Logo AND Algorithmique Numérique Distribuée

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