Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Fix super slow and meaning less code
[simgrid.git] / src / surf / vm_workstation.c
1 /* Copyright (c) 2004, 2005, 2006, 2007, 2008, 2009, 2010. 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 "xbt/ex.h"
8 #include "xbt/dict.h"
9 #include "portable.h"
10 #include "surf_private.h"
11 #include "surf/surf_resource.h"
12 #include "simgrid/sg_config.h"
13 #include "vm_workstation_private.h"
14 #include "cpu_cas01_private.h"
15 #include "maxmin_private.h"
16
17 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_vm_workstation, surf,
18                                 "Logging specific to the SURF VM workstation module");
19
20
21 surf_model_t surf_vm_workstation_model = NULL;
22
23 /* ind means ''indirect'' that this is a reference on the whole dict_elm
24  * structure (i.e not on the surf_resource_private infos) */
25
26 static void vm_ws_create(const char *name, void *ind_phys_workstation)
27 {
28   workstation_CLM03_t sub_ws = surf_workstation_resource_priv(ind_phys_workstation);
29   const char *sub_ws_name = sub_ws->generic_resource.name;
30
31   /* The workstation_VM2013 struct inherits the workstation_CLM03 struct. We
32    * create a physical workstation resource, but specifying the size of
33    * s_workstation_VM2013_t and the vm workstation model object. */
34   workstation_CLM03_t ws = (workstation_CLM03_t) surf_resource_new(sizeof(s_workstation_VM2013_t),
35       surf_vm_workstation_model, name, NULL);
36
37   /* Currently, we assume a VM has no storage. */
38   ws->storage = NULL;
39
40   /* Currently, a VM uses the network resource of its physical host. In
41    * host_lib, this network resource object is refered from two different keys.
42    * When deregistering the reference that points the network resource object
43    * from the VM name, we have to make sure that the system does not call the
44    * free callback for the network resource object. The network resource object
45    * is still used by the physical machine. */
46   ws->net_elm = xbt_lib_get_or_null(host_lib, sub_ws_name, ROUTING_HOST_LEVEL);
47   xbt_lib_set(host_lib, name, ROUTING_HOST_LEVEL, ws->net_elm);
48
49   /* The SURF_WKS_LEVEL at host_lib saves workstation_CLM03 objects. Please
50    * note workstation_VM2013 objects, inheriting the workstation_CLM03
51    * structure, are also saved there. 
52    *
53    * If you want to get a workstation_VM2013 object from host_lib, see
54    * ws->generic_resouce.model->type first. If it is
55    * SURF_MODEL_TYPE_VM_WORKSTATION, you can cast ws to vm_ws. */
56   XBT_INFO("Create VM(%s)@PM(%s) with %ld mounted disks", name, sub_ws_name, xbt_dynar_length(ws->storage));
57   xbt_lib_set(host_lib, name, SURF_WKS_LEVEL, ws);
58
59
60   /* We initialize the VM-specific members. */
61   workstation_VM2013_t vm_ws = (workstation_VM2013_t) ws;
62   vm_ws->sub_ws = sub_ws;
63   vm_ws->current_state = SURF_VM_STATE_CREATED;
64
65
66
67   // //// CPU  RELATED STUFF ////
68   // Roughly, create a vcpu resource by using the values of the sub_cpu one.
69   cpu_Cas01_t sub_cpu = surf_cpu_resource_priv(ind_phys_workstation);
70
71   /* We can assume one core and cas01 cpu for the first step.
72    * Do xbt_lib_set(host_lib, name, SURF_CPU_LEVEL, cpu) if you get the resource. */
73   cpu_cas01_create_resource(name, // name
74       sub_cpu->power_peak,        // host->power_peak,
75       1,                          // host->power_scale,
76       NULL,                       // host->power_trace,
77       1,                          // host->core_amount,
78       SURF_RESOURCE_ON,           // host->initial_state,
79       NULL,                       // host->state_trace,
80       NULL,                       // host->properties,
81       surf_cpu_model_vm);
82
83
84
85   /* We create cpu_action corresponding to a VM process on the host operating system. */
86   /* FIXME: TODO: we have to peridocally input GUESTOS_NOISE to the system? how ? */
87   // vm_ws->cpu_action = surf_cpu_model_pm->extension.cpu.execute(ind_phys_workstation, GUESTOS_NOISE);
88   vm_ws->cpu_action = surf_cpu_model_pm->extension.cpu.execute(ind_phys_workstation, 0);
89
90
91   /* TODO:
92    * - check how network requests are scheduled between distinct processes competing for the same card.
93    */
94 }
95
96 /*
97  * Update the physical host of the given VM
98  */
99 static void vm_ws_migrate(void *ind_vm, void *ind_dst_pm)
100
101    /* ind_phys_workstation equals to smx_host_t */
102    workstation_VM2013_t ws_vm2013 = surf_workstation_resource_priv(ind_vm);
103    workstation_CLM03_t ws_clm03_dst = surf_workstation_resource_priv(ind_dst_pm);
104    const char *vm_name = ws_vm2013->ws.generic_resource.name;
105    const char *pm_name_src = ws_vm2013->sub_ws->generic_resource.name;
106    const char *pm_name_dst = ws_clm03_dst->generic_resource.name;
107
108    xbt_assert(ws_vm2013);
109    xbt_assert(ws_clm03_dst);
110
111    /* do something */
112
113    /* update net_elm with that of the destination physical host */
114    void *old_net_elm = ws_vm2013->ws.net_elm;
115    void *new_net_elm = xbt_lib_get_or_null(host_lib, pm_name_dst, ROUTING_HOST_LEVEL);
116    xbt_assert(new_net_elm);
117
118    /* Unregister the current net_elm from host_lib. Do not call the free callback. */
119    xbt_lib_unset(host_lib, vm_name, ROUTING_HOST_LEVEL, 0);
120
121    /* Then, resister the new one. */
122    ws_vm2013->ws.net_elm = new_net_elm;
123    xbt_lib_set(host_lib, vm_name, ROUTING_HOST_LEVEL, ws_vm2013->ws.net_elm);
124
125    ws_vm2013->sub_ws = ws_clm03_dst;
126
127    /* Update vcpu's action for the new pm */
128    {
129 #if 0
130      XBT_INFO("cpu_action->remains %g", ws_vm2013->cpu_action->remains);
131      XBT_INFO("cost %f remains %f start %f finish %f", ws_vm2013->cpu_action->cost,
132          ws_vm2013->cpu_action->remains,
133          ws_vm2013->cpu_action->start,
134          ws_vm2013->cpu_action->finish
135          );
136      XBT_INFO("cpu_action state %d", surf_action_state_get(ws_vm2013->cpu_action));
137 #endif
138
139      /* create a cpu action bound to the pm model at the destination. */
140      surf_action_t new_cpu_action = surf_cpu_model_pm->extension.cpu.execute(ind_dst_pm, 0);
141
142      e_surf_action_state_t state = surf_action_state_get(ws_vm2013->cpu_action);
143      if (state != SURF_ACTION_DONE)
144        XBT_CRITICAL("FIXME: may need a proper handling, %d", state);
145      if (ws_vm2013->cpu_action->remains > 0)
146        XBT_CRITICAL("FIXME: need copy the state(?), %f", ws_vm2013->cpu_action->remains);
147
148      int ret = surf_cpu_model_pm->action_unref(ws_vm2013->cpu_action);
149      xbt_assert(ret == 1, "Bug: some resource still remains");
150
151      ws_vm2013->cpu_action = new_cpu_action;
152    }
153
154    XBT_DEBUG("migrate VM(%s): change net_elm (%p to %p)", vm_name, old_net_elm, new_net_elm);
155    XBT_DEBUG("migrate VM(%s): change PM (%s to %s)", vm_name, pm_name_src, pm_name_dst);
156 }
157
158 /*
159  * A physical host does not disapper in the current SimGrid code, but a VM may
160  * disapper during a simulation.
161  */
162 static void vm_ws_destroy(void *ind_vm_workstation)
163
164         /* ind_phys_workstation equals to smx_host_t */
165
166   /* Before clearing the entries in host_lib, we have to pick up resources. */
167         workstation_VM2013_t vm_ws = surf_workstation_resource_priv(ind_vm_workstation);
168   cpu_Cas01_t cpu = surf_cpu_resource_priv(ind_vm_workstation);
169         const char *name = vm_ws->ws.generic_resource.name;
170
171         xbt_assert(vm_ws);
172         xbt_assert(vm_ws->ws.generic_resource.model == surf_vm_workstation_model);
173
174
175   /* We deregister objects from host_lib, without invoking the freeing callback
176    * of each level.
177    *
178    * Do not call xbt_lib_remove() here. It deletes all levels of the key,
179    * including MSG_HOST_LEVEL and others. We should unregister only what we know.
180    */
181   xbt_lib_unset(host_lib, name, SURF_CPU_LEVEL, 0);
182   xbt_lib_unset(host_lib, name, ROUTING_HOST_LEVEL, 0);
183   xbt_lib_unset(host_lib, name, SURF_WKS_LEVEL, 0);
184
185   /* TODO: comment out when VM stroage is implemented. */
186   // xbt_lib_unset(host_lib, name, SURF_STORAGE_LEVEL, 0);
187
188
189   /* Free the cpu_action of the VM. */
190   int ret = surf_cpu_model_pm->action_unref(vm_ws->cpu_action);
191   xbt_assert(ret == 1, "Bug: some resource still remains");
192
193   /* Free the cpu resource of the VM. If using power_trace, we will have to
194    * free other objects than lmm_constraint. */
195   surf_model_t cpu_model = cpu->generic_resource.model;
196   lmm_constraint_free(cpu_model->model_private->maxmin_system, cpu->constraint);
197   surf_resource_free(cpu);
198
199   /* Free the network resource of the VM. */
200         // Nothing has to be done, because net_elmts is just a pointer on the physical one
201
202   /* Free the storage resource of the VM. */
203   // Not relevant yet
204
205         /* Free the workstation resource of the VM. */
206   surf_resource_free(vm_ws);
207 }
208
209 static int vm_ws_get_state(void *ind_vm_ws)
210 {
211         return ((workstation_VM2013_t) surf_workstation_resource_priv(ind_vm_ws))->current_state;
212 }
213
214 static void vm_ws_set_state(void *ind_vm_ws, int state)
215 {
216          ((workstation_VM2013_t) surf_workstation_resource_priv(ind_vm_ws))->current_state = state;
217 }
218
219 static void vm_ws_suspend(void *ind_vm_ws)
220 {
221   workstation_VM2013_t vm_ws = surf_workstation_resource_priv(ind_vm_ws);
222
223   surf_action_suspend(vm_ws->cpu_action);
224
225   vm_ws->current_state = SURF_VM_STATE_SUSPENDED;
226 }
227
228 static void vm_ws_resume(void *ind_vm_ws)
229 {
230   workstation_VM2013_t vm_ws = surf_workstation_resource_priv(ind_vm_ws);
231
232   surf_action_resume(vm_ws->cpu_action);
233
234   vm_ws->current_state = SURF_VM_STATE_RUNNING;
235 }
236
237 static void vm_ws_save(void *ind_vm_ws)
238 {
239   workstation_VM2013_t vm_ws = surf_workstation_resource_priv(ind_vm_ws);
240
241   vm_ws->current_state = SURF_VM_STATE_SAVING;
242
243   /* FIXME: do something here */
244   surf_action_suspend(vm_ws->cpu_action);
245
246   vm_ws->current_state = SURF_VM_STATE_SAVED;
247 }
248
249 static void vm_ws_restore(void *ind_vm_ws)
250 {
251   workstation_VM2013_t vm_ws = surf_workstation_resource_priv(ind_vm_ws);
252
253   vm_ws->current_state = SURF_VM_STATE_RESTORING;
254
255   /* FIXME: do something here */
256   surf_action_resume(vm_ws->cpu_action);
257
258   vm_ws->current_state = SURF_VM_STATE_RUNNING;
259 }
260
261 static double get_solved_value(surf_action_t cpu_action)
262 {
263   lmm_variable_t var = ((surf_action_lmm_t) cpu_action)->variable;
264
265   return var->value;
266 }
267
268 /* In the real world, processes on the guest operating system will be somewhat
269  * degraded due to virtualization overhead. The total CPU share that these
270  * processes get is smaller than that of the VM process gets on a host
271  * operating system. */
272 // const double virt_overhead = 0.95;
273 const double virt_overhead = 1;
274
275 static double vm_ws_share_resources(surf_model_t workstation_model, double now)
276 {
277   /* TODO: udpate action's cost with the total cost of processes on the VM. */
278
279
280   /* 0. Make sure that we already calculated the resource share at the physical
281    * machine layer. */
282   {
283     unsigned int index_of_pm_ws_model = xbt_dynar_search(model_list_invoke, &surf_workstation_model);
284     unsigned int index_of_vm_ws_model = xbt_dynar_search(model_list_invoke, &surf_vm_workstation_model);
285     xbt_assert((index_of_pm_ws_model < index_of_vm_ws_model), "Cannot assume surf_workstation_model comes before");
286  
287     /* Another option is that we call sub_ws->share_resource() here. The
288      * share_resource() function has no side-effect. We can call it here to
289      * ensure that. */
290   }
291
292
293   /* 1. Now we know how many resource should be assigned to each virtual
294    * machine. We update constraints of the virtual machine layer.
295    *
296    *
297    * If we have two virtual machine (VM1 and VM2) on a physical machine (PM1).
298    *     X1 + X2 = C       (Equation 1)
299    * where
300    *    the resource share of VM1: X1
301    *    the resource share of VM2: X2
302    *    the capacity of PM1: C
303    *
304    * Then, if we have two process (P1 and P2) on VM1.
305    *     X1_1 + X1_2 = X1  (Equation 2)
306    * where
307    *    the resource share of P1: X1_1
308    *    the resource share of P2: X1_2
309    *    the capacity of VM1: X1
310    *
311    * Equation 1 was solved in the physical machine layer.
312    * Equation 2 is solved in the virtual machine layer (here).
313    * X1 must be passed to the virtual machine laye as a constraint value.
314    *
315    **/
316
317   /* iterate for all hosts including virtual machines */
318   xbt_lib_cursor_t cursor;
319   char *key;
320   void **ind_host;
321   xbt_lib_foreach(host_lib, cursor, key, ind_host) {
322     workstation_CLM03_t ws_clm03 = ind_host[SURF_WKS_LEVEL];
323     cpu_Cas01_t cpu_cas01 = ind_host[SURF_CPU_LEVEL];
324
325     if (!ws_clm03)
326       continue;
327     /* skip if it is not a virtual machine */
328     if (ws_clm03->generic_resource.model != surf_vm_workstation_model)
329       continue;
330     xbt_assert(cpu_cas01, "cpu-less workstation");
331
332     /* It is a virtual machine, so we can cast it to workstation_VM2013_t */
333     workstation_VM2013_t ws_vm2013 = (workstation_VM2013_t) ws_clm03;
334
335     double solved_value = get_solved_value(ws_vm2013->cpu_action);
336     XBT_DEBUG("assign %f to vm %s @ pm %s", solved_value,
337         ws_clm03->generic_resource.name, ws_vm2013->sub_ws->generic_resource.name);
338
339     // TODO: check lmm_update_constraint_bound() works fine instead of the below manual substitution.
340     // cpu_cas01->constraint->bound = solved_value;
341     surf_model_t cpu_model = cpu_cas01->generic_resource.model;
342     xbt_assert(cpu_model == surf_cpu_model_vm);
343     lmm_system_t vcpu_system = cpu_model->model_private->maxmin_system;
344     lmm_update_constraint_bound(vcpu_system, cpu_cas01->constraint, virt_overhead * solved_value);
345   }
346
347
348   /* 2. Calculate resource share at the virtual machine layer. */
349   double ret = ws_share_resources(workstation_model, now);
350
351
352   /* FIXME: 3. do we have to re-initialize our cpu_action object? */
353 #if 0
354   /* iterate for all hosts including virtual machines */
355   xbt_lib_foreach(host_lib, cursor, key, ind_host) {
356     workstation_CLM03_t ws_clm03 = ind_host[SURF_WKS_LEVEL];
357
358     /* skip if it is not a virtual machine */
359     if (!ws_clm03)
360       continue;
361     if (ws_clm03->generic_resource.model != surf_vm_workstation_model)
362       continue;
363
364     /* It is a virtual machine, so we can cast it to workstation_VM2013_t */
365     {
366 #if 0
367       workstation_VM2013_t ws_vm2013 = (workstation_VM2013_t) ws_clm03;     
368       XBT_INFO("cost %f remains %f start %f finish %f", ws_vm2013->cpu_action->cost,
369           ws_vm2013->cpu_action->remains,
370           ws_vm2013->cpu_action->start,
371           ws_vm2013->cpu_action->finish
372           );
373 #endif
374 #if 0
375       void *ind_sub_host = xbt_lib_get_elm_or_null(host_lib, ws_vm2013->sub_ws->generic_resource.name);      
376       surf_cpu_model_pm->action_unref(ws_vm2013->cpu_action);
377       /* FIXME: this means busy loop? */
378       // ws_vm2013->cpu_action = surf_cpu_model_pm->extension.cpu.execute(ind_sub_host, GUESTOS_NOISE);
379       ws_vm2013->cpu_action = surf_cpu_model_pm->extension.cpu.execute(ind_sub_host, 0);
380 #endif
381
382     }
383   }
384 #endif
385
386
387   return ret;
388 }
389
390
391 /*
392  * A surf level object will be useless in the upper layer. Returing the
393  * dict_elm of the host.
394  **/
395 static void *vm_ws_get_pm(void *ind_vm_ws)
396 {
397         workstation_VM2013_t vm_ws = surf_workstation_resource_priv(ind_vm_ws);
398   const char *sub_ws_name = vm_ws->sub_ws->generic_resource.name;
399
400   return xbt_lib_get_elm_or_null(host_lib, sub_ws_name);
401 }
402
403
404 /* Adding a task to a VM updates the VCPU task on its physical machine. */
405 static surf_action_t vm_ws_execute(void *workstation, double size)
406 {
407   surf_resource_t ws = ((surf_resource_t) surf_workstation_resource_priv(workstation));
408
409   xbt_assert(ws->model->type == SURF_MODEL_TYPE_VM_WORKSTATION);
410   workstation_VM2013_t vm_ws = (workstation_VM2013_t) ws;
411
412   double old_cost = vm_ws->cpu_action->cost;
413   double new_cost = old_cost + size;
414
415   XBT_DEBUG("VM(%s)@PM(%s): update dummy action's cost (%f -> %f)",
416       ws->name, vm_ws->sub_ws->generic_resource.name,
417       old_cost, new_cost);
418
419   vm_ws->cpu_action->cost = new_cost;
420
421   return ws_execute(workstation, size);
422 }
423
424 static void vm_ws_action_cancel(surf_action_t action)
425 {
426   XBT_CRITICAL("FIXME: Not yet implemented. Reduce dummy action's cost by %f", action->cost);
427
428   ws_action_cancel(action);
429 }
430
431
432 /* Now we can set bound for each task by using MSG_task_set_bound. But, it does
433  * not work for the dummy CPU action of a VM. Here, we add the set_bound
434  * function for the dummy CPU action. */
435 static void vm_ws_set_vm_bound(void *workstation, double bound)
436 {
437   surf_resource_t ws = ((surf_resource_t) surf_workstation_resource_priv(workstation));
438   xbt_assert(ws->model->type == SURF_MODEL_TYPE_VM_WORKSTATION);
439   workstation_VM2013_t vm_ws = (workstation_VM2013_t) ws;
440
441   surf_action_set_bound(vm_ws->cpu_action, bound);
442 }
443
444
445 static void surf_vm_workstation_model_init_internal(void)
446 {
447   surf_model_t model = surf_model_init();
448
449   model->name = "Virtual Workstation";
450   model->type = SURF_MODEL_TYPE_VM_WORKSTATION;
451
452   model->action_unref     = ws_action_unref;
453   model->action_cancel    = vm_ws_action_cancel;
454   // model->action_state_set = ws_action_state_set;
455
456
457   model->model_private->share_resources       = vm_ws_share_resources;
458   model->model_private->resource_used         = ws_resource_used;
459   model->model_private->update_actions_state  = ws_update_actions_state;
460   model->model_private->update_resource_state = ws_update_resource_state;
461   model->model_private->finalize              = ws_finalize;
462
463
464   /* operation for an action, not for VM it self */
465   model->suspend          = ws_action_suspend;
466   model->resume           = ws_action_resume;
467 //   model->is_suspended     = ws_action_is_suspended;
468 //   model->set_max_duration = ws_action_set_max_duration;
469   model->set_priority     = ws_action_set_priority;
470   model->set_bound        = ws_action_set_bound;
471 // #ifdef HAVE_TRACING
472 //   model->set_category     = ws_action_set_category;
473 // #endif
474   model->get_remains      = ws_action_get_remains;
475 // #ifdef HAVE_LATENCY_BOUND_TRACKING
476 //   model->get_latency_limited = ws_get_latency_limited;
477 // #endif
478
479
480
481
482
483
484
485   xbt_assert(surf_cpu_model_vm);
486   model->extension.workstation.cpu_model = surf_cpu_model_vm;
487
488   model->extension.workstation.execute   = vm_ws_execute;
489   model->extension.workstation.sleep     = ws_action_sleep;
490   model->extension.workstation.get_state = ws_get_state;
491   model->extension.workstation.get_speed = ws_get_speed;
492   // model->extension.workstation.get_available_speed = ws_get_available_speed;
493
494   // model->extension.workstation.communicate           = ws_communicate;
495   // model->extension.workstation.get_route             = ws_get_route;
496   // model->extension.workstation.execute_parallel_task = ws_execute_parallel_task;
497   // model->extension.workstation.get_link_bandwidth    = ws_get_link_bandwidth;
498   // model->extension.workstation.get_link_latency      = ws_get_link_latency;
499   // model->extension.workstation.link_shared           = ws_link_shared;
500   // model->extension.workstation.get_properties        = ws_get_properties;
501
502   // model->extension.workstation.open   = ws_action_open;
503   // model->extension.workstation.close  = ws_action_close;
504   // model->extension.workstation.read   = ws_action_read;
505   // model->extension.workstation.write  = ws_action_write;
506   // model->extension.workstation.stat   = ws_action_stat;
507   // model->extension.workstation.unlink = ws_action_unlink;
508   // model->extension.workstation.ls     = ws_action_ls;
509
510
511   model->extension.vm_workstation.create        = vm_ws_create;
512   model->extension.vm_workstation.set_state     = vm_ws_set_state;
513   model->extension.vm_workstation.get_state     = vm_ws_get_state;
514   model->extension.vm_workstation.migrate       = vm_ws_migrate;
515   model->extension.vm_workstation.destroy       = vm_ws_destroy;
516   model->extension.vm_workstation.suspend       = vm_ws_suspend;
517   model->extension.vm_workstation.resume        = vm_ws_resume;
518   model->extension.vm_workstation.save          = vm_ws_save;
519   model->extension.vm_workstation.restore       = vm_ws_restore;
520   model->extension.vm_workstation.get_pm        = vm_ws_get_pm;
521   model->extension.vm_workstation.set_vm_bound  = vm_ws_set_vm_bound;
522
523   model->extension.workstation.set_params    = ws_set_params;
524   model->extension.workstation.get_params    = ws_get_params;
525
526   surf_vm_workstation_model = model;
527 }
528
529 void surf_vm_workstation_model_init(void)
530 {
531   surf_vm_workstation_model_init_internal();
532   xbt_dynar_push(model_list, &surf_vm_workstation_model);
533   xbt_dynar_push(model_list_invoke, &surf_vm_workstation_model);
534 }