1 /* Copyright (c) 2004-2017. The SimGrid Team. All rights reserved. */
3 /* This program is free software; you can redistribute it and/or modify it
4 * under the terms of the license (GNU LGPL) which comes with this package. */
6 #include "surf_interface.hpp"
8 #include "simgrid/s4u/Engine.hpp"
9 #include "simgrid/sg_config.h"
10 #include "src/instr/instr_private.hpp" // TRACE_is_enabled(). FIXME: remove by subscribing tracing to the surf signals
11 #include "src/kernel/routing/NetPoint.hpp"
12 #include "src/surf/HostImpl.hpp"
19 XBT_LOG_NEW_CATEGORY(surf, "All SURF categories");
20 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_kernel, surf, "Logging specific to SURF (kernel)");
26 std::vector<surf_model_t> * all_existing_models = nullptr; /* to destroy models correctly */
28 simgrid::trace_mgr::future_evt_set *future_evt_set = nullptr;
29 std::vector<std::string> surf_path;
30 std::vector<simgrid::s4u::Host*> host_that_restart;
31 /** set of hosts for which one want to be notified if they ever restart. */
32 std::set<std::string> watched_hosts;
33 extern std::map<std::string, simgrid::surf::StorageType*> storage_types;
38 simgrid::xbt::signal<void()> surfExitCallbacks;
42 #include <simgrid/plugins/energy.h> // FIXME: this plug-in should not be linked to the core
43 #include <simgrid/plugins/load.h> // FIXME: this plug-in should not be linked to the core
45 s_surf_model_description_t surf_plugin_description[] = {
46 {"Energy", "Cpu energy consumption.", &sg_host_energy_plugin_init},
47 {"Load", "Cpu load.", &sg_host_load_plugin_init},
48 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
51 /* Don't forget to update the option description in smx_config when you change this */
52 s_surf_model_description_t surf_network_model_description[] = {
53 {"LV08", "Realistic network analytic model (slow-start modeled by multiplying latency by 13.01, bandwidth by .97; "
54 "bottleneck sharing uses a payload of S=20537 for evaluating RTT). ",
55 &surf_network_model_init_LegrandVelho},
56 {"Constant", "Simplistic network model where all communication take a constant time (one second). This model "
57 "provides the lowest realism, but is (marginally) faster.",
58 &surf_network_model_init_Constant},
59 {"SMPI", "Realistic network model specifically tailored for HPC settings (accurate modeling of slow start with "
60 "correction factors on three intervals: < 1KiB, < 64 KiB, >= 64 KiB)",
61 &surf_network_model_init_SMPI},
62 {"IB", "Realistic network model specifically tailored for HPC settings, with Infiniband contention model",
63 &surf_network_model_init_IB},
64 {"CM02", "Legacy network analytic model (Very similar to LV08, but without corrective factors. The timings of "
65 "small messages are thus poorly modeled).",
66 &surf_network_model_init_CM02},
67 {"NS3", "Network pseudo-model using the NS3 tcp model instead of an analytic model", &surf_network_model_init_NS3},
69 "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
70 &surf_network_model_init_Reno},
72 "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
73 &surf_network_model_init_Reno2},
75 "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
76 &surf_network_model_init_Vegas},
77 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
81 void surf_network_model_init_SMPI() {
82 xbt_die("Please activate SMPI support in cmake to use the SMPI network model.");
84 void surf_network_model_init_IB() {
85 xbt_die("Please activate SMPI support in cmake to use the IB network model.");
89 void surf_network_model_init_NS3() {
90 xbt_die("Please activate NS3 support in cmake and install the dependencies to use the NS3 network model.");
94 s_surf_model_description_t surf_cpu_model_description[] = {
95 {"Cas01", "Simplistic CPU model (time=size/power).", &surf_cpu_model_init_Cas01},
96 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
99 s_surf_model_description_t surf_host_model_description[] = {
100 {"default", "Default host model. Currently, CPU:Cas01 and network:LV08 (with cross traffic enabled)", &surf_host_model_init_current_default},
101 {"compound", "Host model that is automatically chosen if you change the network and CPU models", &surf_host_model_init_compound},
102 {"ptask_L07", "Host model somehow similar to Cas01+CM02 but allowing parallel tasks", &surf_host_model_init_ptask_L07},
103 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
106 s_surf_model_description_t surf_optimization_mode_description[] = {
107 {"Lazy", "Lazy action management (partial invalidation in lmm + heap in action remaining).", nullptr},
108 {"TI", "Trace integration. Highly optimized mode when using availability traces (only available for the Cas01 CPU model for now).", nullptr},
109 {"Full", "Full update of remaining and variables. Slow but may be useful when debugging.", nullptr},
110 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
113 s_surf_model_description_t surf_storage_model_description[] = {
114 {"default", "Simplistic storage model.", &surf_storage_model_init_default},
115 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
120 double surf_get_clock()
126 # define FILE_DELIM "\\"
128 # define FILE_DELIM "/" /* FIXME: move to better location */
131 std::ifstream* surf_ifsopen(std::string name)
133 std::ifstream* fs = new std::ifstream();
134 xbt_assert(not name.empty());
135 if (__surf_is_absolute_file_path(name.c_str())) { /* don't mess with absolute file names */
136 fs->open(name.c_str(), std::ifstream::in);
139 /* search relative files in the path */
140 for (auto const& path_elm : surf_path) {
141 std::string buff = path_elm + FILE_DELIM + name;
142 fs->open(buff.c_str(), std::ifstream::in);
144 if (not fs->fail()) {
145 XBT_DEBUG("Found file at %s", buff.c_str());
153 FILE *surf_fopen(const char *name, const char *mode)
155 FILE *file = nullptr;
159 if (__surf_is_absolute_file_path(name)) /* don't mess with absolute file names */
160 return fopen(name, mode);
162 /* search relative files in the path */
163 for (auto const& path_elm : surf_path) {
164 std::string buff = path_elm + FILE_DELIM + name;
165 file = fopen(buff.c_str(), mode);
173 /* The __surf_is_absolute_file_path() returns 1 if
174 * file_path is a absolute file path, in the other
175 * case the function returns 0.
177 int __surf_is_absolute_file_path(const char *file_path)
180 WIN32_FIND_DATA wfd = { 0 };
181 HANDLE hFile = FindFirstFile(file_path, &wfd);
183 if (INVALID_HANDLE_VALUE == hFile)
189 return (file_path[0] == '/');
193 /** Displays the long description of all registered models, and quit */
194 void model_help(const char *category, s_surf_model_description_t * table)
196 printf("Long description of the %s models accepted by this simulator:\n", category);
197 for (int i = 0; table[i].name; i++)
198 printf(" %s: %s\n", table[i].name, table[i].description);
201 int find_model_description(s_surf_model_description_t* table, std::string name)
203 for (int i = 0; table[i].name; i++)
204 if (name == table[i].name)
207 if (not table[0].name)
208 xbt_die("No model is valid! This is a bug.");
210 std::string name_list = std::string(table[0].name);
211 for (int i = 1; table[i].name; i++)
212 name_list = name_list + ", " + table[i].name;
214 xbt_die("Model '%s' is invalid! Valid models are: %s.", name.c_str(), name_list.c_str());
218 void sg_version_check(int lib_version_major, int lib_version_minor, int lib_version_patch)
220 if ((lib_version_major != SIMGRID_VERSION_MAJOR) || (lib_version_minor != SIMGRID_VERSION_MINOR)) {
221 fprintf(stderr, "FATAL ERROR: Your program was compiled with SimGrid version %d.%d.%d, "
222 "and then linked against SimGrid %d.%d.%d. Please fix this.\n",
223 lib_version_major, lib_version_minor, lib_version_patch, SIMGRID_VERSION_MAJOR, SIMGRID_VERSION_MINOR,
224 SIMGRID_VERSION_PATCH);
227 if (lib_version_patch != SIMGRID_VERSION_PATCH) {
228 if (SIMGRID_VERSION_PATCH >= 90 || lib_version_patch >= 90) {
231 "FATAL ERROR: Your program was compiled with SimGrid version %d.%d.%d, "
232 "and then linked against SimGrid %d.%d.%d. \n"
233 "One of them is a development version, and should not be mixed with the stable release. Please fix this.\n",
234 lib_version_major, lib_version_minor, lib_version_patch, SIMGRID_VERSION_MAJOR, SIMGRID_VERSION_MINOR,
235 SIMGRID_VERSION_PATCH);
238 fprintf(stderr, "Warning: Your program was compiled with SimGrid version %d.%d.%d, "
239 "and then linked against SimGrid %d.%d.%d. Proceeding anyway.\n",
240 lib_version_major, lib_version_minor, lib_version_patch, SIMGRID_VERSION_MAJOR, SIMGRID_VERSION_MINOR,
241 SIMGRID_VERSION_PATCH);
245 void sg_version_get(int* ver_major, int* ver_minor, int* ver_patch)
247 *ver_major = SIMGRID_VERSION_MAJOR;
248 *ver_minor = SIMGRID_VERSION_MINOR;
249 *ver_patch = SIMGRID_VERSION_PATCH;
254 std::printf("This program was linked against %s (git: %s), found in %s.\n",
255 SIMGRID_VERSION_STRING, SIMGRID_GIT_VERSION, SIMGRID_INSTALL_PREFIX);
258 std::printf(" Model-checking support compiled in.\n");
260 std::printf(" Model-checking support disabled at compilation.\n");
264 std::printf(" NS3 support compiled in.\n");
266 std::printf(" NS3 support disabled at compilation.\n");
269 #if SIMGRID_HAVE_JEDULE
270 std::printf(" Jedule support compiled in.\n");
272 std::printf(" Jedule support disabled at compilation.\n");
276 std::printf(" Lua support compiled in.\n");
278 std::printf(" Lua support disabled at compilation.\n");
281 #if SIMGRID_HAVE_MALLOCATOR
282 std::printf(" Mallocator support compiled in.\n");
284 std::printf(" Mallocator support disabled at compilation.\n");
287 std::printf("\nTo cite SimGrid in a publication, please use:\n"
288 " Henri Casanova, Arnaud Giersch, Arnaud Legrand, Martin Quinson, Frédéric Suter. \n"
289 " Versatile, Scalable, and Accurate Simulation of Distributed Applications and Platforms. \n"
290 " Journal of Parallel and Distributed Computing, Elsevier, 2014, 74 (10), pp.2899-2917.\n");
291 std::printf("The pdf file and a BibTeX entry for LaTeX users can be found at http://hal.inria.fr/hal-01017319\n");
294 void surf_init(int *argc, char **argv)
296 if (USER_HOST_LEVEL != -1) // Already initialized
299 XBT_DEBUG("Create all Libs");
300 USER_HOST_LEVEL = simgrid::s4u::Host::extension_create(nullptr);
302 xbt_init(argc, argv);
303 if (not all_existing_models)
304 all_existing_models = new std::vector<simgrid::surf::Model*>();
305 if (not future_evt_set)
306 future_evt_set = new simgrid::trace_mgr::future_evt_set();
308 sg_config_init(argc, argv);
316 TRACE_end(); /* Just in case it was not called by the upper layer (or there is no upper layer) */
320 for (auto const& e : storage_types) {
321 simgrid::surf::StorageType* stype = e.second;
322 delete stype->properties;
323 delete stype->model_properties;
326 for (auto const& s : *simgrid::surf::StorageImpl::storagesMap())
328 delete simgrid::surf::StorageImpl::storagesMap();
330 for (auto const& model : *all_existing_models)
332 delete all_existing_models;
334 simgrid::surf::surfExitCallbacks();
336 if (future_evt_set) {
337 delete future_evt_set;
338 future_evt_set = nullptr;
343 simgrid::s4u::Engine::shutdown();
345 NOW = 0; /* Just in case the user plans to restart the simulation afterward */
356 : maxminSystem_(nullptr)
358 readyActionSet_ = new ActionList();
359 runningActionSet_ = new ActionList();
360 failedActionSet_ = new ActionList();
361 doneActionSet_ = new ActionList();
363 modifiedSet_ = nullptr;
364 actionHeap_ = nullptr;
365 updateMechanism_ = UM_UNDEFINED;
366 selectiveUpdate_ = 0;
370 delete readyActionSet_;
371 delete runningActionSet_;
372 delete failedActionSet_;
373 delete doneActionSet_;
376 double Model::nextOccuringEvent(double now)
378 //FIXME: set the good function once and for all
379 if (updateMechanism_ == UM_LAZY)
380 return nextOccuringEventLazy(now);
381 else if (updateMechanism_ == UM_FULL)
382 return nextOccuringEventFull(now);
384 xbt_die("Invalid cpu update mechanism!");
387 double Model::nextOccuringEventLazy(double now)
389 XBT_DEBUG("Before share resources, the size of modified actions set is %zu", modifiedSet_->size());
390 lmm_solve(maxminSystem_);
391 XBT_DEBUG("After share resources, The size of modified actions set is %zu", modifiedSet_->size());
393 while (not modifiedSet_->empty()) {
394 Action *action = &(modifiedSet_->front());
395 modifiedSet_->pop_front();
396 bool max_dur_flag = false;
398 if (action->getStateSet() != runningActionSet_)
401 /* bogus priority, skip it */
402 if (action->getPriority() <= 0 || action->getHat()==LATENCY)
405 action->updateRemainingLazy(now);
408 double share = lmm_variable_getvalue(action->getVariable());
411 double time_to_completion;
412 if (action->getRemains() > 0) {
413 time_to_completion = action->getRemainsNoUpdate() / share;
415 time_to_completion = 0.0;
417 min = now + time_to_completion; // when the task will complete if nothing changes
420 if ((action->getMaxDuration() > NO_MAX_DURATION) &&
421 (min <= -1 || action->getStartTime() + action->getMaxDuration() < min)) {
422 // when the task will complete anyway because of the deadline if any
423 min = action->getStartTime() + action->getMaxDuration();
427 XBT_DEBUG("Action(%p) corresponds to variable %d", action, action->getVariable()->id_int);
429 XBT_DEBUG("Action(%p) Start %f. May finish at %f (got a share of %f). Max_duration %f", action,
430 action->getStartTime(), min, share,
431 action->getMaxDuration());
434 action->heapUpdate(actionHeap_, min, max_dur_flag ? MAX_DURATION : NORMAL);
435 XBT_DEBUG("Insert at heap action(%p) min %f now %f", action, min, now);
440 //hereafter must have already the min value for this resource model
441 if (xbt_heap_size(actionHeap_) > 0) {
442 double min = xbt_heap_maxkey(actionHeap_) - now;
443 XBT_DEBUG("minimum with the HEAP %f", min);
446 XBT_DEBUG("The HEAP is empty, thus returning -1");
451 double Model::nextOccuringEventFull(double /*now*/) {
452 maxminSystem_->solve_fun(maxminSystem_);
455 for (auto it(getRunningActionSet()->begin()), itend(getRunningActionSet()->end()); it != itend ; ++it) {
456 Action *action = &*it;
457 double value = lmm_variable_getvalue(action->getVariable());
459 if (action->getRemains() > 0)
460 value = action->getRemainsNoUpdate() / value;
463 if (min < 0 || value < min) {
465 XBT_DEBUG("Updating min (value) with %p: %f", action, min);
468 if ((action->getMaxDuration() >= 0) && (min<0 || action->getMaxDuration() < min)) {
469 min = action->getMaxDuration();
470 XBT_DEBUG("Updating min (duration) with %p: %f", action, min);
473 XBT_DEBUG("min value : %f", min);
478 void Model::updateActionsState(double now, double delta)
480 if (updateMechanism_ == UM_FULL)
481 updateActionsStateFull(now, delta);
482 else if (updateMechanism_ == UM_LAZY)
483 updateActionsStateLazy(now, delta);
485 xbt_die("Invalid cpu update mechanism!");
488 void Model::updateActionsStateLazy(double /*now*/, double /*delta*/)
493 void Model::updateActionsStateFull(double /*now*/, double /*delta*/)
508 Resource::Resource(Model* model, const std::string& name, lmm_constraint_t constraint)
509 : name_(name), model_(model), constraint_(constraint)
512 Resource::~Resource() = default;
514 bool Resource::isOn() const {
517 bool Resource::isOff() const {
521 void Resource::turnOn()
526 void Resource::turnOff()
531 Model* Resource::model() const
536 const std::string& Resource::getName() const
541 const char* Resource::getCname() const
543 return name_.c_str();
546 bool Resource::operator==(const Resource &other) const {
547 return name_ == other.name_;
550 lmm_constraint_t Resource::constraint() const
562 const char *surf_action_state_names[6] = {
564 "SURF_ACTION_RUNNING",
565 "SURF_ACTION_FAILED",
567 "SURF_ACTION_TO_FREE",
568 "SURF_ACTION_NOT_IN_THE_SYSTEM"
571 /* added to manage the communication action's heap */
572 void surf_action_lmm_update_index_heap(void *action, int i) {
573 static_cast<simgrid::surf::Action*>(action)->updateIndexHeap(i);
579 Action::Action(simgrid::surf::Model* model, double cost, bool failed) : Action(model, cost, failed, nullptr)
583 Action::Action(simgrid::surf::Model* model, double cost, bool failed, lmm_variable_t var)
584 : remains_(cost), start_(surf_get_clock()), cost_(cost), model_(model), variable_(var)
587 stateSet_ = getModel()->getFailedActionSet();
589 stateSet_ = getModel()->getRunningActionSet();
591 stateSet_->push_back(*this);
598 void Action::finish(Action::State state)
600 finishTime_ = surf_get_clock();
604 Action::State Action::getState()
606 if (stateSet_ == model_->getReadyActionSet())
607 return Action::State::ready;
608 if (stateSet_ == model_->getRunningActionSet())
609 return Action::State::running;
610 if (stateSet_ == model_->getFailedActionSet())
611 return Action::State::failed;
612 if (stateSet_ == model_->getDoneActionSet())
613 return Action::State::done;
614 return Action::State::not_in_the_system;
617 void Action::setState(Action::State state)
619 stateSet_->erase(stateSet_->iterator_to(*this));
621 case Action::State::ready:
622 stateSet_ = model_->getReadyActionSet();
624 case Action::State::running:
625 stateSet_ = model_->getRunningActionSet();
627 case Action::State::failed:
628 stateSet_ = model_->getFailedActionSet();
630 case Action::State::done:
631 stateSet_ = model_->getDoneActionSet();
638 stateSet_->push_back(*this);
641 double Action::getBound()
643 return (variable_) ? lmm_variable_getbound(variable_) : 0;
646 void Action::setBound(double bound)
648 XBT_IN("(%p,%g)", this, bound);
650 lmm_update_variable_bound(getModel()->getMaxminSystem(), variable_, bound);
652 if (getModel()->getUpdateMechanism() == UM_LAZY && getLastUpdate() != surf_get_clock())
653 heapRemove(getModel()->getActionHeap());
657 double Action::getStartTime()
662 double Action::getFinishTime()
667 void Action::setData(void* data)
672 void Action::setCategory(const char *category)
674 category_ = xbt_strdup(category);
681 void Action::setMaxDuration(double duration)
683 maxDuration_ = duration;
684 if (getModel()->getUpdateMechanism() == UM_LAZY) // remove action from the heap
685 heapRemove(getModel()->getActionHeap());
688 void Action::setSharingWeight(double weight)
690 XBT_IN("(%p,%g)", this, weight);
691 sharingWeight_ = weight;
692 lmm_update_variable_weight(getModel()->getMaxminSystem(), getVariable(), weight);
694 if (getModel()->getUpdateMechanism() == UM_LAZY)
695 heapRemove(getModel()->getActionHeap());
699 void Action::cancel(){
700 setState(Action::State::failed);
701 if (getModel()->getUpdateMechanism() == UM_LAZY) {
702 if (action_lmm_hook.is_linked())
703 getModel()->getModifiedSet()->erase(getModel()->getModifiedSet()->iterator_to(*this));
704 heapRemove(getModel()->getActionHeap());
711 if (action_hook.is_linked())
712 stateSet_->erase(stateSet_->iterator_to(*this));
714 lmm_variable_free(getModel()->getMaxminSystem(), getVariable());
715 if (getModel()->getUpdateMechanism() == UM_LAZY) {
716 /* remove from heap */
717 heapRemove(getModel()->getActionHeap());
718 if (action_lmm_hook.is_linked())
719 getModel()->getModifiedSet()->erase(getModel()->getModifiedSet()->iterator_to(*this));
727 void Action::suspend()
729 XBT_IN("(%p)", this);
730 if (suspended_ != 2) {
731 lmm_update_variable_weight(getModel()->getMaxminSystem(), getVariable(), 0.0);
732 if (getModel()->getUpdateMechanism() == UM_LAZY){
733 heapRemove(getModel()->getActionHeap());
734 if (getModel()->getUpdateMechanism() == UM_LAZY && stateSet_ == getModel()->getRunningActionSet() &&
735 sharingWeight_ > 0) {
736 //If we have a lazy model, we need to update the remaining value accordingly
737 updateRemainingLazy(surf_get_clock());
745 void Action::resume()
747 XBT_IN("(%p)", this);
748 if (suspended_ != 2) {
749 lmm_update_variable_weight(getModel()->getMaxminSystem(), getVariable(), getPriority());
751 if (getModel()->getUpdateMechanism() == UM_LAZY)
752 heapRemove(getModel()->getActionHeap());
757 bool Action::isSuspended()
759 return suspended_ == 1;
761 /* insert action on heap using a given key and a hat (heap_action_type)
762 * a hat can be of three types for communications:
764 * NORMAL = this is a normal heap entry stating the date to finish transmitting
765 * LATENCY = this is a heap entry to warn us when the latency is payed
766 * MAX_DURATION =this is a heap entry to warn us when the max_duration limit is reached
768 void Action::heapInsert(xbt_heap_t heap, double key, enum heap_action_type hat)
771 xbt_heap_push(heap, this, key);
774 void Action::heapRemove(xbt_heap_t heap)
777 if (indexHeap_ >= 0) {
778 xbt_heap_remove(heap, indexHeap_);
782 void Action::heapUpdate(xbt_heap_t heap, double key, enum heap_action_type hat)
785 if (indexHeap_ >= 0) {
786 xbt_heap_update(heap, indexHeap_, key);
788 xbt_heap_push(heap, this, key);
792 void Action::updateIndexHeap(int i) {
796 double Action::getRemains()
798 XBT_IN("(%p)", this);
799 /* update remains before return it */
800 if (getModel()->getUpdateMechanism() == UM_LAZY) /* update remains before return it */
801 updateRemainingLazy(surf_get_clock());
806 double Action::getRemainsNoUpdate()