src/surf/cpu_cas01.c

   1 /* Copyright (c) 2009-2011. The SimGrid Team.
   2  * All rights reserved.                                                     */
   3
   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. */
   6
   7 #include "surf_private.h"
   8 #include "surf/surf_resource.h"
   9 #include "maxmin_private.h"
  10
  11 surf_model_t surf_cpu_model = NULL;
  12
  13 #undef GENERIC_LMM_ACTION
  14 #undef GENERIC_ACTION
  15 #undef ACTION_GET_CPU
  16 #define GENERIC_LMM_ACTION(action) action->generic_lmm_action
  17 #define GENERIC_ACTION(action) GENERIC_LMM_ACTION(action).generic_action
  18 #define ACTION_GET_CPU(action) ((surf_action_cpu_Cas01_t) action)->cpu
  19
  20 typedef struct surf_action_cpu_cas01 {
  21   s_surf_action_lmm_t generic_lmm_action;
  22 } s_surf_action_cpu_Cas01_t, *surf_action_cpu_Cas01_t;
  23
  24 typedef struct cpu_Cas01 {
  25   s_surf_resource_t generic_resource;
  26   s_xbt_swag_hookup_t modified_cpu_hookup;
  27   double power_peak;
  28   double power_scale;
  29   tmgr_trace_event_t power_event;
  30   int core;
  31   e_surf_resource_state_t state_current;
  32   tmgr_trace_event_t state_event;
  33   lmm_constraint_t constraint;
  34 } s_cpu_Cas01_t, *cpu_Cas01_t;
  35
  36 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_cpu, surf,
  37                                 "Logging specific to the SURF CPU IMPROVED module");
  38
  39
  40
  41 static xbt_swag_t
  42     cpu_running_action_set_that_does_not_need_being_checked = NULL;
  43
  44
  45 static void *cpu_create_resource(const char *name, double power_peak,
  46                                  double power_scale,
  47                                  tmgr_trace_t power_trace,
  48                                  int core,
  49                                  e_surf_resource_state_t state_initial,
  50                                  tmgr_trace_t state_trace,
  51                                  xbt_dict_t cpu_properties)
  52 {
  53   cpu_Cas01_t cpu = NULL;
  54
  55   xbt_assert(!surf_cpu_resource_by_name(name),
  56              "Host '%s' declared several times in the platform file",
  57              name);
  58   cpu = (cpu_Cas01_t) surf_resource_new(sizeof(s_cpu_Cas01_t),
  59                                         surf_cpu_model, name,
  60                                         cpu_properties);
  61   cpu->power_peak = power_peak;
  62   xbt_assert(cpu->power_peak > 0, "Power has to be >0");
  63   cpu->power_scale = power_scale;
  64   cpu->core = core;
  65   xbt_assert(core > 0, "Invalid number of cores %d", core);
  66
  67   if (power_trace)
  68     cpu->power_event =
  69         tmgr_history_add_trace(history, power_trace, 0.0, 0, cpu);
  70
  71   cpu->state_current = state_initial;
  72   if (state_trace)
  73     cpu->state_event =
  74         tmgr_history_add_trace(history, state_trace, 0.0, 0, cpu);
  75
  76   cpu->constraint =
  77       lmm_constraint_new(surf_cpu_model->model_private->maxmin_system, cpu,
  78                          cpu->core * cpu->power_scale * cpu->power_peak);
  79
  80   xbt_lib_set(host_lib, name, SURF_CPU_LEVEL, cpu);
  81
  82   return cpu;
  83 }
  84
  85
  86 static void parse_cpu_init(sg_platf_host_cbarg_t host)
  87 {
  88   cpu_create_resource(host->id,
  89                       host->power_peak,
  90                       host->power_scale,
  91                       host->power_trace,
  92                       host->core_amount,
  93                       host->initial_state,
  94                       host->state_trace, host->properties);
  95 }
  96
  97 static void cpu_add_traces_cpu(void)
  98 {
  99   xbt_dict_cursor_t cursor = NULL;
 100   char *trace_name, *elm;
 101   static int called = 0;
 102   if (called)
 103     return;
 104   called = 1;
 105
 106   /* connect all traces relative to hosts */
 107   xbt_dict_foreach(trace_connect_list_host_avail, cursor, trace_name, elm) {
 108     tmgr_trace_t trace = xbt_dict_get_or_null(traces_set_list, trace_name);
 109     cpu_Cas01_t host = surf_cpu_resource_by_name(elm);
 110
 111     xbt_assert(host, "Host %s undefined", elm);
 112     xbt_assert(trace, "Trace %s undefined", trace_name);
 113
 114     host->state_event =
 115         tmgr_history_add_trace(history, trace, 0.0, 0, host);
 116   }
 117
 118   xbt_dict_foreach(trace_connect_list_power, cursor, trace_name, elm) {
 119     tmgr_trace_t trace = xbt_dict_get_or_null(traces_set_list, trace_name);
 120     cpu_Cas01_t host = surf_cpu_resource_by_name(elm);
 121
 122     xbt_assert(host, "Host %s undefined", elm);
 123     xbt_assert(trace, "Trace %s undefined", trace_name);
 124
 125     host->power_event =
 126         tmgr_history_add_trace(history, trace, 0.0, 0, host);
 127   }
 128 }
 129
 130 static void cpu_define_callbacks()
 131 {
 132   sg_platf_host_add_cb(parse_cpu_init);
 133   sg_platf_postparse_add_cb(cpu_add_traces_cpu);
 134 }
 135
 136 static int cpu_resource_used(void *resource)
 137 {
 138   return lmm_constraint_used(surf_cpu_model->model_private->maxmin_system,
 139                              ((cpu_Cas01_t) resource)->constraint);
 140 }
 141
 142 static double cpu_share_resources_lazy(double now)
 143 {
 144   return generic_share_resources_lazy(now, surf_cpu_model);
 145 }
 146
 147 static double cpu_share_resources_full(double now)
 148 {
 149   s_surf_action_cpu_Cas01_t action;
 150   return generic_maxmin_share_resources(surf_cpu_model->states.
 151                                         running_action_set,
 152                                         xbt_swag_offset(action,
 153                                                         generic_lmm_action.
 154                                                         variable),
 155                                         surf_cpu_model->model_private->maxmin_system, lmm_solve);
 156 }
 157
 158 static void cpu_update_actions_state_lazy(double now, double delta)
 159 {
 160   surf_action_cpu_Cas01_t action;
 161   while ((xbt_heap_size(surf_cpu_model->model_private->action_heap) > 0)
 162          && (double_equals(xbt_heap_maxkey(surf_cpu_model->model_private->action_heap), now))) {
 163     action = xbt_heap_pop(surf_cpu_model->model_private->action_heap);
 164     XBT_DEBUG("Action %p: finish", action);
 165     GENERIC_ACTION(action).finish = surf_get_clock();
 166 #ifdef HAVE_TRACING
 167     if (TRACE_is_enabled()) {
 168       cpu_Cas01_t cpu =
 169           lmm_constraint_id(lmm_get_cnst_from_var
 170                             (surf_cpu_model->model_private->maxmin_system,
 171                              GENERIC_LMM_ACTION(action).variable, 0));
 172       TRACE_surf_host_set_utilization(cpu->generic_resource.name,
 173                                       ((surf_action_t)action)->category,
 174                                       lmm_variable_getvalue(GENERIC_LMM_ACTION(action).variable),
 175                                       GENERIC_LMM_ACTION(action).last_update,
 176                                       now - GENERIC_LMM_ACTION(action).last_update);
 177     }
 178 #endif
 179     /* set the remains to 0 due to precision problems when updating the remaining amount */
 180     GENERIC_ACTION(action).remains = 0;
 181     surf_action_state_set((surf_action_t) action, SURF_ACTION_DONE);
 182     surf_action_lmm_heap_remove(surf_cpu_model->model_private->action_heap,(surf_action_lmm_t)action); //FIXME: strange call since action was already popped
 183   }
 184 #ifdef HAVE_TRACING
 185   if (TRACE_is_enabled()) {
 186     //defining the last timestamp that we can safely dump to trace file
 187     //without losing the event ascending order (considering all CPU's)
 188     double smaller = -1;
 189     xbt_swag_t running_actions = surf_cpu_model->states.running_action_set;
 190     xbt_swag_foreach(action, running_actions) {
 191         if (smaller < 0) {
 192           smaller = GENERIC_LMM_ACTION(action).last_update;
 193           continue;
 194         }
 195         if (GENERIC_LMM_ACTION(action).last_update < smaller) {
 196           smaller = GENERIC_LMM_ACTION(action).last_update;
 197         }
 198     }
 199     if (smaller > 0) {
 200       TRACE_last_timestamp_to_dump = smaller;
 201     }
 202   }
 203 #endif
 204   return;
 205 }
 206
 207 static void cpu_update_actions_state_full(double now, double delta)
 208 {
 209   surf_action_cpu_Cas01_t action = NULL;
 210   surf_action_cpu_Cas01_t next_action = NULL;
 211   xbt_swag_t running_actions = surf_cpu_model->states.running_action_set;
 212   xbt_swag_foreach_safe(action, next_action, running_actions) {
 213 #ifdef HAVE_TRACING
 214     if (TRACE_is_enabled()) {
 215       cpu_Cas01_t x =
 216           lmm_constraint_id(lmm_get_cnst_from_var
 217                             (surf_cpu_model->model_private->maxmin_system,
 218                              GENERIC_LMM_ACTION(action).variable, 0));
 219
 220       TRACE_surf_host_set_utilization(x->generic_resource.name,
 221                                       ((surf_action_t)action)->category,
 222                                       lmm_variable_getvalue(GENERIC_LMM_ACTION(action).
 223                                        variable),
 224                                       now - delta,
 225                                       delta);
 226       TRACE_last_timestamp_to_dump = now - delta;
 227     }
 228 #endif
 229     double_update(&(GENERIC_ACTION(action).remains),
 230                   lmm_variable_getvalue(GENERIC_LMM_ACTION(action).
 231                                         variable) * delta);
 232     if (GENERIC_LMM_ACTION(action).generic_action.max_duration !=
 233         NO_MAX_DURATION)
 234       double_update(&(GENERIC_ACTION(action).max_duration), delta);
 235     if ((GENERIC_ACTION(action).remains <= 0) &&
 236         (lmm_get_variable_weight(GENERIC_LMM_ACTION(action).variable) >
 237          0)) {
 238       GENERIC_ACTION(action).finish = surf_get_clock();
 239       surf_action_state_set((surf_action_t) action, SURF_ACTION_DONE);
 240     } else if ((GENERIC_ACTION(action).max_duration != NO_MAX_DURATION) &&
 241                (GENERIC_ACTION(action).max_duration <= 0)) {
 242       GENERIC_ACTION(action).finish = surf_get_clock();
 243       surf_action_state_set((surf_action_t) action, SURF_ACTION_DONE);
 244     }
 245   }
 246
 247   return;
 248 }
 249
 250 static void cpu_update_resource_state(void *id,
 251                                       tmgr_trace_event_t event_type,
 252                                       double value, double date)
 253 {
 254   cpu_Cas01_t cpu = id;
 255   lmm_variable_t var = NULL;
 256   lmm_element_t elem = NULL;
 257
 258   if (event_type == cpu->power_event) {
 259     cpu->power_scale = value;
 260     lmm_update_constraint_bound(surf_cpu_model->model_private->maxmin_system, cpu->constraint,
 261                                 cpu->core * cpu->power_scale *
 262                                 cpu->power_peak);
 263 #ifdef HAVE_TRACING
 264     TRACE_surf_host_set_power(date, cpu->generic_resource.name,
 265                               cpu->core * cpu->power_scale *
 266                               cpu->power_peak);
 267 #endif
 268     while ((var = lmm_get_var_from_cnst
 269             (surf_cpu_model->model_private->maxmin_system, cpu->constraint, &elem))) {
 270       surf_action_cpu_Cas01_t action = lmm_variable_id(var);
 271       lmm_update_variable_bound(surf_cpu_model->model_private->maxmin_system,
 272                                 GENERIC_LMM_ACTION(action).variable,
 273                                 cpu->power_scale * cpu->power_peak);
 274     }
 275     if (tmgr_trace_event_free(event_type))
 276       cpu->power_event = NULL;
 277   } else if (event_type == cpu->state_event) {
 278     if (value > 0)
 279       cpu->state_current = SURF_RESOURCE_ON;
 280     else {
 281       lmm_constraint_t cnst = cpu->constraint;
 282
 283       cpu->state_current = SURF_RESOURCE_OFF;
 284
 285       while ((var = lmm_get_var_from_cnst(surf_cpu_model->model_private->maxmin_system, cnst, &elem))) {
 286         surf_action_t action = lmm_variable_id(var);
 287
 288         if (surf_action_state_get(action) == SURF_ACTION_RUNNING ||
 289             surf_action_state_get(action) == SURF_ACTION_READY ||
 290             surf_action_state_get(action) ==
 291             SURF_ACTION_NOT_IN_THE_SYSTEM) {
 292           action->finish = date;
 293           surf_action_state_set(action, SURF_ACTION_FAILED);
 294         }
 295       }
 296     }
 297     if (tmgr_trace_event_free(event_type))
 298       cpu->state_event = NULL;
 299   } else {
 300     XBT_CRITICAL("Unknown event ! \n");
 301     xbt_abort();
 302   }
 303
 304   return;
 305 }
 306
 307 static surf_action_t cpu_execute(void *cpu, double size)
 308 {
 309   surf_action_cpu_Cas01_t action = NULL;
 310   cpu_Cas01_t CPU = cpu;
 311
 312   XBT_IN("(%s,%g)", surf_resource_name(CPU), size);
 313   action =
 314       surf_action_new(sizeof(s_surf_action_cpu_Cas01_t), size,
 315                       surf_cpu_model,
 316                       CPU->state_current != SURF_RESOURCE_ON);
 317
 318   GENERIC_LMM_ACTION(action).suspended = 0;     /* Should be useless because of the
 319                                                    calloc but it seems to help valgrind... */
 320
 321   GENERIC_LMM_ACTION(action).variable =
 322       lmm_variable_new(surf_cpu_model->model_private->maxmin_system, action,
 323                        GENERIC_ACTION(action).priority,
 324                        CPU->power_scale * CPU->power_peak, 1);
 325   if (surf_cpu_model->model_private->update_mechanism == UM_LAZY) {
 326     GENERIC_LMM_ACTION(action).index_heap = -1;
 327     GENERIC_LMM_ACTION(action).last_update = surf_get_clock();
 328     GENERIC_LMM_ACTION(action).last_value = 0.0;
 329   }
 330   lmm_expand(surf_cpu_model->model_private->maxmin_system, CPU->constraint,
 331              GENERIC_LMM_ACTION(action).variable, 1.0);
 332   XBT_OUT();
 333   return (surf_action_t) action;
 334 }
 335
 336 static surf_action_t cpu_action_sleep(void *cpu, double duration)
 337 {
 338   surf_action_cpu_Cas01_t action = NULL;
 339
 340   if (duration > 0)
 341     duration = MAX(duration, MAXMIN_PRECISION);
 342
 343   XBT_IN("(%s,%g)", surf_resource_name(cpu), duration);
 344   action = (surf_action_cpu_Cas01_t) cpu_execute(cpu, 1.0);
 345   // FIXME: sleep variables should not consume 1.0 in lmm_expand
 346   GENERIC_ACTION(action).max_duration = duration;
 347   GENERIC_LMM_ACTION(action).suspended = 2;
 348   if (duration == NO_MAX_DURATION) {
 349     /* Move to the *end* of the corresponding action set. This convention
 350        is used to speed up update_resource_state  */
 351     xbt_swag_remove(action, ((surf_action_t) action)->state_set);
 352     ((surf_action_t) action)->state_set =
 353         cpu_running_action_set_that_does_not_need_being_checked;
 354     xbt_swag_insert(action, ((surf_action_t) action)->state_set);
 355   }
 356
 357   lmm_update_variable_weight(surf_cpu_model->model_private->maxmin_system,
 358                              GENERIC_LMM_ACTION(action).variable, 0.0);
 359   if (surf_cpu_model->model_private->update_mechanism == UM_LAZY) {     // remove action from the heap
 360     surf_action_lmm_heap_remove(surf_cpu_model->model_private->action_heap,(surf_action_lmm_t)action);
 361     // this is necessary for a variable with weight 0 since such
 362     // variables are ignored in lmm and we need to set its max_duration
 363     // correctly at the next call to share_resources
 364     xbt_swag_insert_at_head(action,surf_cpu_model->model_private->modified_set);
 365   }
 366
 367   XBT_OUT();
 368   return (surf_action_t) action;
 369 }
 370
 371 static e_surf_resource_state_t cpu_get_state(void *cpu)
 372 {
 373   return ((cpu_Cas01_t) cpu)->state_current;
 374 }
 375
 376 static double cpu_get_speed(void *cpu, double load)
 377 {
 378   return load * (((cpu_Cas01_t) cpu)->power_peak);
 379 }
 380
 381 static double cpu_get_available_speed(void *cpu)
 382 {
 383   /* number between 0 and 1 */
 384   return ((cpu_Cas01_t) cpu)->power_scale;
 385 }
 386
 387 static void cpu_finalize(void)
 388 {
 389   lmm_system_free(surf_cpu_model->model_private->maxmin_system);
 390   surf_cpu_model->model_private->maxmin_system = NULL;
 391
 392   if (surf_cpu_model->model_private->action_heap)
 393     xbt_heap_free(surf_cpu_model->model_private->action_heap);
 394   xbt_swag_free(surf_cpu_model->model_private->modified_set);
 395
 396   surf_model_exit(surf_cpu_model);
 397   surf_cpu_model = NULL;
 398
 399   xbt_swag_free(cpu_running_action_set_that_does_not_need_being_checked);
 400   cpu_running_action_set_that_does_not_need_being_checked = NULL;
 401 }
 402
 403 static void surf_cpu_model_init_internal()
 404 {
 405   s_surf_action_t action;
 406   s_surf_action_cpu_Cas01_t comp;
 407
 408   char *optim = xbt_cfg_get_string(_surf_cfg_set, "cpu/optim");
 409   int select =
 410       xbt_cfg_get_int(_surf_cfg_set, "cpu/maxmin_selective_update");
 411
 412   surf_cpu_model = surf_model_init();
 413
 414   if (!strcmp(optim, "Full")) {
 415     surf_cpu_model->model_private->update_mechanism = UM_FULL;
 416     surf_cpu_model->model_private->selective_update = select;
 417   } else if (!strcmp(optim, "Lazy")) {
 418     surf_cpu_model->model_private->update_mechanism = UM_LAZY;
 419     surf_cpu_model->model_private->selective_update = 1;
 420     xbt_assert((select == 1)
 421                ||
 422                (xbt_cfg_is_default_value
 423                 (_surf_cfg_set, "cpu/maxmin_selective_update")),
 424                "Disabling selective update while using the lazy update mechanism is dumb!");
 425   } else {
 426     xbt_die("Unsupported optimization (%s) for this model", optim);
 427   }
 428
 429   cpu_running_action_set_that_does_not_need_being_checked =
 430       xbt_swag_new(xbt_swag_offset(action, state_hookup));
 431
 432   surf_cpu_model->name = "cpu";
 433
 434   surf_cpu_model->action_unref = surf_action_unref;
 435   surf_cpu_model->action_cancel = surf_action_cancel;
 436   surf_cpu_model->action_state_set = surf_action_state_set;
 437
 438   surf_cpu_model->model_private->resource_used = cpu_resource_used;
 439
 440   if (surf_cpu_model->model_private->update_mechanism == UM_LAZY) {
 441     surf_cpu_model->model_private->share_resources =
 442         cpu_share_resources_lazy;
 443     surf_cpu_model->model_private->update_actions_state =
 444         cpu_update_actions_state_lazy;
 445   } else if (surf_cpu_model->model_private->update_mechanism == UM_FULL) {
 446     surf_cpu_model->model_private->share_resources =
 447         cpu_share_resources_full;
 448     surf_cpu_model->model_private->update_actions_state =
 449         cpu_update_actions_state_full;
 450   } else
 451     xbt_die("Invalid cpu update mechanism!");
 452
 453   surf_cpu_model->model_private->update_resource_state =
 454       cpu_update_resource_state;
 455   surf_cpu_model->model_private->finalize = cpu_finalize;
 456
 457   surf_cpu_model->suspend = surf_action_suspend;
 458   surf_cpu_model->resume = surf_action_resume;
 459   surf_cpu_model->is_suspended = surf_action_is_suspended;
 460   surf_cpu_model->set_max_duration = surf_action_set_max_duration;
 461   surf_cpu_model->set_priority = surf_action_set_priority;
 462 #ifdef HAVE_TRACING
 463   surf_cpu_model->set_category = surf_action_set_category;
 464 #endif
 465   surf_cpu_model->get_remains = surf_action_get_remains;
 466
 467   surf_cpu_model->extension.cpu.execute = cpu_execute;
 468   surf_cpu_model->extension.cpu.sleep = cpu_action_sleep;
 469
 470   surf_cpu_model->extension.cpu.get_state = cpu_get_state;
 471   surf_cpu_model->extension.cpu.get_speed = cpu_get_speed;
 472   surf_cpu_model->extension.cpu.get_available_speed =
 473       cpu_get_available_speed;
 474   surf_cpu_model->extension.cpu.create_resource = cpu_create_resource;
 475   surf_cpu_model->extension.cpu.add_traces = cpu_add_traces_cpu;
 476
 477   if (!surf_cpu_model->model_private->maxmin_system) {
 478     surf_cpu_model->model_private->maxmin_system = lmm_system_new(surf_cpu_model->model_private->selective_update);
 479   }
 480   if (surf_cpu_model->model_private->update_mechanism == UM_LAZY) {
 481     surf_cpu_model->model_private->action_heap = xbt_heap_new(8, NULL);
 482     xbt_heap_set_update_callback(surf_cpu_model->model_private->action_heap,
 483         surf_action_lmm_update_index_heap);
 484     surf_cpu_model->model_private->modified_set =
 485         xbt_swag_new(xbt_swag_offset(comp, generic_lmm_action.action_list_hookup));
 486     surf_cpu_model->model_private->maxmin_system->keep_track = surf_cpu_model->model_private->modified_set;
 487   }
 488 }
 489
 490 /*********************************************************************/
 491 /* Basic sharing model for CPU: that is where all this started... ;) */
 492 /*********************************************************************/
 493 /* @InProceedings{casanova01simgrid, */
 494 /*   author =       "H. Casanova", */
 495 /*   booktitle =    "Proceedings of the IEEE Symposium on Cluster Computing */
 496 /*                  and the Grid (CCGrid'01)", */
 497 /*   publisher =    "IEEE Computer Society", */
 498 /*   title =        "Simgrid: {A} Toolkit for the Simulation of Application */
 499 /*                  Scheduling", */
 500 /*   year =         "2001", */
 501 /*   month =        may, */
 502 /*   note =         "Available at */
 503 /*                  \url{http://grail.sdsc.edu/papers/simgrid_ccgrid01.ps.gz}." */
 504 /* } */
 505
 506 void surf_cpu_model_init_Cas01()
 507 {
 508   char *optim = xbt_cfg_get_string(_surf_cfg_set, "cpu/optim");
 509
 510   if (surf_cpu_model)
 511     return;
 512
 513   if (!strcmp(optim, "TI")) {
 514     surf_cpu_model_init_ti();
 515     return;
 516   }
 517
 518   surf_cpu_model_init_internal();
 519   cpu_define_callbacks();
 520   xbt_dynar_push(model_list, &surf_cpu_model);
 521 }