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/lmm/maxmin.hpp" // Constraint
12 #include "src/kernel/routing/NetPoint.hpp"
13 #include "src/surf/HostImpl.hpp"
14 #include "xbt/utility.hpp"
25 XBT_LOG_NEW_CATEGORY(surf, "All SURF categories");
26 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_kernel, surf, "Logging specific to SURF (kernel)");
32 std::vector<surf_model_t> * all_existing_models = nullptr; /* to destroy models correctly */
34 simgrid::trace_mgr::future_evt_set *future_evt_set = nullptr;
35 std::vector<std::string> surf_path;
36 std::vector<simgrid::s4u::Host*> host_that_restart;
37 /** set of hosts for which one want to be notified if they ever restart. */
38 std::set<std::string> watched_hosts;
39 extern std::map<std::string, simgrid::surf::StorageType*> storage_types;
44 simgrid::xbt::signal<void()> surfExitCallbacks;
48 #include <simgrid/plugins/energy.h> // FIXME: this plug-in should not be linked to the core
49 #include <simgrid/plugins/load.h> // FIXME: this plug-in should not be linked to the core
51 s_surf_model_description_t surf_plugin_description[] = {
52 {"Energy", "Cpu energy consumption.", &sg_host_energy_plugin_init},
53 {"Load", "Cpu load.", &sg_host_load_plugin_init},
54 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
57 /* Don't forget to update the option description in smx_config when you change this */
58 s_surf_model_description_t surf_network_model_description[] = {
59 {"LV08", "Realistic network analytic model (slow-start modeled by multiplying latency by 13.01, bandwidth by .97; "
60 "bottleneck sharing uses a payload of S=20537 for evaluating RTT). ",
61 &surf_network_model_init_LegrandVelho},
62 {"Constant", "Simplistic network model where all communication take a constant time (one second). This model "
63 "provides the lowest realism, but is (marginally) faster.",
64 &surf_network_model_init_Constant},
65 {"SMPI", "Realistic network model specifically tailored for HPC settings (accurate modeling of slow start with "
66 "correction factors on three intervals: < 1KiB, < 64 KiB, >= 64 KiB)",
67 &surf_network_model_init_SMPI},
68 {"IB", "Realistic network model specifically tailored for HPC settings, with Infiniband contention model",
69 &surf_network_model_init_IB},
70 {"CM02", "Legacy network analytic model (Very similar to LV08, but without corrective factors. The timings of "
71 "small messages are thus poorly modeled).",
72 &surf_network_model_init_CM02},
73 {"NS3", "Network pseudo-model using the NS3 tcp model instead of an analytic model", &surf_network_model_init_NS3},
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_Reno},
78 "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
79 &surf_network_model_init_Reno2},
81 "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
82 &surf_network_model_init_Vegas},
83 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
87 void surf_network_model_init_SMPI() {
88 xbt_die("Please activate SMPI support in cmake to use the SMPI network model.");
90 void surf_network_model_init_IB() {
91 xbt_die("Please activate SMPI support in cmake to use the IB network model.");
95 void surf_network_model_init_NS3() {
96 xbt_die("Please activate NS3 support in cmake and install the dependencies to use the NS3 network model.");
100 s_surf_model_description_t surf_cpu_model_description[] = {
101 {"Cas01", "Simplistic CPU model (time=size/power).", &surf_cpu_model_init_Cas01},
102 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
105 s_surf_model_description_t surf_host_model_description[] = {
106 {"default", "Default host model. Currently, CPU:Cas01 and network:LV08 (with cross traffic enabled)", &surf_host_model_init_current_default},
107 {"compound", "Host model that is automatically chosen if you change the network and CPU models", &surf_host_model_init_compound},
108 {"ptask_L07", "Host model somehow similar to Cas01+CM02 but allowing parallel tasks", &surf_host_model_init_ptask_L07},
109 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
112 s_surf_model_description_t surf_optimization_mode_description[] = {
113 {"Lazy", "Lazy action management (partial invalidation in lmm + heap in action remaining).", nullptr},
114 {"TI", "Trace integration. Highly optimized mode when using availability traces (only available for the Cas01 CPU model for now).", nullptr},
115 {"Full", "Full update of remaining and variables. Slow but may be useful when debugging.", nullptr},
116 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
119 s_surf_model_description_t surf_storage_model_description[] = {
120 {"default", "Simplistic storage model.", &surf_storage_model_init_default},
121 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
126 double surf_get_clock()
132 # define FILE_DELIM "\\"
134 # define FILE_DELIM "/" /* FIXME: move to better location */
137 std::ifstream* surf_ifsopen(std::string name)
139 std::ifstream* fs = new std::ifstream();
140 xbt_assert(not name.empty());
141 if (__surf_is_absolute_file_path(name.c_str())) { /* don't mess with absolute file names */
142 fs->open(name.c_str(), std::ifstream::in);
145 /* search relative files in the path */
146 for (auto const& path_elm : surf_path) {
147 std::string buff = path_elm + FILE_DELIM + name;
148 fs->open(buff.c_str(), std::ifstream::in);
150 if (not fs->fail()) {
151 XBT_DEBUG("Found file at %s", buff.c_str());
159 FILE *surf_fopen(const char *name, const char *mode)
161 FILE *file = nullptr;
165 if (__surf_is_absolute_file_path(name)) /* don't mess with absolute file names */
166 return fopen(name, mode);
168 /* search relative files in the path */
169 for (auto const& path_elm : surf_path) {
170 std::string buff = path_elm + FILE_DELIM + name;
171 file = fopen(buff.c_str(), mode);
179 /* The __surf_is_absolute_file_path() returns 1 if
180 * file_path is a absolute file path, in the other
181 * case the function returns 0.
183 int __surf_is_absolute_file_path(const char *file_path)
186 WIN32_FIND_DATA wfd = { 0 };
187 HANDLE hFile = FindFirstFile(file_path, &wfd);
189 if (INVALID_HANDLE_VALUE == hFile)
195 return (file_path[0] == '/');
199 /** Displays the long description of all registered models, and quit */
200 void model_help(const char *category, s_surf_model_description_t * table)
202 printf("Long description of the %s models accepted by this simulator:\n", category);
203 for (int i = 0; table[i].name; i++)
204 printf(" %s: %s\n", table[i].name, table[i].description);
207 int find_model_description(s_surf_model_description_t* table, std::string name)
209 for (int i = 0; table[i].name; i++)
210 if (name == table[i].name)
213 if (not table[0].name)
214 xbt_die("No model is valid! This is a bug.");
216 std::string name_list = std::string(table[0].name);
217 for (int i = 1; table[i].name; i++)
218 name_list = name_list + ", " + table[i].name;
220 xbt_die("Model '%s' is invalid! Valid models are: %s.", name.c_str(), name_list.c_str());
224 void sg_version_check(int lib_version_major, int lib_version_minor, int lib_version_patch)
226 if ((lib_version_major != SIMGRID_VERSION_MAJOR) || (lib_version_minor != SIMGRID_VERSION_MINOR)) {
227 fprintf(stderr, "FATAL ERROR: Your program was compiled with SimGrid version %d.%d.%d, "
228 "and then linked against SimGrid %d.%d.%d. Please fix this.\n",
229 lib_version_major, lib_version_minor, lib_version_patch, SIMGRID_VERSION_MAJOR, SIMGRID_VERSION_MINOR,
230 SIMGRID_VERSION_PATCH);
233 if (lib_version_patch != SIMGRID_VERSION_PATCH) {
234 if (SIMGRID_VERSION_PATCH >= 90 || lib_version_patch >= 90) {
237 "FATAL ERROR: Your program was compiled with SimGrid version %d.%d.%d, "
238 "and then linked against SimGrid %d.%d.%d. \n"
239 "One of them is a development version, and should not be mixed with the stable release. Please fix this.\n",
240 lib_version_major, lib_version_minor, lib_version_patch, SIMGRID_VERSION_MAJOR, SIMGRID_VERSION_MINOR,
241 SIMGRID_VERSION_PATCH);
244 fprintf(stderr, "Warning: Your program was compiled with SimGrid version %d.%d.%d, "
245 "and then linked against SimGrid %d.%d.%d. Proceeding anyway.\n",
246 lib_version_major, lib_version_minor, lib_version_patch, SIMGRID_VERSION_MAJOR, SIMGRID_VERSION_MINOR,
247 SIMGRID_VERSION_PATCH);
251 void sg_version_get(int* ver_major, int* ver_minor, int* ver_patch)
253 *ver_major = SIMGRID_VERSION_MAJOR;
254 *ver_minor = SIMGRID_VERSION_MINOR;
255 *ver_patch = SIMGRID_VERSION_PATCH;
260 std::printf("This program was linked against %s (git: %s), found in %s.\n",
261 SIMGRID_VERSION_STRING, SIMGRID_GIT_VERSION, SIMGRID_INSTALL_PREFIX);
264 std::printf(" Model-checking support compiled in.\n");
266 std::printf(" Model-checking support disabled at compilation.\n");
270 std::printf(" NS3 support compiled in.\n");
272 std::printf(" NS3 support disabled at compilation.\n");
275 #if SIMGRID_HAVE_JEDULE
276 std::printf(" Jedule support compiled in.\n");
278 std::printf(" Jedule support disabled at compilation.\n");
282 std::printf(" Lua support compiled in.\n");
284 std::printf(" Lua support disabled at compilation.\n");
287 #if SIMGRID_HAVE_MALLOCATOR
288 std::printf(" Mallocator support compiled in.\n");
290 std::printf(" Mallocator support disabled at compilation.\n");
293 std::printf("\nTo cite SimGrid in a publication, please use:\n"
294 " Henri Casanova, Arnaud Giersch, Arnaud Legrand, Martin Quinson, Frédéric Suter. \n"
295 " Versatile, Scalable, and Accurate Simulation of Distributed Applications and Platforms. \n"
296 " Journal of Parallel and Distributed Computing, Elsevier, 2014, 74 (10), pp.2899-2917.\n");
297 std::printf("The pdf file and a BibTeX entry for LaTeX users can be found at http://hal.inria.fr/hal-01017319\n");
300 void surf_init(int *argc, char **argv)
302 if (USER_HOST_LEVEL != -1) // Already initialized
305 XBT_DEBUG("Create all Libs");
306 USER_HOST_LEVEL = simgrid::s4u::Host::extension_create(nullptr);
308 xbt_init(argc, argv);
309 if (not all_existing_models)
310 all_existing_models = new std::vector<simgrid::surf::Model*>();
311 if (not future_evt_set)
312 future_evt_set = new simgrid::trace_mgr::future_evt_set();
314 sg_config_init(argc, argv);
322 TRACE_end(); /* Just in case it was not called by the upper layer (or there is no upper layer) */
326 for (auto const& e : storage_types) {
327 simgrid::surf::StorageType* stype = e.second;
328 delete stype->properties;
329 delete stype->model_properties;
332 for (auto const& s : *simgrid::surf::StorageImpl::storagesMap())
334 delete simgrid::surf::StorageImpl::storagesMap();
336 for (auto const& model : *all_existing_models)
338 delete all_existing_models;
340 simgrid::surf::surfExitCallbacks();
342 if (future_evt_set) {
343 delete future_evt_set;
344 future_evt_set = nullptr;
349 simgrid::s4u::Engine::shutdown();
351 NOW = 0; /* Just in case the user plans to restart the simulation afterward */
362 : maxminSystem_(nullptr)
364 readyActionSet_ = new ActionList();
365 runningActionSet_ = new ActionList();
366 failedActionSet_ = new ActionList();
367 doneActionSet_ = new ActionList();
369 modifiedSet_ = nullptr;
370 updateMechanism_ = UM_UNDEFINED;
371 selectiveUpdate_ = 0;
375 delete readyActionSet_;
376 delete runningActionSet_;
377 delete failedActionSet_;
378 delete doneActionSet_;
380 delete maxminSystem_;
383 Action* Model::actionHeapPop()
385 Action* action = actionHeap_.top().second;
387 action->clearHeapHandle();
391 double Model::nextOccuringEvent(double now)
393 //FIXME: set the good function once and for all
394 if (updateMechanism_ == UM_LAZY)
395 return nextOccuringEventLazy(now);
396 else if (updateMechanism_ == UM_FULL)
397 return nextOccuringEventFull(now);
399 xbt_die("Invalid cpu update mechanism!");
402 double Model::nextOccuringEventLazy(double now)
404 XBT_DEBUG("Before share resources, the size of modified actions set is %zu", modifiedSet_->size());
405 lmm_solve(maxminSystem_);
406 XBT_DEBUG("After share resources, The size of modified actions set is %zu", modifiedSet_->size());
408 while (not modifiedSet_->empty()) {
409 Action *action = &(modifiedSet_->front());
410 modifiedSet_->pop_front();
411 bool max_dur_flag = false;
413 if (action->getStateSet() != runningActionSet_)
416 /* bogus priority, skip it */
417 if (action->getPriority() <= 0 || action->getHat()==LATENCY)
420 action->updateRemainingLazy(now);
423 double share = action->getVariable()->get_value();
426 double time_to_completion;
427 if (action->getRemains() > 0) {
428 time_to_completion = action->getRemainsNoUpdate() / share;
430 time_to_completion = 0.0;
432 min = now + time_to_completion; // when the task will complete if nothing changes
435 if ((action->getMaxDuration() > NO_MAX_DURATION) &&
436 (min <= -1 || action->getStartTime() + action->getMaxDuration() < min)) {
437 // when the task will complete anyway because of the deadline if any
438 min = action->getStartTime() + action->getMaxDuration();
442 XBT_DEBUG("Action(%p) corresponds to variable %d", action, action->getVariable()->id_int);
444 XBT_DEBUG("Action(%p) Start %f. May finish at %f (got a share of %f). Max_duration %f", action,
445 action->getStartTime(), min, share,
446 action->getMaxDuration());
449 action->heapUpdate(actionHeap_, min, max_dur_flag ? MAX_DURATION : NORMAL);
450 XBT_DEBUG("Insert at heap action(%p) min %f now %f", action, min, now);
455 //hereafter must have already the min value for this resource model
456 if (not actionHeapIsEmpty()) {
457 double min = actionHeapTopDate() - now;
458 XBT_DEBUG("minimum with the HEAP %f", min);
461 XBT_DEBUG("The HEAP is empty, thus returning -1");
466 double Model::nextOccuringEventFull(double /*now*/) {
467 maxminSystem_->solve_fun(maxminSystem_);
471 for (Action& action : *getRunningActionSet()) {
472 double value = action.getVariable()->get_value();
474 if (action.getRemains() > 0)
475 value = action.getRemainsNoUpdate() / value;
478 if (min < 0 || value < min) {
480 XBT_DEBUG("Updating min (value) with %p: %f", &action, min);
483 if ((action.getMaxDuration() >= 0) && (min < 0 || action.getMaxDuration() < min)) {
484 min = action.getMaxDuration();
485 XBT_DEBUG("Updating min (duration) with %p: %f", &action, min);
488 XBT_DEBUG("min value : %f", min);
493 void Model::updateActionsState(double now, double delta)
495 if (updateMechanism_ == UM_FULL)
496 updateActionsStateFull(now, delta);
497 else if (updateMechanism_ == UM_LAZY)
498 updateActionsStateLazy(now, delta);
500 xbt_die("Invalid cpu update mechanism!");
503 void Model::updateActionsStateLazy(double /*now*/, double /*delta*/)
508 void Model::updateActionsStateFull(double /*now*/, double /*delta*/)
523 Resource::Resource(Model* model, const std::string& name, lmm_constraint_t constraint)
524 : name_(name), model_(model), constraint_(constraint)
527 Resource::~Resource() = default;
529 bool Resource::isOn() const {
532 bool Resource::isOff() const {
536 void Resource::turnOn()
541 void Resource::turnOff()
546 double Resource::getLoad()
548 return constraint_->get_usage();
551 Model* Resource::model() const
556 const std::string& Resource::getName() const
561 const char* Resource::getCname() const
563 return name_.c_str();
566 bool Resource::operator==(const Resource &other) const {
567 return name_ == other.name_;
570 lmm_constraint_t Resource::constraint() const
582 const char *surf_action_state_names[6] = {
584 "SURF_ACTION_RUNNING",
585 "SURF_ACTION_FAILED",
587 "SURF_ACTION_TO_FREE",
588 "SURF_ACTION_NOT_IN_THE_SYSTEM"
594 Action::Action(simgrid::surf::Model* model, double cost, bool failed) : Action(model, cost, failed, nullptr)
598 Action::Action(simgrid::surf::Model* model, double cost, bool failed, lmm_variable_t var)
599 : remains_(cost), start_(surf_get_clock()), cost_(cost), model_(model), variable_(var)
602 stateSet_ = getModel()->getFailedActionSet();
604 stateSet_ = getModel()->getRunningActionSet();
606 stateSet_->push_back(*this);
613 void Action::finish(Action::State state)
615 finishTime_ = surf_get_clock();
619 Action::State Action::getState() const
621 if (stateSet_ == model_->getReadyActionSet())
622 return Action::State::ready;
623 if (stateSet_ == model_->getRunningActionSet())
624 return Action::State::running;
625 if (stateSet_ == model_->getFailedActionSet())
626 return Action::State::failed;
627 if (stateSet_ == model_->getDoneActionSet())
628 return Action::State::done;
629 return Action::State::not_in_the_system;
632 void Action::setState(Action::State state)
634 simgrid::xbt::intrusive_erase(*stateSet_, *this);
636 case Action::State::ready:
637 stateSet_ = model_->getReadyActionSet();
639 case Action::State::running:
640 stateSet_ = model_->getRunningActionSet();
642 case Action::State::failed:
643 stateSet_ = model_->getFailedActionSet();
645 case Action::State::done:
646 stateSet_ = model_->getDoneActionSet();
653 stateSet_->push_back(*this);
656 double Action::getBound() const
658 return variable_ ? variable_->get_bound() : 0;
661 void Action::setBound(double bound)
663 XBT_IN("(%p,%g)", this, bound);
665 getModel()->getMaxminSystem()->update_variable_bound(variable_, bound);
667 if (getModel()->getUpdateMechanism() == UM_LAZY && getLastUpdate() != surf_get_clock())
668 heapRemove(getModel()->getActionHeap());
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 getModel()->getMaxminSystem()->update_variable_weight(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 simgrid::xbt::intrusive_erase(*getModel()->getModifiedSet(), *this);
704 heapRemove(getModel()->getActionHeap());
711 if (action_hook.is_linked())
712 simgrid::xbt::intrusive_erase(*stateSet_, *this);
714 getModel()->getMaxminSystem()->variable_free(getVariable());
715 if (getModel()->getUpdateMechanism() == UM_LAZY) {
716 /* remove from heap */
717 heapRemove(getModel()->getActionHeap());
718 if (action_lmm_hook.is_linked())
719 simgrid::xbt::intrusive_erase(*getModel()->getModifiedSet(), *this);
727 void Action::suspend()
729 XBT_IN("(%p)", this);
730 if (suspended_ != 2) {
731 getModel()->getMaxminSystem()->update_variable_weight(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 getModel()->getMaxminSystem()->update_variable_weight(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(heap_type& heap, double key, enum heap_action_type hat)
771 heapHandle_ = heap.emplace(std::make_pair(key, this));
774 void Action::heapRemove(heap_type& heap)
778 heap.erase(*heapHandle_);
783 void Action::heapUpdate(heap_type& heap, double key, enum heap_action_type hat)
787 heap.update(*heapHandle_, std::make_pair(key, this));
789 heapHandle_ = heap.emplace(std::make_pair(key, this));
793 double Action::getRemains()
795 XBT_IN("(%p)", this);
796 /* update remains before return it */
797 if (getModel()->getUpdateMechanism() == UM_LAZY) /* update remains before return it */
798 updateRemainingLazy(surf_get_clock());