X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/7f3527d281baac664d10ce7086e0d0138e1389d7..58801abf7076bc6bd09c971104f6dbd3ff236fc6:/src/surf/ptask_L07.cpp diff --git a/src/surf/ptask_L07.cpp b/src/surf/ptask_L07.cpp index 069dc37556..18cba8cae7 100644 --- a/src/surf/ptask_L07.cpp +++ b/src/surf/ptask_L07.cpp @@ -1,18 +1,14 @@ -/* Copyright (c) 2007-2010, 2013-2017. The SimGrid Team. - * All rights reserved. */ +/* Copyright (c) 2007-2019. The SimGrid Team. All rights reserved. */ /* This program is free software; you can redistribute it and/or modify it * under the terms of the license (GNU LGPL) which comes with this package. */ -#include - -#include -#include - #include "ptask_L07.hpp" +#include "src/kernel/resource/profile/Event.hpp" +#include "surf/surf.hpp" +#include "xbt/config.hpp" -#include "cpu_interface.hpp" -#include "xbt/utility.hpp" +#include XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_host); XBT_LOG_EXTERNAL_CATEGORY(xbt_cfg); @@ -23,22 +19,21 @@ XBT_LOG_EXTERNAL_CATEGORY(xbt_cfg); void surf_host_model_init_ptask_L07() { XBT_CINFO(xbt_cfg,"Switching to the L07 model to handle parallel tasks."); - xbt_assert(not surf_cpu_model_pm, "CPU model type already defined"); - xbt_assert(not surf_network_model, "network model type already defined"); + xbt_assert(not surf_cpu_model_pm, "Cannot switch to ptasks: CPU model already defined"); + xbt_assert(not surf_network_model, "Cannot switch to ptasks: network model already defined"); surf_host_model = new simgrid::surf::HostL07Model(); - all_existing_models->push_back(surf_host_model); + all_existing_models.push_back(surf_host_model); } - namespace simgrid { namespace surf { HostL07Model::HostL07Model() : HostModel() { - maxminSystem_ = new simgrid::kernel::lmm::System(true /* lazy */); - maxminSystem_->solve_fun = &simgrid::kernel::lmm::bottleneck_solve; - surf_network_model = new NetworkL07Model(this,maxminSystem_); - surf_cpu_model_pm = new CpuL07Model(this,maxminSystem_); + auto* maxmin_system = new simgrid::kernel::lmm::FairBottleneck(true /* selective update */); + set_maxmin_system(maxmin_system); + surf_network_model = new NetworkL07Model(this, maxmin_system); + surf_cpu_model_pm = new CpuL07Model(this, maxmin_system); } HostL07Model::~HostL07Model() @@ -47,35 +42,37 @@ HostL07Model::~HostL07Model() delete surf_cpu_model_pm; } -CpuL07Model::CpuL07Model(HostL07Model* hmodel, lmm_system_t sys) : CpuModel(), hostModel_(hmodel) +CpuL07Model::CpuL07Model(HostL07Model* hmodel, kernel::lmm::System* sys) + : CpuModel(Model::UpdateAlgo::FULL), hostModel_(hmodel) { - maxminSystem_ = sys; + set_maxmin_system(sys); } CpuL07Model::~CpuL07Model() { - maxminSystem_ = nullptr; + set_maxmin_system(nullptr); } -NetworkL07Model::NetworkL07Model(HostL07Model* hmodel, lmm_system_t sys) : NetworkModel(), hostModel_(hmodel) +NetworkL07Model::NetworkL07Model(HostL07Model* hmodel, kernel::lmm::System* sys) + : NetworkModel(Model::UpdateAlgo::FULL), hostModel_(hmodel) { - maxminSystem_ = sys; - loopback_ = NetworkL07Model::createLink("__loopback__", 498000000, 0.000015, SURF_LINK_FATPIPE); + set_maxmin_system(sys); + loopback_ = NetworkL07Model::create_link("__loopback__", 498000000, 0.000015, s4u::Link::SharingPolicy::FATPIPE); } NetworkL07Model::~NetworkL07Model() { - maxminSystem_ = nullptr; + set_maxmin_system(nullptr); } -double HostL07Model::nextOccuringEvent(double now) +double HostL07Model::next_occuring_event(double now) { - double min = HostModel::nextOccuringEventFull(now); - for (Action const& action : *getRunningActionSet()) { + double min = HostModel::next_occuring_event_full(now); + for (kernel::resource::Action const& action : *get_started_action_set()) { const L07Action& net_action = static_cast(action); if (net_action.latency_ > 0 && (min < 0 || net_action.latency_ < min)) { min = net_action.latency_; - XBT_DEBUG("Updating min with %p (start %f): %f", &net_action, net_action.getStartTime(), min); + XBT_DEBUG("Updating min with %p (start %f): %f", &net_action, net_action.get_start_time(), min); } } XBT_DEBUG("min value: %f", min); @@ -83,9 +80,9 @@ double HostL07Model::nextOccuringEvent(double now) return min; } -void HostL07Model::updateActionsState(double /*now*/, double delta) +void HostL07Model::update_actions_state(double /*now*/, double delta) { - for (auto it = std::begin(*getRunningActionSet()); it != std::end(*getRunningActionSet());) { + for (auto it = std::begin(*get_started_action_set()); it != std::end(*get_started_action_set());) { L07Action& action = static_cast(*it); ++it; // increment iterator here since the following calls to action.finish() may invalidate it if (action.latency_ > 0) { @@ -94,19 +91,18 @@ void HostL07Model::updateActionsState(double /*now*/, double delta) } else { action.latency_ = 0.0; } - if ((action.latency_ <= 0.0) && (action.isSuspended() == 0)) { + if ((action.latency_ <= 0.0) && (action.is_suspended() == 0)) { action.updateBound(); - maxminSystem_->update_variable_weight(action.getVariable(), 1.0); + get_maxmin_system()->update_variable_penalty(action.get_variable(), 1.0); + action.set_last_update(); } } - XBT_DEBUG("Action (%p) : remains (%g) updated by %g.", &action, action.getRemains(), - action.getVariable()->get_value() * delta); - action.updateRemains(action.getVariable()->get_value() * delta); + XBT_DEBUG("Action (%p) : remains (%g) updated by %g.", &action, action.get_remains(), + action.get_variable()->get_value() * delta); + action.update_remains(action.get_variable()->get_value() * delta); + action.update_max_duration(delta); - if (action.getMaxDuration() > NO_MAX_DURATION) - action.updateMaxDuration(delta); - - XBT_DEBUG("Action (%p) : remains (%g).", &action, action.getRemains()); + XBT_DEBUG("Action (%p) : remains (%g).", &action, action.get_remains()); /* In the next if cascade, the action can be finished either because: * - The amount of remaining work reached 0 @@ -114,126 +110,122 @@ void HostL07Model::updateActionsState(double /*now*/, double delta) * If it's not done, it may have failed. */ - if (((action.getRemains() <= 0) && (action.getVariable()->get_weight() > 0)) || - ((action.getMaxDuration() > NO_MAX_DURATION) && (action.getMaxDuration() <= 0))) { - action.finish(Action::State::done); - } else { - /* Need to check that none of the model has failed */ - int i = 0; - kernel::lmm::Constraint* cnst = action.getVariable()->get_constraint(i); - while (cnst != nullptr) { - i++; - void* constraint_id = cnst->get_id(); - if (static_cast(constraint_id)->isOff()) { - XBT_DEBUG("Action (%p) Failed!!", &action); - action.finish(Action::State::failed); - break; - } - cnst = action.getVariable()->get_constraint(i); + if (((action.get_remains() <= 0) && (action.get_variable()->get_penalty() > 0)) || + ((action.get_max_duration() != NO_MAX_DURATION) && (action.get_max_duration() <= 0))) { + action.finish(kernel::resource::Action::State::FINISHED); + continue; + } + + /* Need to check that none of the model has failed */ + int i = 0; + kernel::lmm::Constraint* cnst = action.get_variable()->get_constraint(i); + while (cnst != nullptr) { + i++; + kernel::resource::Resource* constraint_id = cnst->get_id(); + if (not constraint_id->is_on()) { + XBT_DEBUG("Action (%p) Failed!!", &action); + action.finish(kernel::resource::Action::State::FAILED); + break; } + cnst = action.get_variable()->get_constraint(i); } } } -Action *HostL07Model::executeParallelTask(int host_nb, sg_host_t *host_list, - double *flops_amount, double *bytes_amount,double rate) { - return new L07Action(this, host_nb, host_list, flops_amount, bytes_amount, rate); +kernel::resource::CpuAction* HostL07Model::execute_parallel(const std::vector& host_list, + const double* flops_amount, const double* bytes_amount, + double rate) +{ + return new L07Action(this, host_list, flops_amount, bytes_amount, rate); } -L07Action::L07Action(Model *model, int host_nb, sg_host_t *host_list, - double *flops_amount, double *bytes_amount, double rate) - : CpuAction(model, 1, 0) - , computationAmount_(flops_amount) - , communicationAmount_(bytes_amount) - , rate_(rate) +L07Action::L07Action(kernel::resource::Model* model, const std::vector& host_list, + const double* flops_amount, const double* bytes_amount, double rate) + : CpuAction(model, 1, 0), computationAmount_(flops_amount), communicationAmount_(bytes_amount), rate_(rate) { - int nb_link = 0; - int nb_used_host = 0; /* Only the hosts with something to compute (>0 flops) are counted) */ + size_t link_nb = 0; + size_t used_host_nb = 0; /* Only the hosts with something to compute (>0 flops) are counted) */ double latency = 0.0; + this->set_last_update(); - this->hostList_->reserve(host_nb); - for (int i = 0; i < host_nb; i++) { - this->hostList_->push_back(host_list[i]); - if (flops_amount[i] > 0) - nb_used_host++; - } + hostList_.insert(hostList_.end(), host_list.begin(), host_list.end()); + + if (flops_amount != nullptr) + used_host_nb += std::count_if(flops_amount, flops_amount + host_list.size(), [](double x) { return x > 0.0; }); /* Compute the number of affected resources... */ if(bytes_amount != nullptr) { std::unordered_set affected_links; - for (int i = 0; i < host_nb; i++) { - for (int j = 0; j < host_nb; j++) { - - if (bytes_amount[i * host_nb + j] > 0) { - double lat=0.0; + for (size_t k = 0; k < host_list.size() * host_list.size(); k++) { + if (bytes_amount[k] <= 0) + continue; - std::vector route; - hostList_->at(i)->routeTo(hostList_->at(j), route, &lat); - latency = std::max(latency, lat); + double lat = 0.0; + std::vector route; + hostList_[k / host_list.size()]->route_to(hostList_[k % host_list.size()], route, &lat); + latency = std::max(latency, lat); - for (auto const& link : route) - affected_links.insert(link->getCname()); - } - } + for (auto const& link : route) + affected_links.insert(link->get_cname()); } - nb_link = affected_links.size(); + link_nb = affected_links.size(); } - XBT_DEBUG("Creating a parallel task (%p) with %d hosts and %d unique links.", this, host_nb, nb_link); + XBT_DEBUG("Creating a parallel task (%p) with %zu hosts and %zu unique links.", this, host_list.size(), link_nb); latency_ = latency; - setVariable(model->getMaxminSystem()->variable_new(this, 1.0, (rate > 0 ? rate : -1.0), host_nb + nb_link)); + set_variable( + model->get_maxmin_system()->variable_new(this, 1.0, (rate > 0 ? rate : -1.0), host_list.size() + link_nb)); if (latency_ > 0) - model->getMaxminSystem()->update_variable_weight(getVariable(), 0.0); - - for (int i = 0; i < host_nb; i++) - model->getMaxminSystem()->expand(host_list[i]->pimpl_cpu->constraint(), getVariable(), flops_amount[i]); - - if(bytes_amount != nullptr) { - for (int i = 0; i < host_nb; i++) { - for (int j = 0; j < host_nb; j++) { - if (bytes_amount[i * host_nb + j] > 0.0) { - std::vector route; - hostList_->at(i)->routeTo(hostList_->at(j), route, nullptr); - - for (auto const& link : route) - model->getMaxminSystem()->expand_add(link->constraint(), this->getVariable(), - bytes_amount[i * host_nb + j]); - } - } + model->get_maxmin_system()->update_variable_penalty(get_variable(), 0.0); + + /* Expand it for the CPUs even if there is nothing to compute, to make sure that it gets expended even if there is no + * communication either */ + for (size_t i = 0; i < host_list.size(); i++) + model->get_maxmin_system()->expand(host_list[i]->pimpl_cpu->get_constraint(), get_variable(), + (flops_amount == nullptr ? 0.0 : flops_amount[i])); + + if (bytes_amount != nullptr) { + for (size_t k = 0; k < host_list.size() * host_list.size(); k++) { + if (bytes_amount[k] <= 0.0) + continue; + std::vector route; + hostList_[k / host_list.size()]->route_to(hostList_[k % host_list.size()], route, nullptr); + + for (auto const& link : route) + model->get_maxmin_system()->expand_add(link->get_constraint(), this->get_variable(), bytes_amount[k]); } } - if (nb_link + nb_used_host == 0) { - this->setCost(1.0); - this->setRemains(0.0); + if (link_nb + used_host_nb == 0) { + this->set_cost(1.0); + this->set_remains(0.0); } - delete[] host_list; } -Action* NetworkL07Model::communicate(s4u::Host* src, s4u::Host* dst, double size, double rate) +kernel::resource::Action* NetworkL07Model::communicate(s4u::Host* src, s4u::Host* dst, double size, double rate) { - sg_host_t* host_list = new sg_host_t[2](); + std::vector host_list = {src, dst}; double* flops_amount = new double[2](); double* bytes_amount = new double[4](); - host_list[0] = src; - host_list[1] = dst; bytes_amount[1] = size; - return hostModel_->executeParallelTask(2, host_list, flops_amount, bytes_amount, rate); + kernel::resource::Action* res = hostModel_->execute_parallel(host_list, flops_amount, bytes_amount, rate); + static_cast(res)->free_arrays_ = true; + return res; } -Cpu *CpuL07Model::createCpu(simgrid::s4u::Host *host, std::vector *speedPerPstate, int core) +kernel::resource::Cpu* CpuL07Model::create_cpu(s4u::Host* host, const std::vector& speed_per_pstate, int core) { - return new CpuL07(this, host, speedPerPstate, core); + return new CpuL07(this, host, speed_per_pstate, core); } -LinkImpl* NetworkL07Model::createLink(const std::string& name, double bandwidth, double latency, - e_surf_link_sharing_policy_t policy) +kernel::resource::LinkImpl* NetworkL07Model::create_link(const std::string& name, double bandwidth, double latency, + s4u::Link::SharingPolicy policy) { return new LinkL07(this, name, bandwidth, latency, policy); } @@ -242,128 +234,138 @@ LinkImpl* NetworkL07Model::createLink(const std::string& name, double bandwidth, * Resource * ************/ -CpuL07::CpuL07(CpuL07Model* model, simgrid::s4u::Host* host, std::vector* speedPerPstate, int core) - : Cpu(model, host, model->getMaxminSystem()->constraint_new(this, speedPerPstate->front()), speedPerPstate, core) +CpuL07::CpuL07(CpuL07Model* model, simgrid::s4u::Host* host, const std::vector& speed_per_pstate, int core) + : Cpu(model, host, model->get_maxmin_system()->constraint_new(this, speed_per_pstate.front()), speed_per_pstate, + core) { } CpuL07::~CpuL07()=default; LinkL07::LinkL07(NetworkL07Model* model, const std::string& name, double bandwidth, double latency, - e_surf_link_sharing_policy_t policy) - : LinkImpl(model, name, model->getMaxminSystem()->constraint_new(this, bandwidth)) + s4u::Link::SharingPolicy policy) + : LinkImpl(model, name, model->get_maxmin_system()->constraint_new(this, bandwidth)) { bandwidth_.peak = bandwidth; latency_.peak = latency; - if (policy == SURF_LINK_FATPIPE) - constraint()->unshare(); + if (policy == s4u::Link::SharingPolicy::FATPIPE) + get_constraint()->unshare(); - s4u::Link::onCreation(this->piface_); + s4u::Link::on_creation(this->piface_); } -Action *CpuL07::execution_start(double size) +kernel::resource::CpuAction* CpuL07::execution_start(double size) { - sg_host_t* host_list = new sg_host_t[1](); - double* flops_amount = new double[1](); + std::vector host_list = {get_host()}; - host_list[0] = getHost(); + double* flops_amount = new double[host_list.size()](); flops_amount[0] = size; - return static_cast(model())->hostModel_->executeParallelTask(1, host_list, flops_amount, nullptr, -1); + kernel::resource::CpuAction* res = + static_cast(get_model())->hostModel_->execute_parallel(host_list, flops_amount, nullptr, -1); + static_cast(res)->free_arrays_ = true; + return res; } -Action *CpuL07::sleep(double duration) +kernel::resource::CpuAction* CpuL07::sleep(double duration) { L07Action *action = static_cast(execution_start(1.0)); - action->setMaxDuration(duration); - action->suspended_ = Action::SuspendStates::sleeping; - model()->getMaxminSystem()->update_variable_weight(action->getVariable(), 0.0); + action->set_max_duration(duration); + action->set_suspend_state(kernel::resource::Action::SuspendStates::SLEEPING); + get_model()->get_maxmin_system()->update_variable_penalty(action->get_variable(), 0.0); return action; } -bool CpuL07::isUsed(){ - return model()->getMaxminSystem()->constraint_used(constraint()); +bool CpuL07::is_used() +{ + return get_model()->get_maxmin_system()->constraint_used(get_constraint()); } /** @brief take into account changes of speed (either load or max) */ -void CpuL07::onSpeedChange() { +void CpuL07::on_speed_change() +{ kernel::lmm::Variable* var = nullptr; - const_lmm_element_t elem = nullptr; + const kernel::lmm::Element* elem = nullptr; - model()->getMaxminSystem()->update_constraint_bound(constraint(), speed_.peak * speed_.scale); - while ((var = constraint()->get_variable(&elem))) { - Action* action = static_cast(var->get_id()); + get_model()->get_maxmin_system()->update_constraint_bound(get_constraint(), speed_.peak * speed_.scale); + while ((var = get_constraint()->get_variable(&elem))) { + kernel::resource::Action* action = static_cast(var->get_id()); - model()->getMaxminSystem()->update_variable_bound(action->getVariable(), speed_.scale * speed_.peak); + get_model()->get_maxmin_system()->update_variable_bound(action->get_variable(), speed_.scale * speed_.peak); } - Cpu::onSpeedChange(); + Cpu::on_speed_change(); } - -bool LinkL07::isUsed(){ - return model()->getMaxminSystem()->constraint_used(constraint()); +bool LinkL07::is_used() +{ + return get_model()->get_maxmin_system()->constraint_used(get_constraint()); } -void CpuL07::apply_event(tmgr_trace_event_t triggered, double value) +void CpuL07::apply_event(kernel::profile::Event* triggered, double value) { - XBT_DEBUG("Updating cpu %s (%p) with value %g", getCname(), this, value); + XBT_DEBUG("Updating cpu %s (%p) with value %g", get_cname(), this, value); if (triggered == speed_.event) { speed_.scale = value; - onSpeedChange(); + on_speed_change(); tmgr_trace_event_unref(&speed_.event); - } else if (triggered == stateEvent_) { - if (value > 0) - turnOn(); - else - turnOff(); - tmgr_trace_event_unref(&stateEvent_); + } else if (triggered == state_event_) { + if (value > 0) { + if (not is_on()) { + XBT_VERB("Restart processes on host %s", get_host()->get_cname()); + get_host()->turn_on(); + } + } else + get_host()->turn_off(); + tmgr_trace_event_unref(&state_event_); } else { xbt_die("Unknown event!\n"); } } -void LinkL07::apply_event(tmgr_trace_event_t triggered, double value) +void LinkL07::apply_event(kernel::profile::Event* triggered, double value) { - XBT_DEBUG("Updating link %s (%p) with value=%f", getCname(), this, value); + XBT_DEBUG("Updating link %s (%p) with value=%f", get_cname(), this, value); if (triggered == bandwidth_.event) { - setBandwidth(value); + set_bandwidth(value); tmgr_trace_event_unref(&bandwidth_.event); } else if (triggered == latency_.event) { - setLatency(value); + set_latency(value); tmgr_trace_event_unref(&latency_.event); - } else if (triggered == stateEvent_) { + } else if (triggered == state_event_) { if (value > 0) - turnOn(); + turn_on(); else - turnOff(); - tmgr_trace_event_unref(&stateEvent_); + turn_off(); + tmgr_trace_event_unref(&state_event_); } else { xbt_die("Unknown event ! \n"); } } -void LinkL07::setBandwidth(double value) +void LinkL07::set_bandwidth(double value) { bandwidth_.peak = value; - model()->getMaxminSystem()->update_constraint_bound(constraint(), bandwidth_.peak * bandwidth_.scale); + LinkImpl::on_bandwidth_change(); + + get_model()->get_maxmin_system()->update_constraint_bound(get_constraint(), bandwidth_.peak * bandwidth_.scale); } -void LinkL07::setLatency(double value) +void LinkL07::set_latency(double value) { kernel::lmm::Variable* var = nullptr; L07Action *action; - const_lmm_element_t elem = nullptr; + const kernel::lmm::Element* elem = nullptr; latency_.peak = value; - while ((var = constraint()->get_variable(&elem))) { + while ((var = get_constraint()->get_variable(&elem))) { action = static_cast(var->get_id()); action->updateBound(); } @@ -374,54 +376,42 @@ LinkL07::~LinkL07() = default; * Action * **********/ -L07Action::~L07Action(){ - delete hostList_; - delete[] communicationAmount_; - delete[] computationAmount_; +L07Action::~L07Action() +{ + if (free_arrays_) { + delete[] computationAmount_; + delete[] communicationAmount_; + } } void L07Action::updateBound() { double lat_current = 0.0; - int hostNb = hostList_->size(); + size_t host_count = hostList_.size(); if (communicationAmount_ != nullptr) { - for (int i = 0; i < hostNb; i++) { - for (int j = 0; j < hostNb; j++) { + for (size_t i = 0; i < host_count; i++) { + for (size_t j = 0; j < host_count; j++) { - if (communicationAmount_[i * hostNb + j] > 0) { + if (communicationAmount_[i * host_count + j] > 0) { double lat = 0.0; - std::vector route; - hostList_->at(i)->routeTo(hostList_->at(j), route, &lat); + std::vector route; + hostList_.at(i)->route_to(hostList_.at(j), route, &lat); - lat_current = std::max(lat_current, lat * communicationAmount_[i * hostNb + j]); + lat_current = std::max(lat_current, lat * communicationAmount_[i * host_count + j]); } } } } - double lat_bound = sg_tcp_gamma / (2.0 * lat_current); + double lat_bound = kernel::resource::NetworkModel::cfg_tcp_gamma / (2.0 * lat_current); XBT_DEBUG("action (%p) : lat_bound = %g", this, lat_bound); - if ((latency_ <= 0.0) && (suspended_ == Action::SuspendStates::not_suspended)) { + if ((latency_ <= 0.0) && is_running()) { if (rate_ < 0) - getModel()->getMaxminSystem()->update_variable_bound(getVariable(), lat_bound); + get_model()->get_maxmin_system()->update_variable_bound(get_variable(), lat_bound); else - getModel()->getMaxminSystem()->update_variable_bound(getVariable(), std::min(rate_, lat_bound)); - } -} - -int L07Action::unref() -{ - refcount_--; - if (not refcount_) { - if (action_hook.is_linked()) - simgrid::xbt::intrusive_erase(*stateSet_, *this); - if (getVariable()) - getModel()->getMaxminSystem()->variable_free(getVariable()); - delete this; - return 1; + get_model()->get_maxmin_system()->update_variable_bound(get_variable(), std::min(rate_, lat_bound)); } - return 0; } }