource_list, i, resource) { */
 470 /*     if(resource==surf_cpu_resource) { */
 471 /*       xbt_dynar_remove_at(resource_list, i, NULL); */
 472 /*       i--;  */
 473 /*       continue; */
 474 /*     } */
 475 /*     if(resource==surf_network_resource) { */
 476 /*       xbt_dynar_remove_at(resource_list, i, NULL); */
 477 /*       i--;  */
 478 /*       continue; */
 479 /*     } */
 480 /*   } */
 481 }