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"
23 XBT_LOG_NEW_CATEGORY(surf, "All SURF categories");
24 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_kernel, surf, "Logging specific to SURF (kernel)");
30 std::vector<surf_model_t> * all_existing_models = nullptr; /* to destroy models correctly */
32 simgrid::trace_mgr::future_evt_set *future_evt_set = nullptr;
33 std::vector<std::string> surf_path;
34 std::vector<simgrid::s4u::Host*> host_that_restart;
35 /** set of hosts for which one want to be notified if they ever restart. */
36 std::set<std::string> watched_hosts;
37 extern std::map<std::string, simgrid::surf::StorageType*> storage_types;
42 simgrid::xbt::signal<void()> surfExitCallbacks;
46 #include <simgrid/plugins/energy.h> // FIXME: this plug-in should not be linked to the core
47 #include <simgrid/plugins/load.h> // FIXME: this plug-in should not be linked to the core
49 s_surf_model_description_t surf_plugin_description[] = {
50 {"Energy", "Cpu energy consumption.", &sg_host_energy_plugin_init},
51 {"Load", "Cpu load.", &sg_host_load_plugin_init},
52 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
55 /* Don't forget to update the option description in smx_config when you change this */
56 s_surf_model_description_t surf_network_model_description[] = {
57 {"LV08", "Realistic network analytic model (slow-start modeled by multiplying latency by 13.01, bandwidth by .97; "
58 "bottleneck sharing uses a payload of S=20537 for evaluating RTT). ",
59 &surf_network_model_init_LegrandVelho},
60 {"Constant", "Simplistic network model where all communication take a constant time (one second). This model "
61 "provides the lowest realism, but is (marginally) faster.",
62 &surf_network_model_init_Constant},
63 {"SMPI", "Realistic network model specifically tailored for HPC settings (accurate modeling of slow start with "
64 "correction factors on three intervals: < 1KiB, < 64 KiB, >= 64 KiB)",
65 &surf_network_model_init_SMPI},
66 {"IB", "Realistic network model specifically tailored for HPC settings, with Infiniband contention model",
67 &surf_network_model_init_IB},
68 {"CM02", "Legacy network analytic model (Very similar to LV08, but without corrective factors. The timings of "
69 "small messages are thus poorly modeled).",
70 &surf_network_model_init_CM02},
71 {"NS3", "Network pseudo-model using the NS3 tcp model instead of an analytic model", &surf_network_model_init_NS3},
73 "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
74 &surf_network_model_init_Reno},
76 "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
77 &surf_network_model_init_Reno2},
79 "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
80 &surf_network_model_init_Vegas},
81 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
85 void surf_network_model_init_SMPI() {
86 xbt_die("Please activate SMPI support in cmake to use the SMPI network model.");
88 void surf_network_model_init_IB() {
89 xbt_die("Please activate SMPI support in cmake to use the IB network model.");
93 void surf_network_model_init_NS3() {
94 xbt_die("Please activate NS3 support in cmake and install the dependencies to use the NS3 network model.");
98 s_surf_model_description_t surf_cpu_model_description[] = {
99 {"Cas01", "Simplistic CPU model (time=size/power).", &surf_cpu_model_init_Cas01},
100 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
103 s_surf_model_description_t surf_host_model_description[] = {
104 {"default", "Default host model. Currently, CPU:Cas01 and network:LV08 (with cross traffic enabled)", &surf_host_model_init_current_default},
105 {"compound", "Host model that is automatically chosen if you change the network and CPU models", &surf_host_model_init_compound},
106 {"ptask_L07", "Host model somehow similar to Cas01+CM02 but allowing parallel tasks", &surf_host_model_init_ptask_L07},
107 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
110 s_surf_model_description_t surf_optimization_mode_description[] = {
111 {"Lazy", "Lazy action management (partial invalidation in lmm + heap in action remaining).", nullptr},
112 {"TI", "Trace integration. Highly optimized mode when using availability traces (only available for the Cas01 CPU model for now).", nullptr},
113 {"Full", "Full update of remaining and variables. Slow but may be useful when debugging.", nullptr},
114 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
117 s_surf_model_description_t surf_storage_model_description[] = {
118 {"default", "Simplistic storage model.", &surf_storage_model_init_default},
119 {nullptr, nullptr, nullptr} /* this array must be nullptr terminated */
124 double surf_get_clock()
130 # define FILE_DELIM "\\"
132 # define FILE_DELIM "/" /* FIXME: move to better location */
135 std::ifstream* surf_ifsopen(std::string name)
137 std::ifstream* fs = new std::ifstream();
138 xbt_assert(not name.empty());
139 if (__surf_is_absolute_file_path(name.c_str())) { /* don't mess with absolute file names */
140 fs->open(name.c_str(), std::ifstream::in);
143 /* search relative files in the path */
144 for (auto const& path_elm : surf_path) {
145 std::string buff = path_elm + FILE_DELIM + name;
146 fs->open(buff.c_str(), std::ifstream::in);
148 if (not fs->fail()) {
149 XBT_DEBUG("Found file at %s", buff.c_str());
157 FILE *surf_fopen(const char *name, const char *mode)
159 FILE *file = nullptr;
163 if (__surf_is_absolute_file_path(name)) /* don't mess with absolute file names */
164 return fopen(name, mode);
166 /* search relative files in the path */
167 for (auto const& path_elm : surf_path) {
168 std::string buff = path_elm + FILE_DELIM + name;
169 file = fopen(buff.c_str(), mode);
177 /* The __surf_is_absolute_file_path() returns 1 if
178 * file_path is a absolute file path, in the other
179 * case the function returns 0.
181 int __surf_is_absolute_file_path(const char *file_path)
184 WIN32_FIND_DATA wfd = { 0 };
185 HANDLE hFile = FindFirstFile(file_path, &wfd);
187 if (INVALID_HANDLE_VALUE == hFile)
193 return (file_path[0] == '/');
197 /** Displays the long description of all registered models, and quit */
198 void model_help(const char *category, s_surf_model_description_t * table)
200 printf("Long description of the %s models accepted by this simulator:\n", category);
201 for (int i = 0; table[i].name; i++)
202 printf(" %s: %s\n", table[i].name, table[i].description);
205 int find_model_description(s_surf_model_description_t* table, std::string name)
207 for (int i = 0; table[i].name; i++)
208 if (name == table[i].name)
211 if (not table[0].name)
212 xbt_die("No model is valid! This is a bug.");
214 std::string name_list = std::string(table[0].name);
215 for (int i = 1; table[i].name; i++)
216 name_list = name_list + ", " + table[i].name;
218 xbt_die("Model '%s' is invalid! Valid models are: %s.", name.c_str(), name_list.c_str());
222 void sg_version_check(int lib_version_major, int lib_version_minor, int lib_version_patch)
224 if ((lib_version_major != SIMGRID_VERSION_MAJOR) || (lib_version_minor != SIMGRID_VERSION_MINOR)) {
225 fprintf(stderr, "FATAL ERROR: Your program was compiled with SimGrid version %d.%d.%d, "
226 "and then linked against SimGrid %d.%d.%d. Please fix this.\n",
227 lib_version_major, lib_version_minor, lib_version_patch, SIMGRID_VERSION_MAJOR, SIMGRID_VERSION_MINOR,
228 SIMGRID_VERSION_PATCH);
231 if (lib_version_patch != SIMGRID_VERSION_PATCH) {
232 if (SIMGRID_VERSION_PATCH >= 90 || lib_version_patch >= 90) {
235 "FATAL ERROR: Your program was compiled with SimGrid version %d.%d.%d, "
236 "and then linked against SimGrid %d.%d.%d. \n"
237 "One of them is a development version, and should not be mixed with the stable release. Please fix this.\n",
238 lib_version_major, lib_version_minor, lib_version_patch, SIMGRID_VERSION_MAJOR, SIMGRID_VERSION_MINOR,
239 SIMGRID_VERSION_PATCH);
242 fprintf(stderr, "Warning: Your program was compiled with SimGrid version %d.%d.%d, "
243 "and then linked against SimGrid %d.%d.%d. Proceeding anyway.\n",
244 lib_version_major, lib_version_minor, lib_version_patch, SIMGRID_VERSION_MAJOR, SIMGRID_VERSION_MINOR,
245 SIMGRID_VERSION_PATCH);
249 void sg_version_get(int* ver_major, int* ver_minor, int* ver_patch)
251 *ver_major = SIMGRID_VERSION_MAJOR;
252 *ver_minor = SIMGRID_VERSION_MINOR;
253 *ver_patch = SIMGRID_VERSION_PATCH;
258 std::printf("This program was linked against %s (git: %s), found in %s.\n",
259 SIMGRID_VERSION_STRING, SIMGRID_GIT_VERSION, SIMGRID_INSTALL_PREFIX);
262 std::printf(" Model-checking support compiled in.\n");
264 std::printf(" Model-checking support disabled at compilation.\n");
268 std::printf(" NS3 support compiled in.\n");
270 std::printf(" NS3 support disabled at compilation.\n");
273 #if SIMGRID_HAVE_JEDULE
274 std::printf(" Jedule support compiled in.\n");
276 std::printf(" Jedule support disabled at compilation.\n");
280 std::printf(" Lua support compiled in.\n");
282 std::printf(" Lua support disabled at compilation.\n");
285 #if SIMGRID_HAVE_MALLOCATOR
286 std::printf(" Mallocator support compiled in.\n");
288 std::printf(" Mallocator support disabled at compilation.\n");
291 std::printf("\nTo cite SimGrid in a publication, please use:\n"
292 " Henri Casanova, Arnaud Giersch, Arnaud Legrand, Martin Quinson, Frédéric Suter. \n"
293 " Versatile, Scalable, and Accurate Simulation of Distributed Applications and Platforms. \n"
294 " Journal of Parallel and Distributed Computing, Elsevier, 2014, 74 (10), pp.2899-2917.\n");
295 std::printf("The pdf file and a BibTeX entry for LaTeX users can be found at http://hal.inria.fr/hal-01017319\n");
298 void surf_init(int *argc, char **argv)
300 if (USER_HOST_LEVEL != -1) // Already initialized
303 XBT_DEBUG("Create all Libs");
304 USER_HOST_LEVEL = simgrid::s4u::Host::extension_create(nullptr);
306 xbt_init(argc, argv);
307 if (not all_existing_models)
308 all_existing_models = new std::vector<simgrid::surf::Model*>();
309 if (not future_evt_set)
310 future_evt_set = new simgrid::trace_mgr::future_evt_set();
312 sg_config_init(argc, argv);
320 TRACE_end(); /* Just in case it was not called by the upper layer (or there is no upper layer) */
324 for (auto const& e : storage_types) {
325 simgrid::surf::StorageType* stype = e.second;
326 delete stype->properties;
327 delete stype->model_properties;
330 for (auto const& s : *simgrid::surf::StorageImpl::storagesMap())
332 delete simgrid::surf::StorageImpl::storagesMap();
334 for (auto const& model : *all_existing_models)
336 delete all_existing_models;
338 simgrid::surf::surfExitCallbacks();
340 if (future_evt_set) {
341 delete future_evt_set;
342 future_evt_set = nullptr;
347 simgrid::s4u::Engine::shutdown();
349 NOW = 0; /* Just in case the user plans to restart the simulation afterward */
360 : maxminSystem_(nullptr)
362 readyActionSet_ = new ActionList();
363 runningActionSet_ = new ActionList();
364 failedActionSet_ = new ActionList();
365 doneActionSet_ = new ActionList();
367 modifiedSet_ = nullptr;
368 updateMechanism_ = UM_UNDEFINED;
369 selectiveUpdate_ = 0;
373 delete readyActionSet_;
374 delete runningActionSet_;
375 delete failedActionSet_;
376 delete doneActionSet_;
379 Action* Model::actionHeapPop()
381 Action* action = actionHeap_.top().second;
383 action->clearHeapHandle();
387 double Model::nextOccuringEvent(double now)
389 //FIXME: set the good function once and for all
390 if (updateMechanism_ == UM_LAZY)
391 return nextOccuringEventLazy(now);
392 else if (updateMechanism_ == UM_FULL)
393 return nextOccuringEventFull(now);
395 xbt_die("Invalid cpu update mechanism!");
398 double Model::nextOccuringEventLazy(double now)
400 XBT_DEBUG("Before share resources, the size of modified actions set is %zu", modifiedSet_->size());
401 lmm_solve(maxminSystem_);
402 XBT_DEBUG("After share resources, The size of modified actions set is %zu", modifiedSet_->size());
404 while (not modifiedSet_->empty()) {
405 Action *action = &(modifiedSet_->front());
406 modifiedSet_->pop_front();
407 bool max_dur_flag = false;
409 if (action->getStateSet() != runningActionSet_)
412 /* bogus priority, skip it */
413 if (action->getPriority() <= 0 || action->getHat()==LATENCY)
416 action->updateRemainingLazy(now);
419 double share = action->getVariable()->get_value();
422 double time_to_completion;
423 if (action->getRemains() > 0) {
424 time_to_completion = action->getRemainsNoUpdate() / share;
426 time_to_completion = 0.0;
428 min = now + time_to_completion; // when the task will complete if nothing changes
431 if ((action->getMaxDuration() > NO_MAX_DURATION) &&
432 (min <= -1 || action->getStartTime() + action->getMaxDuration() < min)) {
433 // when the task will complete anyway because of the deadline if any
434 min = action->getStartTime() + action->getMaxDuration();
438 XBT_DEBUG("Action(%p) corresponds to variable %d", action, action->getVariable()->id_int);
440 XBT_DEBUG("Action(%p) Start %f. May finish at %f (got a share of %f). Max_duration %f", action,
441 action->getStartTime(), min, share,
442 action->getMaxDuration());
445 action->heapUpdate(actionHeap_, min, max_dur_flag ? MAX_DURATION : NORMAL);
446 XBT_DEBUG("Insert at heap action(%p) min %f now %f", action, min, now);
451 //hereafter must have already the min value for this resource model
452 if (not actionHeapIsEmpty()) {
453 double min = actionHeapTopDate() - now;
454 XBT_DEBUG("minimum with the HEAP %f", min);
457 XBT_DEBUG("The HEAP is empty, thus returning -1");
462 double Model::nextOccuringEventFull(double /*now*/) {
463 maxminSystem_->solve_fun(maxminSystem_);
467 for (Action& action : *getRunningActionSet()) {
468 double value = action.getVariable()->get_value();
470 if (action.getRemains() > 0)
471 value = action.getRemainsNoUpdate() / value;
474 if (min < 0 || value < min) {
476 XBT_DEBUG("Updating min (value) with %p: %f", &action, min);
479 if ((action.getMaxDuration() >= 0) && (min < 0 || action.getMaxDuration() < min)) {
480 min = action.getMaxDuration();
481 XBT_DEBUG("Updating min (duration) with %p: %f", &action, min);
484 XBT_DEBUG("min value : %f", min);
489 void Model::updateActionsState(double now, double delta)
491 if (updateMechanism_ == UM_FULL)
492 updateActionsStateFull(now, delta);
493 else if (updateMechanism_ == UM_LAZY)
494 updateActionsStateLazy(now, delta);
496 xbt_die("Invalid cpu update mechanism!");
499 void Model::updateActionsStateLazy(double /*now*/, double /*delta*/)
504 void Model::updateActionsStateFull(double /*now*/, double /*delta*/)
519 Resource::Resource(Model* model, const std::string& name, lmm_constraint_t constraint)
520 : name_(name), model_(model), constraint_(constraint)
523 Resource::~Resource() = default;
525 bool Resource::isOn() const {
528 bool Resource::isOff() const {
532 void Resource::turnOn()
537 void Resource::turnOff()
542 Model* Resource::model() const
547 const std::string& Resource::getName() const
552 const char* Resource::getCname() const
554 return name_.c_str();
557 bool Resource::operator==(const Resource &other) const {
558 return name_ == other.name_;
561 lmm_constraint_t Resource::constraint() const
573 const char *surf_action_state_names[6] = {
575 "SURF_ACTION_RUNNING",
576 "SURF_ACTION_FAILED",
578 "SURF_ACTION_TO_FREE",
579 "SURF_ACTION_NOT_IN_THE_SYSTEM"
585 Action::Action(simgrid::surf::Model* model, double cost, bool failed) : Action(model, cost, failed, nullptr)
589 Action::Action(simgrid::surf::Model* model, double cost, bool failed, lmm_variable_t var)
590 : remains_(cost), start_(surf_get_clock()), cost_(cost), model_(model), variable_(var)
593 stateSet_ = getModel()->getFailedActionSet();
595 stateSet_ = getModel()->getRunningActionSet();
597 stateSet_->push_back(*this);
604 void Action::finish(Action::State state)
606 finishTime_ = surf_get_clock();
610 Action::State Action::getState() const
612 if (stateSet_ == model_->getReadyActionSet())
613 return Action::State::ready;
614 if (stateSet_ == model_->getRunningActionSet())
615 return Action::State::running;
616 if (stateSet_ == model_->getFailedActionSet())
617 return Action::State::failed;
618 if (stateSet_ == model_->getDoneActionSet())
619 return Action::State::done;
620 return Action::State::not_in_the_system;
623 void Action::setState(Action::State state)
625 stateSet_->erase(stateSet_->iterator_to(*this));
627 case Action::State::ready:
628 stateSet_ = model_->getReadyActionSet();
630 case Action::State::running:
631 stateSet_ = model_->getRunningActionSet();
633 case Action::State::failed:
634 stateSet_ = model_->getFailedActionSet();
636 case Action::State::done:
637 stateSet_ = model_->getDoneActionSet();
644 stateSet_->push_back(*this);
647 double Action::getBound() const
649 return variable_ ? variable_->get_bound() : 0;
652 void Action::setBound(double bound)
654 XBT_IN("(%p,%g)", this, bound);
656 getModel()->getMaxminSystem()->update_variable_bound(variable_, bound);
658 if (getModel()->getUpdateMechanism() == UM_LAZY && getLastUpdate() != surf_get_clock())
659 heapRemove(getModel()->getActionHeap());
663 void Action::setCategory(const char *category)
665 category_ = xbt_strdup(category);
672 void Action::setMaxDuration(double duration)
674 maxDuration_ = duration;
675 if (getModel()->getUpdateMechanism() == UM_LAZY) // remove action from the heap
676 heapRemove(getModel()->getActionHeap());
679 void Action::setSharingWeight(double weight)
681 XBT_IN("(%p,%g)", this, weight);
682 sharingWeight_ = weight;
683 getModel()->getMaxminSystem()->update_variable_weight(getVariable(), weight);
685 if (getModel()->getUpdateMechanism() == UM_LAZY)
686 heapRemove(getModel()->getActionHeap());
690 void Action::cancel(){
691 setState(Action::State::failed);
692 if (getModel()->getUpdateMechanism() == UM_LAZY) {
693 if (action_lmm_hook.is_linked())
694 getModel()->getModifiedSet()->erase(getModel()->getModifiedSet()->iterator_to(*this));
695 heapRemove(getModel()->getActionHeap());
702 if (action_hook.is_linked())
703 stateSet_->erase(stateSet_->iterator_to(*this));
705 getModel()->getMaxminSystem()->variable_free(getVariable());
706 if (getModel()->getUpdateMechanism() == UM_LAZY) {
707 /* remove from heap */
708 heapRemove(getModel()->getActionHeap());
709 if (action_lmm_hook.is_linked())
710 getModel()->getModifiedSet()->erase(getModel()->getModifiedSet()->iterator_to(*this));
718 void Action::suspend()
720 XBT_IN("(%p)", this);
721 if (suspended_ != 2) {
722 getModel()->getMaxminSystem()->update_variable_weight(getVariable(), 0.0);
723 if (getModel()->getUpdateMechanism() == UM_LAZY){
724 heapRemove(getModel()->getActionHeap());
725 if (getModel()->getUpdateMechanism() == UM_LAZY && stateSet_ == getModel()->getRunningActionSet() &&
726 sharingWeight_ > 0) {
727 //If we have a lazy model, we need to update the remaining value accordingly
728 updateRemainingLazy(surf_get_clock());
736 void Action::resume()
738 XBT_IN("(%p)", this);
739 if (suspended_ != 2) {
740 getModel()->getMaxminSystem()->update_variable_weight(getVariable(), getPriority());
742 if (getModel()->getUpdateMechanism() == UM_LAZY)
743 heapRemove(getModel()->getActionHeap());
748 bool Action::isSuspended()
750 return suspended_ == 1;
752 /* insert action on heap using a given key and a hat (heap_action_type)
753 * a hat can be of three types for communications:
755 * NORMAL = this is a normal heap entry stating the date to finish transmitting
756 * LATENCY = this is a heap entry to warn us when the latency is payed
757 * MAX_DURATION =this is a heap entry to warn us when the max_duration limit is reached
759 void Action::heapInsert(heap_type& heap, double key, enum heap_action_type hat)
762 heapHandle_ = heap.emplace(std::make_pair(key, this));
765 void Action::heapRemove(heap_type& heap)
769 heap.erase(*heapHandle_);
774 void Action::heapUpdate(heap_type& heap, double key, enum heap_action_type hat)
778 heap.update(*heapHandle_, std::make_pair(key, this));
780 heapHandle_ = heap.emplace(std::make_pair(key, this));
784 double Action::getRemains()
786 XBT_IN("(%p)", this);
787 /* update remains before return it */
788 if (getModel()->getUpdateMechanism() == UM_LAZY) /* update remains before return it */
789 updateRemainingLazy(surf_get_clock());