Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Fixed full routing support for routers, no routing complexity is added routers are...
[simgrid.git] / src / surf / network.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 "network_private.h"
9 #include "xbt/log.h"
10 #include "xbt/str.h"
11
12 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_network, surf,
13                                 "Logging specific to the SURF network module");
14
15 surf_model_t surf_network_model = NULL;
16 static lmm_system_t network_maxmin_system = NULL;
17 static void (*network_solve) (lmm_system_t) = NULL;
18
19 double latency_factor = 1.0;    /* default value */
20 double bandwidth_factor = 1.0;  /* default value */
21 double weight_S_parameter = 0.0;        /* default value */
22
23 double sg_tcp_gamma = 0.0;
24
25
26 static link_CM02_t link_new(char *name,
27                             double bw_initial,
28                             tmgr_trace_t bw_trace,
29                             double lat_initial,
30                             tmgr_trace_t lat_trace,
31                             e_surf_resource_state_t
32                             state_initial,
33                             tmgr_trace_t state_trace,
34                             e_surf_link_sharing_policy_t
35                             policy, xbt_dict_t properties)
36 {
37   link_CM02_t nw_link = (link_CM02_t)
38     surf_resource_lmm_new(sizeof(s_link_CM02_t),
39                           surf_network_model, name, properties,
40                           network_maxmin_system,
41                           bandwidth_factor * bw_initial,
42                           history,
43                           state_initial, state_trace,
44                           bw_initial, bw_trace);
45
46   xbt_assert1(!xbt_dict_get_or_null(surf_network_model->resource_set, name),
47               "Link '%s' declared several times in the platform file.", name);
48
49   nw_link->lat_current = lat_initial;
50   if (lat_trace)
51     nw_link->lat_event =
52       tmgr_history_add_trace(history, lat_trace, 0.0, 0, nw_link);
53
54   if (policy == SURF_LINK_FATPIPE)
55     lmm_constraint_shared(nw_link->lmm_resource.constraint);
56
57   xbt_dict_set(surf_network_model->resource_set, name, nw_link,
58                surf_resource_free);
59
60   return nw_link;
61 }
62
63 static void parse_link_init(void)
64 {
65   char *name_link;
66   double bw_initial;
67   tmgr_trace_t bw_trace;
68   double lat_initial;
69   tmgr_trace_t lat_trace;
70   e_surf_resource_state_t state_initial_link = SURF_RESOURCE_ON;
71   e_surf_link_sharing_policy_t policy_initial_link = SURF_LINK_SHARED;
72   tmgr_trace_t state_trace;
73
74   name_link = xbt_strdup(A_surfxml_link_id);
75   surf_parse_get_double(&bw_initial, A_surfxml_link_bandwidth);
76   bw_trace = tmgr_trace_new(A_surfxml_link_bandwidth_file);
77   surf_parse_get_double(&lat_initial, A_surfxml_link_latency);
78   lat_trace = tmgr_trace_new(A_surfxml_link_latency_file);
79
80   xbt_assert0((A_surfxml_link_state == A_surfxml_link_state_ON)
81               || (A_surfxml_link_state ==
82                   A_surfxml_link_state_OFF), "Invalid state");
83   if (A_surfxml_link_state == A_surfxml_link_state_ON)
84     state_initial_link = SURF_RESOURCE_ON;
85   else if (A_surfxml_link_state == A_surfxml_link_state_OFF)
86     state_initial_link = SURF_RESOURCE_OFF;
87
88   if (A_surfxml_link_sharing_policy == A_surfxml_link_sharing_policy_SHARED)
89     policy_initial_link = SURF_LINK_SHARED;
90   else if (A_surfxml_link_sharing_policy ==
91            A_surfxml_link_sharing_policy_FATPIPE)
92     policy_initial_link = SURF_LINK_FATPIPE;
93
94   state_trace = tmgr_trace_new(A_surfxml_link_state_file);
95
96   link_new(name_link, bw_initial, bw_trace,
97            lat_initial, lat_trace, state_initial_link, state_trace,
98            policy_initial_link, xbt_dict_new());
99
100 }
101
102 static void add_traces(void)
103 {
104   xbt_dict_cursor_t cursor = NULL;
105   char *trace_name, *elm;
106
107   static int called = 0;
108   if (called)
109     return;
110   called = 1;
111
112   /* connect all traces relative to network */
113   xbt_dict_foreach(trace_connect_list_link_avail, cursor, trace_name, elm) {
114     tmgr_trace_t trace = xbt_dict_get_or_null(traces_set_list, trace_name);
115     link_CM02_t link =
116       xbt_dict_get_or_null(surf_network_model->resource_set, elm);
117
118     xbt_assert2(link, "Cannot connect trace %s to link %s: link undefined",
119                 trace_name, elm);
120     xbt_assert2(trace, "Cannot connect trace %s to link %s: trace undefined",
121                 trace_name, elm);
122
123     link->lmm_resource.state_event =
124       tmgr_history_add_trace(history, trace, 0.0, 0, link);
125   }
126
127   xbt_dict_foreach(trace_connect_list_bandwidth, cursor, trace_name, elm) {
128     tmgr_trace_t trace = xbt_dict_get_or_null(traces_set_list, trace_name);
129     link_CM02_t link =
130       xbt_dict_get_or_null(surf_network_model->resource_set, elm);
131
132     xbt_assert2(link, "Cannot connect trace %s to link %s: link undefined",
133                 trace_name, elm);
134     xbt_assert2(trace, "Cannot connect trace %s to link %s: trace undefined",
135                 trace_name, elm);
136
137     link->lmm_resource.power.event =
138       tmgr_history_add_trace(history, trace, 0.0, 0, link);
139   }
140
141   xbt_dict_foreach(trace_connect_list_latency, cursor, trace_name, elm) {
142     tmgr_trace_t trace = xbt_dict_get_or_null(traces_set_list, trace_name);
143     link_CM02_t link =
144       xbt_dict_get_or_null(surf_network_model->resource_set, elm);
145
146     xbt_assert2(link, "Cannot connect trace %s to link %s: link undefined",
147                 trace_name, elm);
148     xbt_assert2(trace, "Cannot connect trace %s to link %s: trace undefined",
149                 trace_name, elm);
150
151     link->lat_event = tmgr_history_add_trace(history, trace, 0.0, 0, link);
152   }
153 }
154
155 static void define_callbacks(const char *file)
156 {
157   /* Figuring out the network links */
158   surfxml_add_callback(STag_surfxml_link_cb_list, &parse_link_init);
159   surfxml_add_callback(ETag_surfxml_platform_cb_list, &add_traces);
160 }
161
162 static int resource_used(void *resource_id)
163 {
164   return lmm_constraint_used(network_maxmin_system,
165                              ((surf_resource_lmm_t) resource_id)->constraint);
166 }
167
168 static int action_unref(surf_action_t action)
169 {
170   action->refcount--;
171   if (!action->refcount) {
172     xbt_swag_remove(action, action->state_set);
173     if (((surf_action_network_CM02_t) action)->variable)
174       lmm_variable_free(network_maxmin_system,
175                         ((surf_action_network_CM02_t) action)->variable);
176     free(action);
177     return 1;
178   }
179   return 0;
180 }
181
182 static void action_cancel(surf_action_t action)
183 {
184   return;
185 }
186
187 static void action_recycle(surf_action_t action)
188 {
189   return;
190 }
191
192 static double action_get_remains(surf_action_t action)
193 {
194   return action->remains;
195 }
196
197 static double share_resources(double now)
198 {
199   s_surf_action_network_CM02_t s_action;
200   surf_action_network_CM02_t action = NULL;
201   xbt_swag_t running_actions = surf_network_model->states.running_action_set;
202   double min;
203
204   min = generic_maxmin_share_resources(running_actions,
205                                        xbt_swag_offset(s_action,
206                                                        variable),
207                                        network_maxmin_system, network_solve);
208
209 #define VARIABLE(action) (*((lmm_variable_t*)(((char *) (action)) + xbt_swag_offset(s_action, variable)  )))
210
211   xbt_swag_foreach(action, running_actions) {
212     if (action->latency > 0) {
213       if (min < 0)
214         min = action->latency;
215       else if (action->latency < min)
216         min = action->latency;
217     }
218   }
219
220   return min;
221 }
222
223 static void update_actions_state(double now, double delta)
224 {
225   double deltap = 0.0;
226   surf_action_network_CM02_t action = NULL;
227   surf_action_network_CM02_t next_action = NULL;
228   xbt_swag_t running_actions = surf_network_model->states.running_action_set;
229   /*
230      xbt_swag_t failed_actions =
231      surf_network_model->states.failed_action_set;
232    */
233
234   xbt_swag_foreach_safe(action, next_action, running_actions) {
235     deltap = delta;
236     if (action->latency > 0) {
237       if (action->latency > deltap) {
238         double_update(&(action->latency), deltap);
239         deltap = 0.0;
240       } else {
241         double_update(&(deltap), action->latency);
242         action->latency = 0.0;
243       }
244       if ((action->latency == 0.0) && !(action->suspended))
245         lmm_update_variable_weight(network_maxmin_system, action->variable,
246                                    action->weight);
247     }
248     double_update(&(action->generic_action.remains),
249                   lmm_variable_getvalue(action->variable) * deltap);
250     if (action->generic_action.max_duration != NO_MAX_DURATION)
251       double_update(&(action->generic_action.max_duration), delta);
252
253     if ((action->generic_action.remains <= 0) &&
254         (lmm_get_variable_weight(action->variable) > 0)) {
255       action->generic_action.finish = surf_get_clock();
256       surf_network_model->action_state_set((surf_action_t) action,
257                                            SURF_ACTION_DONE);
258     } else if ((action->generic_action.max_duration != NO_MAX_DURATION)
259                && (action->generic_action.max_duration <= 0)) {
260       action->generic_action.finish = surf_get_clock();
261       surf_network_model->action_state_set((surf_action_t) action,
262                                            SURF_ACTION_DONE);
263     }
264   }
265
266   return;
267 }
268
269 static void update_resource_state(void *id,
270                                   tmgr_trace_event_t event_type,
271                                   double value, double date)
272 {
273   link_CM02_t nw_link = id;
274   /*   printf("[" "%lg" "] Asking to update network card \"%s\" with value " */
275   /*     "%lg" " for event %p\n", surf_get_clock(), nw_link->name, */
276   /*     value, event_type); */
277
278   if (event_type == nw_link->lmm_resource.power.event) {
279     double delta =
280       weight_S_parameter / value - weight_S_parameter /
281       (nw_link->lmm_resource.power.peak * nw_link->lmm_resource.power.scale);
282     lmm_variable_t var = NULL;
283     lmm_element_t elem = NULL;
284     surf_action_network_CM02_t action = NULL;
285
286     nw_link->lmm_resource.power.peak = value;
287     lmm_update_constraint_bound(network_maxmin_system,
288                                 nw_link->lmm_resource.constraint,
289                                 bandwidth_factor *
290                                 (nw_link->lmm_resource.power.peak *
291                                  nw_link->lmm_resource.power.scale));
292     if (weight_S_parameter > 0) {
293       while ((var = lmm_get_var_from_cnst
294               (network_maxmin_system, nw_link->lmm_resource.constraint,
295                &elem))) {
296         action = lmm_variable_id(var);
297         action->weight += delta;
298         if (!(action->suspended))
299           lmm_update_variable_weight(network_maxmin_system, action->variable,
300                                      action->weight);
301       }
302     }
303     if (tmgr_trace_event_free(event_type))
304       nw_link->lmm_resource.power.event = NULL;
305   } else if (event_type == nw_link->lat_event) {
306     double delta = value - nw_link->lat_current;
307     lmm_variable_t var = NULL;
308     lmm_element_t elem = NULL;
309     surf_action_network_CM02_t action = NULL;
310
311     nw_link->lat_current = value;
312     while ((var = lmm_get_var_from_cnst
313             (network_maxmin_system, nw_link->lmm_resource.constraint,
314              &elem))) {
315       action = lmm_variable_id(var);
316       action->lat_current += delta;
317       action->weight += delta;
318       if (action->rate < 0)
319         lmm_update_variable_bound(network_maxmin_system, action->variable,
320                                   sg_tcp_gamma / (2.0 * action->lat_current));
321       else
322         lmm_update_variable_bound(network_maxmin_system, action->variable,
323                                   min(action->rate,
324                                       sg_tcp_gamma / (2.0 *
325                                                       action->lat_current)));
326       if (!(action->suspended))
327         lmm_update_variable_weight(network_maxmin_system, action->variable,
328                                    action->weight);
329
330     }
331     if (tmgr_trace_event_free(event_type))
332       nw_link->lat_event = NULL;
333   } else if (event_type == nw_link->lmm_resource.state_event) {
334     if (value > 0)
335       nw_link->lmm_resource.state_current = SURF_RESOURCE_ON;
336     else {
337       lmm_constraint_t cnst = nw_link->lmm_resource.constraint;
338       lmm_variable_t var = NULL;
339       lmm_element_t elem = NULL;
340
341       nw_link->lmm_resource.state_current = SURF_RESOURCE_OFF;
342       while ((var = lmm_get_var_from_cnst
343               (network_maxmin_system, cnst, &elem))) {
344         surf_action_t action = lmm_variable_id(var);
345
346         if (surf_action_state_get(action) == SURF_ACTION_RUNNING ||
347             surf_action_state_get(action) == SURF_ACTION_READY) {
348           action->finish = date;
349           surf_network_model->action_state_set(action, SURF_ACTION_FAILED);
350         }
351       }
352     }
353     if (tmgr_trace_event_free(event_type))
354       nw_link->lmm_resource.state_event = NULL;
355   } else {
356     CRITICAL0("Unknown event ! \n");
357     xbt_abort();
358   }
359
360   return;
361 }
362
363 static surf_action_t communicate(const char *src_name, const char *dst_name,
364                                  int src, int dst, double size, double rate)
365 {
366   surf_action_network_CM02_t action = NULL;
367   /* LARGE PLATFORMS HACK:
368      Add a link_CM02_t *link and a int link_nb to network_card_CM02_t. It will represent local links for this node
369      Use the cluster_id for ->id */
370   xbt_dynar_t route = used_routing->get_route(src, dst);
371   /* LARGE PLATFORMS HACK:
372      total_route_size = route_size + src->link_nb + dst->nb */
373   unsigned int i;
374
375   XBT_IN4("(%s,%s,%g,%g)", src_name, dst_name, size, rate);
376   /* LARGE PLATFORMS HACK:
377      assert on total_route_size */
378   xbt_assert2(xbt_dynar_length(route),
379               "You're trying to send data from %s to %s but there is no connection between these two hosts.",
380               src_name, dst_name);
381
382   link_CM02_t link;
383   int failed = 0;
384   xbt_dynar_foreach(route, i, link) {
385     if (link->lmm_resource.state_current == SURF_RESOURCE_OFF) {
386       failed = 1;
387       break;
388     }
389   }
390   action =
391     surf_action_new(sizeof(s_surf_action_network_CM02_t), size,
392                     surf_network_model, failed);
393
394   xbt_swag_insert(action, action->generic_action.state_set);
395   action->rate = rate;
396
397   action->latency = 0.0;
398   action->weight = 0.0;
399   xbt_dynar_foreach(route, i, link) {
400     action->latency += link->lat_current;
401     action->weight +=
402       link->lat_current +
403       weight_S_parameter /
404       (link->lmm_resource.power.peak * link->lmm_resource.power.scale);
405   }
406   /* LARGE PLATFORMS HACK:
407      Add src->link and dst->link latencies */
408   action->lat_current = action->latency;
409   action->latency *= latency_factor;
410
411   /* LARGE PLATFORMS HACK:
412      lmm_variable_new(..., total_route_size) */
413   if (action->latency > 0)
414     action->variable =
415       lmm_variable_new(network_maxmin_system, action, 0.0, -1.0,
416                        xbt_dynar_length(route));
417   else
418     action->variable =
419       lmm_variable_new(network_maxmin_system, action, 1.0, -1.0,
420                        xbt_dynar_length(route));
421
422   if (action->rate < 0) {
423     if (action->lat_current > 0)
424       lmm_update_variable_bound(network_maxmin_system, action->variable,
425                                 sg_tcp_gamma / (2.0 * action->lat_current));
426     else
427       lmm_update_variable_bound(network_maxmin_system, action->variable,
428                                 -1.0);
429   } else {
430     if (action->lat_current > 0)
431       lmm_update_variable_bound(network_maxmin_system, action->variable,
432                                 min(action->rate,
433                                     sg_tcp_gamma / (2.0 *
434                                                     action->lat_current)));
435     else
436       lmm_update_variable_bound(network_maxmin_system, action->variable,
437                                 action->rate);
438   }
439
440   xbt_dynar_foreach(route, i, link) {
441     lmm_expand(network_maxmin_system, link->lmm_resource.constraint,
442                action->variable, 1.0);
443   }
444   /* LARGE PLATFORMS HACK:
445      expand also with src->link and dst->link */
446
447   XBT_OUT;
448
449   return (surf_action_t) action;
450 }
451
452 static double get_link_bandwidth(const void *link)
453 {
454   surf_resource_lmm_t lmm = (surf_resource_lmm_t) link;
455   return lmm->power.peak * lmm->power.scale;
456 }
457
458 static double get_link_latency(const void *link)
459 {
460   return ((link_CM02_t) link)->lat_current;
461 }
462
463 static int link_shared(const void *link)
464 {
465   return lmm_constraint_is_shared(((surf_resource_lmm_t) link)->constraint);
466 }
467
468 static void action_suspend(surf_action_t action)
469 {
470   ((surf_action_network_CM02_t) action)->suspended = 1;
471   lmm_update_variable_weight(network_maxmin_system,
472                              ((surf_action_network_CM02_t) action)->variable,
473                              0.0);
474 }
475
476 static void action_resume(surf_action_t action)
477 {
478   if (((surf_action_network_CM02_t) action)->suspended) {
479     lmm_update_variable_weight(network_maxmin_system,
480                                ((surf_action_network_CM02_t)
481                                 action)->variable,
482                                ((surf_action_network_CM02_t) action)->weight);
483     ((surf_action_network_CM02_t) action)->suspended = 0;
484   }
485 }
486
487 static int action_is_suspended(surf_action_t action)
488 {
489   return ((surf_action_network_CM02_t) action)->suspended;
490 }
491
492 static void action_set_max_duration(surf_action_t action, double duration)
493 {
494   action->max_duration = duration;
495 }
496
497 static void finalize(void)
498 {
499   surf_model_exit(surf_network_model);
500   surf_network_model = NULL;
501
502   used_routing->finalize();
503   lmm_system_free(network_maxmin_system);
504   network_maxmin_system = NULL;
505 }
506
507 static void surf_network_model_init_internal(void)
508 {
509   surf_network_model = surf_model_init();
510
511   surf_network_model->name = "network";
512   surf_network_model->action_unref = action_unref;
513   surf_network_model->action_cancel = action_cancel;
514   surf_network_model->action_recycle = action_recycle;
515   surf_network_model->get_remains = action_get_remains;
516
517   surf_network_model->model_private->resource_used = resource_used;
518   surf_network_model->model_private->share_resources = share_resources;
519   surf_network_model->model_private->update_actions_state =
520     update_actions_state;
521   surf_network_model->model_private->update_resource_state =
522     update_resource_state;
523   surf_network_model->model_private->finalize = finalize;
524
525   surf_network_model->suspend = action_suspend;
526   surf_network_model->resume = action_resume;
527   surf_network_model->is_suspended = action_is_suspended;
528   surf_cpu_model->set_max_duration = action_set_max_duration;
529
530   surf_network_model->extension.network.communicate = communicate;
531   surf_network_model->extension.network.get_link_bandwidth =
532     get_link_bandwidth;
533   surf_network_model->extension.network.get_link_latency = get_link_latency;
534   surf_network_model->extension.network.link_shared = link_shared;
535
536   if (!network_maxmin_system)
537     network_maxmin_system = lmm_system_new();
538
539   routing_model_create(sizeof(link_CM02_t),
540                        link_new(xbt_strdup("__loopback__"),
541                                 498000000, NULL, 0.000015, NULL,
542                                 SURF_RESOURCE_ON, NULL, SURF_LINK_FATPIPE,
543                                 NULL));
544 }
545
546 /************************************************************************/
547 /* New model based on optimizations discussed during this thesis        */
548 /************************************************************************/
549 void surf_network_model_init_LegrandVelho(const char *filename)
550 {
551
552   if (surf_network_model)
553     return;
554   surf_network_model_init_internal();
555   define_callbacks(filename);
556   xbt_dynar_push(model_list, &surf_network_model);
557   network_solve = lmm_solve;
558
559   latency_factor = 10.4;
560   bandwidth_factor = 0.92;
561   weight_S_parameter = 8775;
562
563   update_model_description(surf_network_model_description,
564                            "LV08", surf_network_model);
565 }
566
567 /***************************************************************************/
568 /* The nice TCP sharing model designed by Loris Marchal and Henri Casanova */
569 /***************************************************************************/
570 /* @TechReport{      rr-lip2002-40, */
571 /*   author        = {Henri Casanova and Loris Marchal}, */
572 /*   institution   = {LIP}, */
573 /*   title         = {A Network Model for Simulation of Grid Application}, */
574 /*   number        = {2002-40}, */
575 /*   month         = {oct}, */
576 /*   year          = {2002} */
577 /* } */
578 void surf_network_model_init_CM02(const char *filename)
579 {
580
581   if (surf_network_model)
582     return;
583   surf_network_model_init_internal();
584   define_callbacks(filename);
585   xbt_dynar_push(model_list, &surf_network_model);
586   network_solve = lmm_solve;
587
588   update_model_description(surf_network_model_description,
589                            "CM02", surf_network_model);
590 }
591
592 void surf_network_model_init_Reno(const char *filename)
593 {
594   if (surf_network_model)
595     return;
596   surf_network_model_init_internal();
597   define_callbacks(filename);
598
599   xbt_dynar_push(model_list, &surf_network_model);
600   lmm_set_default_protocol_function(func_reno_f, func_reno_fp, func_reno_fpi);
601   network_solve = lagrange_solve;
602
603   latency_factor = 10.4;
604   bandwidth_factor = 0.92;
605   weight_S_parameter = 8775;
606
607   update_model_description(surf_network_model_description,
608                            "Reno", surf_network_model);
609 }
610
611
612 void surf_network_model_init_Reno2(const char *filename)
613 {
614   if (surf_network_model)
615     return;
616   surf_network_model_init_internal();
617   define_callbacks(filename);
618
619   xbt_dynar_push(model_list, &surf_network_model);
620   lmm_set_default_protocol_function(func_reno2_f, func_reno2_fp,
621                                     func_reno2_fpi);
622   network_solve = lagrange_solve;
623
624   latency_factor = 10.4;
625   bandwidth_factor = 0.92;
626   weight_S_parameter = 8775;
627
628   update_model_description(surf_network_model_description,
629                            "Reno2", surf_network_model);
630 }
631
632 void surf_network_model_init_Vegas(const char *filename)
633 {
634   if (surf_network_model)
635     return;
636   surf_network_model_init_internal();
637   define_callbacks(filename);
638
639   xbt_dynar_push(model_list, &surf_network_model);
640   lmm_set_default_protocol_function(func_vegas_f, func_vegas_fp,
641                                     func_vegas_fpi);
642   network_solve = lagrange_solve;
643
644   latency_factor = 10.4;
645   bandwidth_factor = 0.92;
646   weight_S_parameter = 8775;
647
648   update_model_description(surf_network_model_description,
649                            "Vegas", surf_network_model);
650 }