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"
13 #include "xbt/utility.hpp"
24 XBT_LOG_NEW_CATEGORY(surf, "All SURF categories");
25 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_kernel, surf, "Logging specific to SURF (kernel)");
31 std::vector<surf_model_t> * all_existing_models = nullptr; /* to destroy models correctly */
33 simgrid::trace_mgr::future_evt_set *future_evt_set = nullptr;
34 std::vector<std::string> surf_path;
35 std::vector<simgrid::s4u::Host*> host_that_restart;
36 /** set of hosts for which one want to be notified if they ever restart. */
37 std::set<std::string> watched_hosts;
38 extern std::map<std::string, simgrid::surf::StorageType*> storage_types;
43 simgrid::xbt::signal<void()> surfExitCallbacks;
47 #include <simgrid/plugins/energy.h> // FIXME: this plug-in should not be linked to the core
48 #include <simgrid/plugins/load.h> // FIXME: this plug-in should not be linked to the core
50 s_surf_model_description_t surf_plugin_description[] = {
51 {"Energy", "Cpu energy consumption.", &sg_host_energy_plugin_init},
52 {"Load", "Cpu load.", &sg_host_load_plugin_init},
53 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
56 /* Don't forget to update the option description in smx_config when you change this */
57 s_surf_model_description_t surf_network_model_description[] = {
58 {"LV08", "Realistic network analytic model (slow-start modeled by multiplying latency by 13.01, bandwidth by .97; "
59 "bottleneck sharing uses a payload of S=20537 for evaluating RTT). ",
60 &surf_network_model_init_LegrandVelho},
61 {"Constant", "Simplistic network model where all communication take a constant time (one second). This model "
62 "provides the lowest realism, but is (marginally) faster.",
63 &surf_network_model_init_Constant},
64 {"SMPI", "Realistic network model specifically tailored for HPC settings (accurate modeling of slow start with "
65 "correction factors on three intervals: < 1KiB, < 64 KiB, >= 64 KiB)",
66 &surf_network_model_init_SMPI},
67 {"IB", "Realistic network model specifically tailored for HPC settings, with Infiniband contention model",
68 &surf_network_model_init_IB},
69 {"CM02", "Legacy network analytic model (Very similar to LV08, but without corrective factors. The timings of "
70 "small messages are thus poorly modeled).",
71 &surf_network_model_init_CM02},
72 {"NS3", "Network pseudo-model using the NS3 tcp model instead of an analytic model", &surf_network_model_init_NS3},
74 "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
75 &surf_network_model_init_Reno},
77 "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
78 &surf_network_model_init_Reno2},
80 "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
81 &surf_network_model_init_Vegas},
82 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
86 void surf_network_model_init_SMPI() {
87 xbt_die("Please activate SMPI support in cmake to use the SMPI network model.");
89 void surf_network_model_init_IB() {
90 xbt_die("Please activate SMPI support in cmake to use the IB network model.");
94 void surf_network_model_init_NS3() {
95 xbt_die("Please activate NS3 support in cmake and install the dependencies to use the NS3 network model.");
99 s_surf_model_description_t surf_cpu_model_description[] = {
100 {"Cas01", "Simplistic CPU model (time=size/power).", &surf_cpu_model_init_Cas01},
101 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
104 s_surf_model_description_t surf_host_model_description[] = {
105 {"default", "Default host model. Currently, CPU:Cas01 and network:LV08 (with cross traffic enabled)", &surf_host_model_init_current_default},
106 {"compound", "Host model that is automatically chosen if you change the network and CPU models", &surf_host_model_init_compound},
107 {"ptask_L07", "Host model somehow similar to Cas01+CM02 but allowing parallel tasks", &surf_host_model_init_ptask_L07},
108 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
111 s_surf_model_description_t surf_optimization_mode_description[] = {
112 {"Lazy", "Lazy action management (partial invalidation in lmm + heap in action remaining).", nullptr},
113 {"TI", "Trace integration. Highly optimized mode when using availability traces (only available for the Cas01 CPU model for now).", nullptr},
114 {"Full", "Full update of remaining and variables. Slow but may be useful when debugging.", nullptr},
115 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
118 s_surf_model_description_t surf_storage_model_description[] = {
119 {"default", "Simplistic storage model.", &surf_storage_model_init_default},
120 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
125 double surf_get_clock()
131 # define FILE_DELIM "\\"
133 # define FILE_DELIM "/" /* FIXME: move to better location */
136 std::ifstream* surf_ifsopen(std::string name)
138 std::ifstream* fs = new std::ifstream();
139 xbt_assert(not name.empty());
140 if (__surf_is_absolute_file_path(name.c_str())) { /* don't mess with absolute file names */
141 fs->open(name.c_str(), std::ifstream::in);
144 /* search relative files in the path */
145 for (auto const& path_elm : surf_path) {
146 std::string buff = path_elm + FILE_DELIM + name;
147 fs->open(buff.c_str(), std::ifstream::in);
149 if (not fs->fail()) {
150 XBT_DEBUG("Found file at %s", buff.c_str());
158 FILE *surf_fopen(const char *name, const char *mode)
160 FILE *file = nullptr;
164 if (__surf_is_absolute_file_path(name)) /* don't mess with absolute file names */
165 return fopen(name, mode);
167 /* search relative files in the path */
168 for (auto const& path_elm : surf_path) {
169 std::string buff = path_elm + FILE_DELIM + name;
170 file = fopen(buff.c_str(), mode);
178 /* The __surf_is_absolute_file_path() returns 1 if
179 * file_path is a absolute file path, in the other
180 * case the function returns 0.
182 int __surf_is_absolute_file_path(const char *file_path)
185 WIN32_FIND_DATA wfd = { 0 };
186 HANDLE hFile = FindFirstFile(file_path, &wfd);
188 if (INVALID_HANDLE_VALUE == hFile)
194 return (file_path[0] == '/');
198 /** Displays the long description of all registered models, and quit */
199 void model_help(const char *category, s_surf_model_description_t * table)
201 printf("Long description of the %s models accepted by this simulator:\n", category);
202 for (int i = 0; table[i].name; i++)
203 printf(" %s: %s\n", table[i].name, table[i].description);
206 int find_model_description(s_surf_model_description_t* table, std::string name)
208 for (int i = 0; table[i].name; i++)
209 if (name == table[i].name)
212 if (not table[0].name)
213 xbt_die("No model is valid! This is a bug.");
215 std::string name_list = std::string(table[0].name);
216 for (int i = 1; table[i].name; i++)
217 name_list = name_list + ", " + table[i].name;
219 xbt_die("Model '%s' is invalid! Valid models are: %s.", name.c_str(), name_list.c_str());
223 void sg_version_check(int lib_version_major, int lib_version_minor, int lib_version_patch)
225 if ((lib_version_major != SIMGRID_VERSION_MAJOR) || (lib_version_minor != SIMGRID_VERSION_MINOR)) {
226 fprintf(stderr, "FATAL ERROR: Your program was compiled with SimGrid version %d.%d.%d, "
227 "and then linked against SimGrid %d.%d.%d. Please fix this.\n",
228 lib_version_major, lib_version_minor, lib_version_patch, SIMGRID_VERSION_MAJOR, SIMGRID_VERSION_MINOR,
229 SIMGRID_VERSION_PATCH);
232 if (lib_version_patch != SIMGRID_VERSION_PATCH) {
233 if (SIMGRID_VERSION_PATCH >= 90 || lib_version_patch >= 90) {
236 "FATAL ERROR: Your program was compiled with SimGrid version %d.%d.%d, "
237 "and then linked against SimGrid %d.%d.%d. \n"
238 "One of them is a development version, and should not be mixed with the stable release. Please fix this.\n",
239 lib_version_major, lib_version_minor, lib_version_patch, SIMGRID_VERSION_MAJOR, SIMGRID_VERSION_MINOR,
240 SIMGRID_VERSION_PATCH);
243 fprintf(stderr, "Warning: Your program was compiled with SimGrid version %d.%d.%d, "
244 "and then linked against SimGrid %d.%d.%d. Proceeding anyway.\n",
245 lib_version_major, lib_version_minor, lib_version_patch, SIMGRID_VERSION_MAJOR, SIMGRID_VERSION_MINOR,
246 SIMGRID_VERSION_PATCH);
250 void sg_version_get(int* ver_major, int* ver_minor, int* ver_patch)
252 *ver_major = SIMGRID_VERSION_MAJOR;
253 *ver_minor = SIMGRID_VERSION_MINOR;
254 *ver_patch = SIMGRID_VERSION_PATCH;
259 std::printf("This program was linked against %s (git: %s), found in %s.\n",
260 SIMGRID_VERSION_STRING, SIMGRID_GIT_VERSION, SIMGRID_INSTALL_PREFIX);
263 std::printf(" Model-checking support compiled in.\n");
265 std::printf(" Model-checking support disabled at compilation.\n");
269 std::printf(" NS3 support compiled in.\n");
271 std::printf(" NS3 support disabled at compilation.\n");
274 #if SIMGRID_HAVE_JEDULE
275 std::printf(" Jedule support compiled in.\n");
277 std::printf(" Jedule support disabled at compilation.\n");
281 std::printf(" Lua support compiled in.\n");
283 std::printf(" Lua support disabled at compilation.\n");
286 #if SIMGRID_HAVE_MALLOCATOR
287 std::printf(" Mallocator support compiled in.\n");
289 std::printf(" Mallocator support disabled at compilation.\n");
292 std::printf("\nTo cite SimGrid in a publication, please use:\n"
293 " Henri Casanova, Arnaud Giersch, Arnaud Legrand, Martin Quinson, Frédéric Suter. \n"
294 " Versatile, Scalable, and Accurate Simulation of Distributed Applications and Platforms. \n"
295 " Journal of Parallel and Distributed Computing, Elsevier, 2014, 74 (10), pp.2899-2917.\n");
296 std::printf("The pdf file and a BibTeX entry for LaTeX users can be found at http://hal.inria.fr/hal-01017319\n");
299 void surf_init(int *argc, char **argv)
301 if (USER_HOST_LEVEL != -1) // Already initialized
304 XBT_DEBUG("Create all Libs");
305 USER_HOST_LEVEL = simgrid::s4u::Host::extension_create(nullptr);
307 xbt_init(argc, argv);
308 if (not all_existing_models)
309 all_existing_models = new std::vector<simgrid::surf::Model*>();
310 if (not future_evt_set)
311 future_evt_set = new simgrid::trace_mgr::future_evt_set();
313 sg_config_init(argc, argv);
321 TRACE_end(); /* Just in case it was not called by the upper layer (or there is no upper layer) */
325 for (auto const& e : storage_types) {
326 simgrid::surf::StorageType* stype = e.second;
327 delete stype->properties;
328 delete stype->model_properties;
331 for (auto const& s : *simgrid::surf::StorageImpl::storagesMap())
333 delete simgrid::surf::StorageImpl::storagesMap();
335 for (auto const& model : *all_existing_models)
337 delete all_existing_models;
339 simgrid::surf::surfExitCallbacks();
341 if (future_evt_set) {
342 delete future_evt_set;
343 future_evt_set = nullptr;
348 simgrid::s4u::Engine::shutdown();
350 NOW = 0; /* Just in case the user plans to restart the simulation afterward */
361 : maxminSystem_(nullptr)
363 readyActionSet_ = new ActionList();
364 runningActionSet_ = new ActionList();
365 failedActionSet_ = new ActionList();
366 doneActionSet_ = new ActionList();
368 modifiedSet_ = nullptr;
369 updateMechanism_ = UM_UNDEFINED;
370 selectiveUpdate_ = 0;
374 delete readyActionSet_;
375 delete runningActionSet_;
376 delete failedActionSet_;
377 delete doneActionSet_;
380 Action* Model::actionHeapPop()
382 Action* action = actionHeap_.top().second;
384 action->clearHeapHandle();
388 double Model::nextOccuringEvent(double now)
390 //FIXME: set the good function once and for all
391 if (updateMechanism_ == UM_LAZY)
392 return nextOccuringEventLazy(now);
393 else if (updateMechanism_ == UM_FULL)
394 return nextOccuringEventFull(now);
396 xbt_die("Invalid cpu update mechanism!");
399 double Model::nextOccuringEventLazy(double now)
401 XBT_DEBUG("Before share resources, the size of modified actions set is %zu", modifiedSet_->size());
402 lmm_solve(maxminSystem_);
403 XBT_DEBUG("After share resources, The size of modified actions set is %zu", modifiedSet_->size());
405 while (not modifiedSet_->empty()) {
406 Action *action = &(modifiedSet_->front());
407 modifiedSet_->pop_front();
408 bool max_dur_flag = false;
410 if (action->getStateSet() != runningActionSet_)
413 /* bogus priority, skip it */
414 if (action->getPriority() <= 0 || action->getHat()==LATENCY)
417 action->updateRemainingLazy(now);
420 double share = action->getVariable()->get_value();
423 double time_to_completion;
424 if (action->getRemains() > 0) {
425 time_to_completion = action->getRemainsNoUpdate() / share;
427 time_to_completion = 0.0;
429 min = now + time_to_completion; // when the task will complete if nothing changes
432 if ((action->getMaxDuration() > NO_MAX_DURATION) &&
433 (min <= -1 || action->getStartTime() + action->getMaxDuration() < min)) {
434 // when the task will complete anyway because of the deadline if any
435 min = action->getStartTime() + action->getMaxDuration();
439 XBT_DEBUG("Action(%p) corresponds to variable %d", action, action->getVariable()->id_int);
441 XBT_DEBUG("Action(%p) Start %f. May finish at %f (got a share of %f). Max_duration %f", action,
442 action->getStartTime(), min, share,
443 action->getMaxDuration());
446 action->heapUpdate(actionHeap_, min, max_dur_flag ? MAX_DURATION : NORMAL);
447 XBT_DEBUG("Insert at heap action(%p) min %f now %f", action, min, now);
452 //hereafter must have already the min value for this resource model
453 if (not actionHeapIsEmpty()) {
454 double min = actionHeapTopDate() - now;
455 XBT_DEBUG("minimum with the HEAP %f", min);
458 XBT_DEBUG("The HEAP is empty, thus returning -1");
463 double Model::nextOccuringEventFull(double /*now*/) {
464 maxminSystem_->solve_fun(maxminSystem_);
468 for (Action& action : *getRunningActionSet()) {
469 double value = action.getVariable()->get_value();
471 if (action.getRemains() > 0)
472 value = action.getRemainsNoUpdate() / value;
475 if (min < 0 || value < min) {
477 XBT_DEBUG("Updating min (value) with %p: %f", &action, min);
480 if ((action.getMaxDuration() >= 0) && (min < 0 || action.getMaxDuration() < min)) {
481 min = action.getMaxDuration();
482 XBT_DEBUG("Updating min (duration) with %p: %f", &action, min);
485 XBT_DEBUG("min value : %f", min);
490 void Model::updateActionsState(double now, double delta)
492 if (updateMechanism_ == UM_FULL)
493 updateActionsStateFull(now, delta);
494 else if (updateMechanism_ == UM_LAZY)
495 updateActionsStateLazy(now, delta);
497 xbt_die("Invalid cpu update mechanism!");
500 void Model::updateActionsStateLazy(double /*now*/, double /*delta*/)
505 void Model::updateActionsStateFull(double /*now*/, double /*delta*/)
520 Resource::Resource(Model* model, const std::string& name, lmm_constraint_t constraint)
521 : name_(name), model_(model), constraint_(constraint)
524 Resource::~Resource() = default;
526 bool Resource::isOn() const {
529 bool Resource::isOff() const {
533 void Resource::turnOn()
538 void Resource::turnOff()
543 Model* Resource::model() const
548 const std::string& Resource::getName() const
553 const char* Resource::getCname() const
555 return name_.c_str();
558 bool Resource::operator==(const Resource &other) const {
559 return name_ == other.name_;
562 lmm_constraint_t Resource::constraint() const
574 const char *surf_action_state_names[6] = {
576 "SURF_ACTION_RUNNING",
577 "SURF_ACTION_FAILED",
579 "SURF_ACTION_TO_FREE",
580 "SURF_ACTION_NOT_IN_THE_SYSTEM"
586 Action::Action(simgrid::surf::Model* model, double cost, bool failed) : Action(model, cost, failed, nullptr)
590 Action::Action(simgrid::surf::Model* model, double cost, bool failed, lmm_variable_t var)
591 : remains_(cost), start_(surf_get_clock()), cost_(cost), model_(model), variable_(var)
594 stateSet_ = getModel()->getFailedActionSet();
596 stateSet_ = getModel()->getRunningActionSet();
598 stateSet_->push_back(*this);
605 void Action::finish(Action::State state)
607 finishTime_ = surf_get_clock();
611 Action::State Action::getState() const
613 if (stateSet_ == model_->getReadyActionSet())
614 return Action::State::ready;
615 if (stateSet_ == model_->getRunningActionSet())
616 return Action::State::running;
617 if (stateSet_ == model_->getFailedActionSet())
618 return Action::State::failed;
619 if (stateSet_ == model_->getDoneActionSet())
620 return Action::State::done;
621 return Action::State::not_in_the_system;
624 void Action::setState(Action::State state)
626 simgrid::xbt::intrusive_erase(*stateSet_, *this);
628 case Action::State::ready:
629 stateSet_ = model_->getReadyActionSet();
631 case Action::State::running:
632 stateSet_ = model_->getRunningActionSet();
634 case Action::State::failed:
635 stateSet_ = model_->getFailedActionSet();
637 case Action::State::done:
638 stateSet_ = model_->getDoneActionSet();
645 stateSet_->push_back(*this);
648 double Action::getBound() const
650 return variable_ ? variable_->get_bound() : 0;
653 void Action::setBound(double bound)
655 XBT_IN("(%p,%g)", this, bound);
657 getModel()->getMaxminSystem()->update_variable_bound(variable_, bound);
659 if (getModel()->getUpdateMechanism() == UM_LAZY && getLastUpdate() != surf_get_clock())
660 heapRemove(getModel()->getActionHeap());
664 void Action::setCategory(const char *category)
666 category_ = xbt_strdup(category);
673 void Action::setMaxDuration(double duration)
675 maxDuration_ = duration;
676 if (getModel()->getUpdateMechanism() == UM_LAZY) // remove action from the heap
677 heapRemove(getModel()->getActionHeap());
680 void Action::setSharingWeight(double weight)
682 XBT_IN("(%p,%g)", this, weight);
683 sharingWeight_ = weight;
684 getModel()->getMaxminSystem()->update_variable_weight(getVariable(), weight);
686 if (getModel()->getUpdateMechanism() == UM_LAZY)
687 heapRemove(getModel()->getActionHeap());
691 void Action::cancel(){
692 setState(Action::State::failed);
693 if (getModel()->getUpdateMechanism() == UM_LAZY) {
694 if (action_lmm_hook.is_linked())
695 simgrid::xbt::intrusive_erase(*getModel()->getModifiedSet(), *this);
696 heapRemove(getModel()->getActionHeap());
703 if (action_hook.is_linked())
704 simgrid::xbt::intrusive_erase(*stateSet_, *this);
706 getModel()->getMaxminSystem()->variable_free(getVariable());
707 if (getModel()->getUpdateMechanism() == UM_LAZY) {
708 /* remove from heap */
709 heapRemove(getModel()->getActionHeap());
710 if (action_lmm_hook.is_linked())
711 simgrid::xbt::intrusive_erase(*getModel()->getModifiedSet(), *this);
719 void Action::suspend()
721 XBT_IN("(%p)", this);
722 if (suspended_ != 2) {
723 getModel()->getMaxminSystem()->update_variable_weight(getVariable(), 0.0);
724 if (getModel()->getUpdateMechanism() == UM_LAZY){
725 heapRemove(getModel()->getActionHeap());
726 if (getModel()->getUpdateMechanism() == UM_LAZY && stateSet_ == getModel()->getRunningActionSet() &&
727 sharingWeight_ > 0) {
728 //If we have a lazy model, we need to update the remaining value accordingly
729 updateRemainingLazy(surf_get_clock());
737 void Action::resume()
739 XBT_IN("(%p)", this);
740 if (suspended_ != 2) {
741 getModel()->getMaxminSystem()->update_variable_weight(getVariable(), getPriority());
743 if (getModel()->getUpdateMechanism() == UM_LAZY)
744 heapRemove(getModel()->getActionHeap());
749 bool Action::isSuspended()
751 return suspended_ == 1;
753 /* insert action on heap using a given key and a hat (heap_action_type)
754 * a hat can be of three types for communications:
756 * NORMAL = this is a normal heap entry stating the date to finish transmitting
757 * LATENCY = this is a heap entry to warn us when the latency is payed
758 * MAX_DURATION =this is a heap entry to warn us when the max_duration limit is reached
760 void Action::heapInsert(heap_type& heap, double key, enum heap_action_type hat)
763 heapHandle_ = heap.emplace(std::make_pair(key, this));
766 void Action::heapRemove(heap_type& heap)
770 heap.erase(*heapHandle_);
775 void Action::heapUpdate(heap_type& heap, double key, enum heap_action_type hat)
779 heap.update(*heapHandle_, std::make_pair(key, this));
781 heapHandle_ = heap.emplace(std::make_pair(key, this));
785 double Action::getRemains()
787 XBT_IN("(%p)", this);
788 /* update remains before return it */
789 if (getModel()->getUpdateMechanism() == UM_LAZY) /* update remains before return it */
790 updateRemainingLazy(surf_get_clock());