X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/4c18091c2618f718dcd672ed8391572eeb80ad99..5ed37babb2fa9097abe82df299c0aa259ed84d5a:/src/surf/ptask_L07.cpp diff --git a/src/surf/ptask_L07.cpp b/src/surf/ptask_L07.cpp index 4746890ac0..ac76b6ccdf 100644 --- a/src/surf/ptask_L07.cpp +++ b/src/surf/ptask_L07.cpp @@ -1,49 +1,68 @@ -/* Copyright (c) 2007-2021. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2007-2023. 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 "ptask_L07.hpp" +#include +#include +#include + +#include "simgrid/config.h" +#include "src/kernel/EngineImpl.hpp" +#if SIMGRID_HAVE_EIGEN3 +#include "src/kernel/lmm/bmf.hpp" +#endif #include "src/kernel/resource/profile/Event.hpp" -#include "surf/surf.hpp" -#include "xbt/config.hpp" +#include "src/surf/ptask_L07.hpp" #include XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(res_host); XBT_LOG_EXTERNAL_CATEGORY(xbt_cfg); +/*********** + * Options * + ***********/ +static simgrid::config::Flag cfg_ptask_solver("host/solver", + "Set linear equations solver used by ptask model", + "fairbottleneck", + &simgrid::kernel::lmm::System::validate_solver); + /**************************************/ /*** Resource Creation & Destruction **/ /**************************************/ 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, "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); + XBT_CINFO(xbt_cfg, "Switching to the L07 model to handle parallel tasks."); + xbt_assert(cfg_ptask_solver != "maxmin", "Invalid configuration. Cannot use maxmin solver with parallel tasks."); + + auto* system = simgrid::kernel::lmm::System::build(cfg_ptask_solver.get(), true /* selective update */); + auto host_model = std::make_shared("Host_Ptask", system); + auto* engine = simgrid::kernel::EngineImpl::get_instance(); + engine->add_model(host_model); + engine->get_netzone_root()->set_host_model(host_model); } -namespace simgrid { -namespace surf { +namespace simgrid::kernel::resource { -HostL07Model::HostL07Model() : HostModel() { - 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() +HostL07Model::HostL07Model(const std::string& name, lmm::System* sys) : HostModel(name) { - delete surf_network_model; - delete surf_cpu_model_pm; + set_maxmin_system(sys); + + auto net_model = std::make_shared("Network_Ptask", this, sys); + auto engine = EngineImpl::get_instance(); + engine->add_model(net_model); + engine->get_netzone_root()->set_network_model(net_model); + + auto cpu_model = std::make_shared("Cpu_Ptask", this, sys); + engine->add_model(cpu_model); + engine->get_netzone_root()->set_cpu_pm_model(cpu_model); + + surf_disk_model_init_S19(); } -CpuL07Model::CpuL07Model(HostL07Model* hmodel, kernel::lmm::System* sys) - : CpuModel(Model::UpdateAlgo::FULL), hostModel_(hmodel) +CpuL07Model::CpuL07Model(const std::string& name, HostL07Model* hmodel, lmm::System* sys) + : CpuModel(name), hostModel_(hmodel) { set_maxmin_system(sys); } @@ -53,14 +72,14 @@ CpuL07Model::~CpuL07Model() set_maxmin_system(nullptr); } -NetworkL07Model::NetworkL07Model(HostL07Model* hmodel, kernel::lmm::System* sys) - : NetworkModel(Model::UpdateAlgo::FULL), hostModel_(hmodel) +NetworkL07Model::NetworkL07Model(const std::string& name, HostL07Model* hmodel, lmm::System* sys) + : NetworkModel(name), hostModel_(hmodel) { set_maxmin_system(sys); - loopback_ = NetworkL07Model::create_link("__loopback__", - std::vector{simgrid::config::get_value("network/loopback-bw")}, - s4u::Link::SharingPolicy::FATPIPE)->set_latency(simgrid::config::get_value("network/loopback-lat")); - loopback_->seal(); + loopback_.reset(create_link("__loopback__", {simgrid::config::get_value("network/loopback-bw")})); + loopback_->set_sharing_policy(s4u::Link::SharingPolicy::FATPIPE, {}); + loopback_->set_latency(simgrid::config::get_value("network/loopback-lat")); + loopback_->get_iface()->seal(); } NetworkL07Model::~NetworkL07Model() @@ -71,7 +90,7 @@ NetworkL07Model::~NetworkL07Model() double HostL07Model::next_occurring_event(double now) { double min = HostModel::next_occurring_event_full(now); - for (kernel::resource::Action const& action : *get_started_action_set()) { + for (Action const& action : *get_started_action_set()) { const auto& net_action = static_cast(action); if (net_action.get_latency() > 0 && (min < 0 || net_action.get_latency() < min)) { min = net_action.get_latency(); @@ -95,14 +114,13 @@ void HostL07Model::update_actions_state(double /*now*/, double delta) action.set_latency(0.0); } if ((action.get_latency() <= 0.0) && (action.is_suspended() == 0)) { - action.updateBound(); + action.update_bound(); 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.get_remains(), - action.get_variable()->get_value() * delta); - action.update_remains(action.get_variable()->get_value() * delta); + XBT_DEBUG("Action (%p) : remains (%g) updated by %g.", &action, action.get_remains(), action.get_rate() * delta); + action.update_remains(action.get_rate() * delta); action.update_max_duration(delta); XBT_DEBUG("Action (%p) : remains (%g).", &action, action.get_remains()); @@ -115,19 +133,18 @@ void HostL07Model::update_actions_state(double /*now*/, double delta) 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); + action.finish(Action::State::FINISHED); continue; } /* Need to check that none of the model has failed */ - int i = 0; - const kernel::lmm::Constraint* cnst = action.get_variable()->get_constraint(i); + int i = 0; + const lmm::Constraint* cnst = action.get_variable()->get_constraint(i); while (cnst != nullptr) { i++; - const kernel::resource::Resource* constraint_id = cnst->get_id(); - if (not constraint_id->is_on()) { + if (not cnst->get_id()->is_on()) { XBT_DEBUG("Action (%p) Failed!!", &action); - action.finish(kernel::resource::Action::State::FAILED); + action.finish(Action::State::FAILED); break; } cnst = action.get_variable()->get_constraint(i); @@ -135,38 +152,40 @@ void HostL07Model::update_actions_state(double /*now*/, double delta) } } -kernel::resource::CpuAction* HostL07Model::execute_parallel(const std::vector& host_list, - const double* flops_amount, const double* bytes_amount, - double rate) +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(kernel::resource::Model* model, const std::vector& host_list, - const double* flops_amount, const double* bytes_amount, double rate) - : CpuAction(model, 1.0, false), computationAmount_(flops_amount), communicationAmount_(bytes_amount), rate_(rate) +L07Action::L07Action(Model* model, const std::vector& host_list, const double* flops_amount, + const double* bytes_amount, double rate) + : CpuAction(model, 1.0, false) + , host_list_(host_list) + , computation_amount_(flops_amount) + , communication_amount_(bytes_amount) + , rate_(rate) { - 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; + size_t link_nb = 0; + const size_t host_nb = host_list_.size(); + 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(); - 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; }); + used_host_nb += std::count_if(flops_amount, flops_amount + host_nb, [](double x) { return x > 0.0; }); /* Compute the number of affected resources... */ - if(bytes_amount != nullptr) { + if (bytes_amount != nullptr) { std::unordered_set affected_links; - for (size_t k = 0; k < host_list.size() * host_list.size(); k++) { + for (size_t k = 0; k < host_nb * host_nb; k++) { if (bytes_amount[k] <= 0) continue; double lat = 0.0; - std::vector route; - hostList_[k / host_list.size()]->route_to(hostList_[k % host_list.size()], route, &lat); + std::vector route; + host_list_[k / host_nb]->route_to(host_list_[k % host_nb], route, &lat); latency = std::max(latency, lat); for (auto const& link : route) @@ -176,30 +195,30 @@ L07Action::L07Action(kernel::resource::Model* model, const std::vectorget_maxmin_system()->variable_new(this, 1.0, (rate > 0 ? rate : -1.0), host_list.size() + link_nb)); + set_variable(model->get_maxmin_system()->variable_new(this, 1.0, (rate > 0 ? rate : -1.0), host_nb + link_nb)); if (latency_ > 0) 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])); + for (size_t i = 0; i < host_nb; i++) { + model->get_maxmin_system()->expand(host_list[i]->get_cpu()->get_constraint(), get_variable(), + (flops_amount == nullptr ? 0.0 : flops_amount[i]), true); + } if (bytes_amount != nullptr) { - for (size_t k = 0; k < host_list.size() * host_list.size(); k++) { + for (size_t k = 0; k < host_nb * host_nb; 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); + std::vector route; + host_list_[k / host_nb]->route_to(host_list_[k % host_nb], route, nullptr); for (auto const& link : route) - model->get_maxmin_system()->expand_add(link->get_constraint(), this->get_variable(), bytes_amount[k]); + model->get_maxmin_system()->expand(link->get_constraint(), this->get_variable(), bytes_amount[k]); } } @@ -207,9 +226,11 @@ L07Action::L07Action(kernel::resource::Model* model, const std::vectorset_cost(1.0); this->set_remains(0.0); } + /* finally calculate the initial bound value */ + update_bound(); } -kernel::resource::Action* NetworkL07Model::communicate(s4u::Host* src, s4u::Host* dst, double size, double rate) +Action* NetworkL07Model::communicate(s4u::Host* src, s4u::Host* dst, double size, double rate, bool /* streamed */) { std::vector host_list = {src, dst}; const auto* flops_amount = new double[2](); @@ -217,95 +238,79 @@ kernel::resource::Action* NetworkL07Model::communicate(s4u::Host* src, s4u::Host bytes_amount[1] = size; - kernel::resource::Action* res = hostModel_->execute_parallel(host_list, flops_amount, bytes_amount, rate); + Action* res = hostModel_->execute_parallel(host_list, flops_amount, bytes_amount, rate); static_cast(res)->free_arrays_ = true; return res; } -kernel::resource::Cpu* CpuL07Model::create_cpu(s4u::Host* host, const std::vector& speed_per_pstate, int core) +CpuImpl* CpuL07Model::create_cpu(s4u::Host* host, const std::vector& speed_per_pstate) { - return new CpuL07(this, host, speed_per_pstate, core); + return (new CpuL07(host, speed_per_pstate))->set_model(this); } -kernel::resource::LinkImpl* NetworkL07Model::create_link(const std::string& name, const std::vector& bandwidths, - s4u::Link::SharingPolicy policy) +StandardLinkImpl* NetworkL07Model::create_link(const std::string& name, const std::vector& bandwidths) { xbt_assert(bandwidths.size() == 1, "Non WIFI link must have only 1 bandwidth."); - return new LinkL07(this, name, bandwidths[0], policy); + auto link = new LinkL07(name, bandwidths[0], get_maxmin_system()); + link->set_model(this); + return link; } -/************ - * Resource * - ************/ - -CpuL07::CpuL07(CpuL07Model* model, simgrid::s4u::Host* host, const std::vector& speed_per_pstate, int core) - : Cpu(host, speed_per_pstate) +StandardLinkImpl* NetworkL07Model::create_wifi_link(const std::string& name, const std::vector& bandwidths) { - this->set_core_count(core)->set_model(model)->set_constraint( - model->get_maxmin_system()->constraint_new(this, speed_per_pstate.front())); + THROW_UNIMPLEMENTED; } -CpuL07::~CpuL07()=default; - -LinkL07::LinkL07(NetworkL07Model* model, const std::string& name, double bandwidth, s4u::Link::SharingPolicy policy) - : LinkImpl(model, name, model->get_maxmin_system()->constraint_new(this, bandwidth)) -{ - bandwidth_.peak = bandwidth; - - if (policy == s4u::Link::SharingPolicy::FATPIPE) - get_constraint()->unshare(); -} +/************ + * Resource * + ************/ -kernel::resource::CpuAction* CpuL07::execution_start(double size) +CpuAction* CpuL07::execution_start(double size, double user_bound) { std::vector host_list = {get_iface()}; + xbt_assert(user_bound <= 0, "User bound not supported by ptask model"); auto* flops_amount = new double[host_list.size()](); - flops_amount[0] = size; + flops_amount[0] = size; - kernel::resource::CpuAction* res = + CpuAction* res = static_cast(get_model())->hostModel_->execute_parallel(host_list, flops_amount, nullptr, -1); static_cast(res)->free_arrays_ = true; return res; } -kernel::resource::CpuAction* CpuL07::sleep(double duration) +CpuAction* CpuL07::sleep(double duration) { - auto* action = static_cast(execution_start(1.0)); + auto* action = static_cast(execution_start(1.0, -1)); action->set_max_duration(duration); - action->set_suspend_state(kernel::resource::Action::SuspendStates::SLEEPING); + action->set_suspend_state(Action::SuspendStates::SLEEPING); get_model()->get_maxmin_system()->update_variable_penalty(action->get_variable(), 0.0); return action; } -bool CpuL07::is_used() const -{ - return get_model()->get_maxmin_system()->constraint_used(get_constraint()); -} - /** @brief take into account changes of speed (either load or max) */ void CpuL07::on_speed_change() { - const kernel::lmm::Variable* var; - const kernel::lmm::Element* elem = nullptr; + const lmm::Element* elem = nullptr; - get_model()->get_maxmin_system()->update_constraint_bound(get_constraint(), speed_.peak * speed_.scale); - while ((var = get_constraint()->get_variable(&elem))) { - const kernel::resource::Action* action = var->get_id(); + get_model()->get_maxmin_system()->update_constraint_bound(get_constraint(), get_core_count() * speed_.peak * speed_.scale); - get_model()->get_maxmin_system()->update_variable_bound(action->get_variable(), speed_.scale * speed_.peak); + while (const auto* var = get_constraint()->get_variable(&elem)) { + const auto* action = static_cast(var->get_id()); + action->update_bound(); } - Cpu::on_speed_change(); + CpuImpl::on_speed_change(); } -bool LinkL07::is_used() const +LinkL07::LinkL07(const std::string& name, double bandwidth, lmm::System* system) : StandardLinkImpl(name) { - return get_model()->get_maxmin_system()->constraint_used(get_constraint()); + this->set_constraint(system->constraint_new(this, bandwidth)); + bandwidth_.peak = bandwidth; } -void CpuL07::apply_event(kernel::profile::Event* triggered, double value) +void CpuL07::apply_event(profile::Event* triggered, double value) { XBT_DEBUG("Updating cpu %s (%p) with value %g", get_cname(), this, value); if (triggered == speed_.event) { @@ -313,7 +318,7 @@ void CpuL07::apply_event(kernel::profile::Event* triggered, double value) on_speed_change(); tmgr_trace_event_unref(&speed_.event); - } else if (triggered == state_event_) { + } else if (triggered == get_state_event()) { if (value > 0) { if (not is_on()) { XBT_VERB("Restart actors on host %s", get_iface()->get_cname()); @@ -321,14 +326,14 @@ void CpuL07::apply_event(kernel::profile::Event* triggered, double value) } } else get_iface()->turn_off(); - tmgr_trace_event_unref(&state_event_); + unref_state_event(); } else { xbt_die("Unknown event!\n"); } } -void LinkL07::apply_event(kernel::profile::Event* triggered, double value) +void LinkL07::apply_event(profile::Event* triggered, double value) { XBT_DEBUG("Updating link %s (%p) with value=%f", get_cname(), this, value); if (triggered == bandwidth_.event) { @@ -339,13 +344,12 @@ void LinkL07::apply_event(kernel::profile::Event* triggered, double value) set_latency(value); tmgr_trace_event_unref(&latency_.event); - } else if (triggered == state_event_) { + } else if (triggered == get_state_event()) { if (value > 0) turn_on(); else turn_off(); - tmgr_trace_event_unref(&state_event_); - + unref_state_event(); } else { xbt_die("Unknown event ! \n"); } @@ -354,24 +358,21 @@ void LinkL07::apply_event(kernel::profile::Event* triggered, double value) void LinkL07::set_bandwidth(double value) { bandwidth_.peak = value; - LinkImpl::on_bandwidth_change(); + StandardLinkImpl::on_bandwidth_change(); get_model()->get_maxmin_system()->update_constraint_bound(get_constraint(), bandwidth_.peak * bandwidth_.scale); } -kernel::resource::LinkImpl* LinkL07::set_latency(double value) +void LinkL07::set_latency(double value) { latency_check(value); - const kernel::lmm::Variable* var; - L07Action *action; - const kernel::lmm::Element* elem = nullptr; + const lmm::Element* elem = nullptr; latency_.peak = value; - while ((var = get_constraint()->get_variable(&elem))) { - action = static_cast(var->get_id()); - action->updateBound(); + while (const auto* var = get_constraint()->get_variable(&elem)) { + const auto* action = static_cast(var->get_id()); + action->update_bound(); } - return this; } LinkL07::~LinkL07() = default; @@ -382,39 +383,69 @@ LinkL07::~LinkL07() = default; L07Action::~L07Action() { if (free_arrays_) { - delete[] computationAmount_; - delete[] communicationAmount_; + delete[] computation_amount_; + delete[] communication_amount_; } } -void L07Action::updateBound() +double L07Action::calculate_network_bound() const { double lat_current = 0.0; + double lat_bound = std::numeric_limits::max(); + + size_t host_count = host_list_.size(); - size_t host_count = hostList_.size(); + if (communication_amount_ == nullptr) { + return lat_bound; + } - if (communicationAmount_ != nullptr) { - for (size_t i = 0; i < host_count; i++) { - for (size_t j = 0; j < host_count; j++) { - if (communicationAmount_[i * host_count + j] > 0) { - double lat = 0.0; - std::vector route; - hostList_.at(i)->route_to(hostList_.at(j), route, &lat); + for (size_t i = 0; i < host_count; i++) { + for (size_t j = 0; j < host_count; j++) { + if (communication_amount_[i * host_count + j] > 0) { + double lat = 0.0; + std::vector route; + host_list_.at(i)->route_to(host_list_.at(j), route, &lat); - lat_current = std::max(lat_current, lat * communicationAmount_[i * host_count + j]); - } + lat_current = std::max(lat_current, lat * communication_amount_[i * host_count + j]); } } } - 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) && is_running()) { - if (rate_ < 0) - get_model()->get_maxmin_system()->update_variable_bound(get_variable(), lat_bound); - else - get_model()->get_maxmin_system()->update_variable_bound(get_variable(), std::min(rate_, lat_bound)); + if (lat_current > 0) { + lat_bound = NetworkModel::cfg_tcp_gamma / (2.0 * lat_current); } + return lat_bound; } +double L07Action::calculate_cpu_bound() const +{ + double cpu_bound = std::numeric_limits::max(); + + if (computation_amount_ == nullptr) { + return cpu_bound; + } + + for (size_t i = 0; i < host_list_.size(); i++) { + if (computation_amount_[i] > 0) { + cpu_bound = std::min(cpu_bound, host_list_[i]->get_cpu()->get_speed(1.0) * + host_list_[i]->get_cpu()->get_speed_ratio() / computation_amount_[i]); + } + } + return cpu_bound; } + +void L07Action::update_bound() const +{ + double bound = std::min(calculate_network_bound(), calculate_cpu_bound()); + + XBT_DEBUG("action (%p) : bound = %g", this, bound); + + /* latency has been paid (or no latency), we can set the appropriate bound for multicore or network limit */ + if ((bound < std::numeric_limits::max()) && (latency_ <= 0.0)) { + if (rate_ < 0) + get_model()->get_maxmin_system()->update_variable_bound(get_variable(), bound); + else + get_model()->get_maxmin_system()->update_variable_bound(get_variable(), std::min(rate_, bound)); + } } + +} // namespace simgrid::kernel::resource