- SimDAG++: Automatic dependencies on S4U activities (experimental)
- Some features are already implemented but not all of them
- Cannot block an activity until it's scheduled on a resource
- - Only Exec and Io so far. Comms are not handled yet.
- - No heterogeneous wait_any() that would mix Exec and Io activities.
- - See examples/s4u/{io-,exec-}dependent for what's already there.
+ - No heterogeneous wait_any() that would mix Exec/Comm/Io activities.
+ - See examples/s4u/{io,exec,comm}-dependent for what's already there.
include examples/s4u/cloud-migration/s4u-cloud-migration.tesh
include examples/s4u/cloud-simple/s4u-cloud-simple.cpp
include examples/s4u/cloud-simple/s4u-cloud-simple.tesh
+include examples/s4u/comm-dependent/s4u-comm-dependent.cpp
+include examples/s4u/comm-dependent/s4u-comm-dependent.tesh
include examples/s4u/dht-chord/s4u-dht-chord-node.cpp
include examples/s4u/dht-chord/s4u-dht-chord.cpp
include examples/s4u/dht-chord/s4u-dht-chord.hpp
include examples/s4u/exec-async/s4u-exec-async.tesh
include examples/s4u/exec-basic/s4u-exec-basic.cpp
include examples/s4u/exec-basic/s4u-exec-basic.tesh
+include examples/s4u/exec-dependent/s4u-exec-dependent.cpp
+include examples/s4u/exec-dependent/s4u-exec-dependent.tesh
include examples/s4u/exec-dvfs/s4u-exec-dvfs.cpp
include examples/s4u/exec-dvfs/s4u-exec-dvfs.tesh
include examples/s4u/exec-ptask/s4u-exec-ptask.cpp
include examples/s4u/exec-waitfor/s4u-exec-waitfor.tesh
include examples/s4u/io-async/s4u-io-async.cpp
include examples/s4u/io-async/s4u-io-async.tesh
+include examples/s4u/io-dependent/s4u-io-dependent.cpp
+include examples/s4u/io-dependent/s4u-io-dependent.tesh
include examples/s4u/io-disk-raw/s4u-io-disk-raw.cpp
include examples/s4u/io-disk-raw/s4u-io-disk-raw.tesh
include examples/s4u/io-file-remote/s4u-io-file-remote.cpp
actor-lifetime actor-migrate actor-suspend actor-yield
app-chainsend app-pingpong app-token-ring
async-ready async-wait async-waitany async-waitall async-waituntil
+ comm-dependent
cloud-capping cloud-migration cloud-simple
energy-exec energy-boot energy-link energy-vm energy-exec-ptask
engine-filtering
- exec-async exec-basic exec-dvfs exec-ptask exec-remote exec-waitany exec-waitfor
- io-async io-file-system io-file-remote io-disk-raw
+ exec-async exec-basic exec-dvfs exec-ptask exec-remote exec-waitany exec-waitfor exec-dependent
+ io-async io-file-system io-file-remote io-disk-raw io-dependent
platform-failures platform-profile platform-properties
plugin-hostload
replay-comm replay-io
> [ 0.110000] (3:peer@Perl) I got a 'Message 5 from peer 1'.
> [ 0.110000] (3:peer@Perl) I got a 'finalize'.
> [ 0.110000] (3:peer@Perl) I'm done, just waiting for my peers to receive the messages before exiting
-> [ 0.110000] (3:peer@Perl) Goodbye now!
> [ 0.110000] (1:peer@Tremblay) Goodbye now!
> [ 0.110000] (2:peer@Ruby) Goodbye now!
+> [ 0.110000] (3:peer@Perl) Goodbye now!
--- /dev/null
+/* Copyright (c) 2007-2020. 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 <simgrid/s4u.hpp>
+
+XBT_LOG_NEW_DEFAULT_CATEGORY(s4u_comm_dependent, "Messages specific for this s4u example");
+
+static void sender(simgrid::s4u::Mailbox* mailbox)
+{
+ double* computation_amount = new double();
+ *computation_amount = simgrid::s4u::this_actor::get_host()->get_speed();
+ simgrid::s4u::ExecPtr exec = simgrid::s4u::this_actor::exec_init(2 * (*computation_amount));
+ simgrid::s4u::CommPtr comm = mailbox->put_init(computation_amount, 7e6);
+
+ exec->set_name("exec on sender")->add_successor(comm)->start();
+ comm->set_name("comm to receiver")->vetoable_start();
+ exec->wait();
+ comm->wait();
+}
+
+static void receiver(simgrid::s4u::Mailbox* mailbox)
+{
+ double* received = nullptr;
+ double computation_amount = simgrid::s4u::this_actor::get_host()->get_speed();
+ simgrid::s4u::ExecPtr exec = simgrid::s4u::this_actor::exec_init(2 * computation_amount);
+ simgrid::s4u::CommPtr comm = mailbox->get_init()->set_dst_data((void**)&received, sizeof(double));
+
+ comm->set_name("comm from sender")->add_successor(exec)->start();
+ exec->set_name("exec on receiver")->vetoable_start();
+
+ comm->wait();
+ exec->wait();
+ XBT_INFO("Received: %.0f flops were computed on sender", *received);
+ delete received;
+}
+
+int main(int argc, char* argv[])
+{
+ simgrid::s4u::Engine e(&argc, argv);
+ e.load_platform(argv[1]);
+
+ simgrid::s4u::Mailbox* mbox = simgrid::s4u::Mailbox::by_name("Mailbox");
+
+ simgrid::s4u::Actor::create("sender", simgrid::s4u::Host::by_name("Tremblay"), sender, mbox);
+ simgrid::s4u::Actor::create("receiver", simgrid::s4u::Host::by_name("Jupiter"), receiver, mbox);
+
+ e.run();
+
+ XBT_INFO("Simulation time: %.3f", e.get_clock());
+
+ return 0;
+}
--- /dev/null
+#!/usr/bin/env tesh
+
+p Testing with default compound
+
+$ ${bindir:=.}/s4u-comm-dependent ${platfdir}/small_platform.xml --log=s4u_activity.t:debug "--log=root.fmt:[%6.2r]%e(%i:%P@%h)%e%m%n"
+> [ 2.00] (1:sender@Tremblay) Remove a dependency from 'exec on sender' on 'comm to receiver'
+> [ 2.00] (1:sender@Tremblay) All dependencies are solved, let's start 'comm to receiver'
+> [ 3.07] (2:receiver@Jupiter) Remove a dependency from 'comm from sender' on 'exec on receiver'
+> [ 3.07] (2:receiver@Jupiter) All dependencies are solved, let's start 'exec on receiver'
+> [ 5.07] (2:receiver@Jupiter) Received: 98095000 flops were computed on sender
+> [ 5.07] (0:maestro@) Simulation time: 5.070
--- /dev/null
+/* 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 "simgrid/s4u.hpp"
+#include <vector>
+
+XBT_LOG_NEW_DEFAULT_CATEGORY(s4u_test, "Messages specific for this s4u example");
+
+static void worker()
+{
+ // Define an amount of work that should take 1 second to execute.
+ double computation_amount = simgrid::s4u::this_actor::get_host()->get_speed();
+
+ std::vector<simgrid::s4u::ExecPtr> pending_execs;
+ // Create a small DAG
+ // + Two parents and a child
+ // + First parent ends after 1 second and the Second parent after 2 seconds.
+ simgrid::s4u::ExecPtr first_parent = simgrid::s4u::this_actor::exec_init(computation_amount);
+ pending_execs.push_back(first_parent);
+ simgrid::s4u::ExecPtr second_parent = simgrid::s4u::this_actor::exec_init(2 * computation_amount);
+ pending_execs.push_back(second_parent);
+ simgrid::s4u::ExecPtr child = simgrid::s4u::this_actor::exec_init(computation_amount);
+ pending_execs.push_back(child);
+
+ // Name the activities (for logging purposes only)
+ first_parent->set_name("parent 1");
+ second_parent->set_name("parent 2");
+ child->set_name("child");
+
+ // Create the dependencies by declaring 'child' as a successor of first_parent and second_parent
+ first_parent->add_successor(child);
+ second_parent->add_successor(child);
+
+ // Start the activities.
+ first_parent->start();
+ second_parent->start();
+ // child uses a vetoable start to force it to wait for the completion of its predecessors
+ child->vetoable_start();
+
+ // wait for the completion of all activities
+ while (not pending_execs.empty()) {
+ int changed_pos = simgrid::s4u::Exec::wait_any_for(&pending_execs, -1);
+ XBT_INFO("Exec '%s' is complete", pending_execs[changed_pos]->get_cname());
+ pending_execs.erase(pending_execs.begin() + changed_pos);
+ }
+}
+
+int main(int argc, char* argv[])
+{
+ simgrid::s4u::Engine e(&argc, argv);
+ e.load_platform(argv[1]);
+
+ simgrid::s4u::Actor::create("worker", simgrid::s4u::Host::by_name("Fafard"), worker);
+
+ e.run();
+
+ XBT_INFO("Simulation time %g", e.get_clock());
+
+ return 0;
+}
--- /dev/null
+#!/usr/bin/env tesh
+
+! output sort
+$ ${bindir:=.}/s4u-exec-dependent ${platfdir}/small_platform.xml --log=s4u_activity.t:debug "--log=root.fmt:[%10.6r]%e(%i:%P@%h)%e%m%n"
+> [ 2.000000] (1:worker@Fafard) Remove a dependency from 'parent 1' on 'child'
+> [ 2.000000] (1:worker@Fafard) Exec 'parent 1' is complete
+> [ 3.000000] (1:worker@Fafard) Remove a dependency from 'parent 2' on 'child'
+> [ 3.000000] (1:worker@Fafard) All dependencies are solved, let's start 'child'
+> [ 3.000000] (1:worker@Fafard) Exec 'parent 2' is complete
+> [ 4.000000] (1:worker@Fafard) Exec 'child' is complete
+> [ 4.000000] (0:maestro@) Simulation time 4
--- /dev/null
+/* 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 "simgrid/s4u.hpp"
+#include <simgrid/plugins/file_system.h>
+#include <vector>
+
+XBT_LOG_NEW_DEFAULT_CATEGORY(s4u_test, "Messages specific for this s4u example");
+
+static void test()
+{
+ simgrid::s4u::ExecPtr bob_compute = simgrid::s4u::this_actor::exec_init(1e9);
+ simgrid::s4u::IoPtr bob_write =
+ simgrid::s4u::Host::current()->get_disks().front()->io_init(4000000, simgrid::s4u::Io::OpType::WRITE);
+ simgrid::s4u::IoPtr carl_read =
+ simgrid::s4u::Host::by_name("carl")->get_disks().front()->io_init(4000000, simgrid::s4u::Io::OpType::READ);
+ simgrid::s4u::ExecPtr carl_compute = simgrid::s4u::Host::by_name("carl")->exec_async(1e9);
+
+ // Name the activities (for logging purposes only)
+ bob_compute->set_name("bob compute");
+ bob_write->set_name("bob write");
+ carl_read->set_name("carl read");
+ carl_compute->set_name("carl compute");
+
+ // Create the dependencies:
+ // 'bob_write' is a successor of 'bob_compute'
+ // 'carl_read' is a successor of 'bob_write'
+ // 'carl_compute' is a successor of 'carl_read'
+ bob_compute->add_successor(bob_write);
+ bob_write->add_successor(carl_read);
+ carl_read->add_successor(carl_compute);
+
+ // Start the activities.
+ bob_compute->start();
+ bob_write->vetoable_start();
+ carl_read->vetoable_start();
+ carl_compute->vetoable_start();
+
+ // Wait for their completion (should be replaced by a wait_any_for at some point)
+ bob_compute->wait();
+ bob_write->wait();
+ carl_read->wait();
+ carl_compute->wait();
+}
+
+int main(int argc, char* argv[])
+{
+ simgrid::s4u::Engine e(&argc, argv);
+ sg_storage_file_system_init();
+ e.load_platform(argv[1]);
+
+ simgrid::s4u::Actor::create("bob", simgrid::s4u::Host::by_name("bob"), test);
+
+ e.run();
+
+ XBT_INFO("Simulation time %g", e.get_clock());
+
+ return 0;
+}
--- /dev/null
+#!/usr/bin/env tesh
+
+! output sort
+$ ${bindir:=.}/s4u-io-dependent ${platfdir}/hosts_with_disks.xml --log=s4u_activity.t:debug "--log=root.fmt:[%10.6r]%e(%i:%P@%h)%e%m%n"
+> [ 1.000000] (1:bob@bob) All dependencies are solved, let's start 'bob write'
+> [ 1.000000] (1:bob@bob) Remove a dependency from 'bob compute' on 'bob write'
+> [ 1.100000] (1:bob@bob) All dependencies are solved, let's start 'carl read'
+> [ 1.100000] (1:bob@bob) Remove a dependency from 'bob write' on 'carl read'
+> [ 1.166667] (1:bob@bob) All dependencies are solved, let's start 'carl compute'
+> [ 1.166667] (1:bob@bob) Remove a dependency from 'carl read' on 'carl compute'
+> [ 2.166667] (0:maestro@) Simulation time 2.16667
#include "xbt/asserts.h"
#include <atomic>
+#include <set>
#include <simgrid/forward.h>
#include <string>
+#include <vector>
#include <xbt/signal.hpp>
+XBT_LOG_EXTERNAL_CATEGORY(s4u_activity);
+
namespace simgrid {
namespace s4u {
Activity() = default;
virtual ~Activity() = default;
+ void release_dependencies()
+ {
+ while (not successors_.empty()) {
+ ActivityPtr b = successors_.back();
+ XBT_CDEBUG(s4u_activity, "Remove a dependency from '%s' on '%s'", get_cname(), b->get_cname());
+ b->dependencies_.erase(this);
+ if (b->dependencies_.empty()) {
+ b->vetoable_start();
+ }
+ successors_.pop_back();
+ }
+ }
+
+ void add_successor(ActivityPtr a)
+ {
+ successors_.push_back(a);
+ a->dependencies_.insert({this});
+ }
+
public:
+ void vetoable_start()
+ {
+ state_ = State::STARTING;
+ if (dependencies_.empty()) {
+ XBT_CDEBUG(s4u_activity, "All dependencies are solved, let's start '%s'", get_cname());
+ start();
+ }
+ }
+
#ifndef DOXYGEN
Activity(Activity const&) = delete;
Activity& operator=(Activity const&) = delete;
#endif
- enum class State { INITED = 0, STARTED, CANCELED, ERRORED, FINISHED };
+ enum class State { INITED = 0, STARTING, STARTED, CANCELED, ERRORED, FINISHED };
/** Starts a previously created activity.
*
virtual Activity* cancel() = 0;
/** Retrieve the current state of the activity */
Activity::State get_state() { return state_; }
+ void set_state(Activity::State state) { state_ = state; }
/** Tests whether the given activity is terminated yet. This is a pure function. */
virtual bool test() = 0;
+ virtual const char* get_cname() = 0;
+ virtual const std::string& get_name() = 0;
/** Get the remaining amount of work that this Activity entails. When it's 0, it's done. */
virtual double get_remaining();
-
/** Set the [remaining] amount of work that this Activity will entail
*
* It is forbidden to change the amount of work once the Activity is started */
kernel::activity::ActivityImplPtr pimpl_ = nullptr;
Activity::State state_ = Activity::State::INITED;
double remains_ = 0;
+ std::vector<ActivityPtr> successors_;
+ std::set<ActivityPtr> dependencies_;
std::atomic_int_fast32_t refcount_{0};
};
template <class AnyActivity> class Activity_T : public Activity {
-private:
- std::string name_ = "";
+ std::string name_ = "unnamed";
std::string tracing_category_ = "";
void* user_data_ = nullptr;
public:
+ AnyActivity* add_successor(ActivityPtr a)
+ {
+ Activity::add_successor(a);
+ return static_cast<AnyActivity*>(this);
+ }
+
AnyActivity* set_name(const std::string& name)
{
xbt_assert(get_state() == State::INITED, "Cannot change the name of an activity after its start");
synchro->state_ = simgrid::kernel::activity::State::TIMEOUT;
}
synchro->finish();
+ return;
}
/* If the synchro is already finished then perform the error handling */
std::unique_ptr<kernel::activity::CommImpl* []> rcomms(new kernel::activity::CommImpl*[comms->size()]);
std::transform(begin(*comms), end(*comms), rcomms.get(),
[](const CommPtr& comm) { return static_cast<kernel::activity::CommImpl*>(comm->pimpl_.get()); });
- return simcall_comm_waitany(rcomms.get(), comms->size(), timeout);
+ int changed_pos = simcall_comm_waitany(rcomms.get(), comms->size(), timeout);
+ if (changed_pos != -1)
+ comms->at(changed_pos)->release_dependencies();
+ return changed_pos;
}
void Comm::wait_all(const std::vector<CommPtr>* comms)
src_buff_size_ = size;
return this;
}
+
CommPtr Comm::set_dst_data(void** buff)
{
xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
Comm* Comm::start()
{
- xbt_assert(get_state() == State::INITED, "You cannot use %s() once your communication started (not implemented)",
- __FUNCTION__);
+ xbt_assert(get_state() == State::INITED || get_state() == State::STARTING,
+ "You cannot use %s() once your communication started (not implemented)", __FUNCTION__);
if (src_buff_ != nullptr) { // Sender side
on_sender_start(*Actor::self());
case State::FINISHED:
break;
- case State::INITED: // It's not started yet. Do it in one simcall
+ case State::INITED:
+ case State::STARTING: // It's not started yet. Do it in one simcall
if (src_buff_ != nullptr) {
on_sender_start(*Actor::self());
simcall_comm_send(sender_, mailbox_->get_impl(), remains_, rate_, src_buff_, src_buff_size_, match_fun_,
get_user_data(), timeout, rate_);
}
state_ = State::FINISHED;
+ this->release_dependencies();
break;
case State::STARTED:
simcall_comm_wait(pimpl_, timeout);
on_completion(*Actor::self());
state_ = State::FINISHED;
+ this->release_dependencies();
break;
case State::CANCELED:
}
return this;
}
+
int Comm::test_any(const std::vector<CommPtr>* comms)
{
std::unique_ptr<kernel::activity::CommImpl* []> rcomms(new kernel::activity::CommImpl*[comms->size()]);
std::transform(begin(*comms), end(*comms), rcomms.get(),
[](const CommPtr& comm) { return static_cast<kernel::activity::CommImpl*>(comm->pimpl_.get()); });
- return simcall_comm_testany(rcomms.get(), comms->size());
+ int changed_pos = simcall_comm_testany(rcomms.get(), comms->size());
+ if (changed_pos != -1)
+ comms->at(changed_pos)->release_dependencies();
+ return changed_pos;
}
Comm* Comm::detach()
__FUNCTION__);
xbt_assert(src_buff_ != nullptr && src_buff_size_ != 0, "You can only detach sends, not recvs");
detached_ = true;
- return start();
+ vetoable_start();
+ return this;
}
Comm* Comm::cancel()
if (state_ == State::FINISHED)
return true;
- if (state_ == State::INITED)
- this->start();
+ if (state_ == State::INITED || state_ == State::STARTING)
+ this->vetoable_start();
if (simcall_comm_test(pimpl_)) {
state_ = State::FINISHED;
+ this->release_dependencies();
return true;
}
return false;
bool Exec::test()
{
- xbt_assert(state_ == State::INITED || state_ == State::STARTED || state_ == State::FINISHED);
+ xbt_assert(state_ == State::INITED || state_ == State::STARTED || state_ == State::STARTING ||
+ state_ == State::FINISHED);
if (state_ == State::FINISHED)
return true;
- if (state_ == State::INITED)
- this->start();
+ if (state_ == State::INITED || state_ == State::STARTING)
+ this->vetoable_start();
if (simcall_execution_test(pimpl_)) {
state_ = State::FINISHED;
+ this->release_dependencies();
return true;
}
Exec* Exec::wait_for(double timeout)
{
if (state_ == State::INITED)
- start();
+ vetoable_start();
simcall_execution_wait(pimpl_, timeout);
state_ = State::FINISHED;
on_completion(*Actor::self(), *this);
+ this->release_dependencies();
return this;
}
std::unique_ptr<kernel::activity::ExecImpl* []> rexecs(new kernel::activity::ExecImpl*[execs->size()]);
std::transform(begin(*execs), end(*execs), rexecs.get(),
[](const ExecPtr& exec) { return static_cast<kernel::activity::ExecImpl*>(exec->pimpl_.get()); });
- return simcall_execution_waitany_for(rexecs.get(), execs->size(), timeout);
+
+ int changed_pos = simcall_execution_waitany_for(rexecs.get(), execs->size(), timeout);
+ if (changed_pos != -1)
+ execs->at(changed_pos)->release_dependencies();
+ return changed_pos;
}
Exec* Exec::cancel()
Io* Io::wait_for(double timeout)
{
+ if (state_ == State::INITED)
+ vetoable_start();
simcall_io_wait(pimpl_, timeout);
state_ = State::FINISHED;
+ this->release_dependencies();
return this;
}
bool Io::test()
{
- xbt_assert(state_ == State::INITED || state_ == State::STARTED || state_ == State::FINISHED);
+ xbt_assert(state_ == State::INITED || state_ == State::STARTED || state_ == State::STARTING ||
+ state_ == State::FINISHED);
if (state_ == State::FINISHED)
return true;
- if (state_ == State::INITED)
- this->start();
+ if (state_ == State::INITED || state_ == State::STARTING)
+ this->vetoable_start();
if (simcall_io_test(pimpl_)) {
state_ = State::FINISHED;
+ this->release_dependencies();
return true;
}
xbt_assert(payload != nullptr, "You cannot send nullptr");
s4u::CommPtr res = put_init(payload, simulated_size_in_bytes);
- res->start();
+ res->vetoable_start();
return res;
}
void Mailbox::put(void* payload, uint64_t simulated_size_in_bytes)
CommPtr c = put_init();
c->set_remaining(simulated_size_in_bytes);
c->set_src_data(payload);
+ c->vetoable_start();
c->wait();
}
/** Blocking send with timeout */
CommPtr c = put_init();
c->set_remaining(simulated_size_in_bytes);
c->set_src_data(payload);
- // c->start() is optional.
+ c->vetoable_start();
c->wait_for(timeout);
}
{
s4u::CommPtr res = get_init();
res->set_dst_data(data, sizeof(*data));
- res->start();
+ res->vetoable_start();
return res;
}
void* res = nullptr;
CommPtr c = get_init();
c->set_dst_data(&res, sizeof(res));
+ c->vetoable_start();
c->wait();
return res;
}
void* res = nullptr;
CommPtr c = get_init();
c->set_dst_data(&res, sizeof(res));
+ c->vetoable_start();
c->wait_for(timeout);
return res;
}