#include "simgrid/host.h"
#include "simgrid/simix.hpp"
+#include "xbt/string.hpp"
extern "C" {
struct s_dirty_page {
- double prev_clock;
- double prev_remaining;
- msg_task_t task;
+ double prev_clock = 0.0;
+ double prev_remaining = 0.0;
+ msg_task_t task = nullptr;
};
typedef s_dirty_page* dirty_page_t;
XBT_LOG_NEW_DEFAULT_SUBCATEGORY(msg_vm, msg, "Cloud-oriented parts of the MSG API");
-/* **** ******** GENERAL ********* **** */
const char* MSG_vm_get_name(msg_vm_t vm)
{
return vm->getCname();
vm->getParameters(params);
}
+void MSG_vm_set_ramsize(msg_vm_t vm, size_t size)
+{
+ vm->setRamsize(size);
+}
+size_t MSG_vm_get_ramsize(msg_vm_t vm)
+{
+ return vm->getRamsize();
+}
+
/* **** Check state of a VM **** */
static inline int __MSG_vm_is_state(msg_vm_t vm, e_surf_vm_state_t state)
{
- return vm->pimpl_vm_ != nullptr && vm->pimpl_vm_->getState() == state;
+ return vm->pimpl_vm_ != nullptr && vm->getState() == state;
}
/** @brief Returns whether the given VM has just created, not running.
/* For the moment, intensity_rate is the percentage against the migration bandwidth */
- msg_vm_t vm = new simgrid::s4u::VirtualMachine(name, pm, coreAmount);
+ msg_vm_t vm = new simgrid::s4u::VirtualMachine(name, pm, coreAmount, static_cast<sg_size_t>(ramsize) * 1024 * 1024);
s_vm_params_t params;
- memset(¶ms, 0, sizeof(params));
- params.ramsize = static_cast<sg_size_t>(ramsize) * 1024 * 1024;
- params.devsize = 0;
- params.skip_stage2 = 0;
params.max_downtime = 0.03;
- params.mig_speed = static_cast<double>(mig_netspeed) * 1024 * 1024; // mig_speed
+ params.mig_speed = static_cast<double>(mig_netspeed) * 1024 * 1024; // mig_speed
params.dp_intensity = static_cast<double>(dp_intensity) / 100;
- params.dp_cap = params.ramsize * 0.9; // assume working set memory is 90% of ramsize
+ params.dp_cap = vm->getRamsize() * 0.9; // assume working set memory is 90% of ramsize
XBT_DEBUG("migspeed : %f intensity mem : %d", params.mig_speed, dp_intensity);
vm->setParameters(¶ms);
xbt_assert(sg_host_by_name(name) == nullptr,
"Cannot create a VM named %s: this name is already used by an host or a VM", name);
- return new simgrid::s4u::VirtualMachine(name, pm, 1);
+ msg_vm_t vm = new simgrid::s4u::VirtualMachine(name, pm, 1);
+ s_vm_params_t params;
+ memset(¶ms, 0, sizeof(params));
+ vm->setParameters(¶ms);
+ return vm;
}
/** @brief Create a new VM object with the default parameters, but with a specified amount of cores
* @ingroup msg_VMs*
xbt_assert(sg_host_by_name(name) == nullptr,
"Cannot create a VM named %s: this name is already used by an host or a VM", name);
- return new simgrid::s4u::VirtualMachine(name, pm, coreAmount);
+ msg_vm_t vm = new simgrid::s4u::VirtualMachine(name, pm, coreAmount);
+ s_vm_params_t params;
+ memset(¶ms, 0, sizeof(params));
+ vm->setParameters(¶ms);
+ return vm;
}
/** @brief Destroy a VM. Destroy the VM object from the simulation.
{
vm->start();
if (TRACE_msg_vm_is_enabled()) {
- container_t vm_container = simgrid::instr::Container::byName(vm->getName());
- simgrid::instr::Type* type = vm_container->type_->byName("MSG_VM_STATE");
- simgrid::instr::Value* val = simgrid::instr::Value::byNameOrCreate("start", "0 0 1", type); // start is blue
- new simgrid::instr::PushStateEvent(MSG_get_clock(), vm_container, type, val);
+ simgrid::instr::StateType* state = simgrid::instr::Container::byName(vm->getName())->getState("MSG_VM_STATE");
+ state->addEntityValue("start", "0 0 1"); // start is blue
+ state->pushEvent("start");
}
}
/* The miration_rx process uses mbox_ctl to let the caller of do_migration()
* know the completion of the migration. */
- char *mbox_ctl;
+ std::string mbox_ctl;
/* The migration_rx and migration_tx processes use mbox to transfer migration data. */
- char *mbox;
+ std::string mbox;
};
static int migration_rx_fun(int argc, char *argv[])
std::string finalize_task_name = get_mig_task_name(ms->vm, ms->src_pm, ms->dst_pm, 3);
while (not received_finalize) {
msg_task_t task = nullptr;
- int ret = MSG_task_recv(&task, ms->mbox);
+ int ret = MSG_task_recv(&task, ms->mbox.c_str());
if (ret != MSG_OK) {
// An error occurred, clean the code and return
simgrid::simix::kernelImmediate([vm, dst_pm]() {
/* Update the vm location */
/* precopy migration makes the VM temporally paused */
- xbt_assert(vm->pimpl_vm_->getState() == SURF_VM_STATE_SUSPENDED);
+ xbt_assert(vm->getState() == SURF_VM_STATE_SUSPENDED);
/* Update the vm location and resume it */
vm->pimpl_vm_->setPm(dst_pm);
- vm->pimpl_vm_->resume();
+ vm->resume();
});
if (TRACE_msg_vm_is_enabled()) {
static long long int counter = 0;
- char key[INSTR_DEFAULT_STR_SIZE];
- snprintf(key, INSTR_DEFAULT_STR_SIZE, "%lld", counter);
+ std::string key = std::to_string(counter);
counter++;
// start link
- container_t msg = simgrid::instr::Container::byName(vm->getName());
- simgrid::instr::Type* type = simgrid::instr::Type::getRootType()->byName("MSG_VM_LINK");
- new simgrid::instr::StartLinkEvent(MSG_get_clock(), PJ_container_get_root(), type, msg, "M", key);
+ container_t msg = simgrid::instr::Container::byName(vm->getName());
+ simgrid::instr::Container::getRoot()->getLink("MSG_VM_LINK")->startEvent(msg, "M", key);
// destroy existing container of this vm
container_t existing_container = simgrid::instr::Container::byName(vm->getName());
delete existing_container;
// create new container on the new_host location
- new simgrid::instr::Container(vm->getCname(), simgrid::instr::INSTR_MSG_VM,
- simgrid::instr::Container::byName(ms->dst_pm->getName()));
+ new simgrid::instr::Container(vm->getCname(), "MSG_VM", simgrid::instr::Container::byName(ms->dst_pm->getName()));
// end link
msg = simgrid::instr::Container::byName(vm->getName());
- type = simgrid::instr::Type::getRootType()->byName("MSG_VM_LINK");
- new simgrid::instr::EndLinkEvent(MSG_get_clock(), PJ_container_get_root(), type, msg, "M", key);
+ simgrid::instr::Container::getRoot()->getLink("MSG_VM_LINK")->endEvent(msg, "M", key);
}
// Inform the SRC that the migration has been correctly performed
std::string task_name = get_mig_task_name(ms->vm, ms->src_pm, ms->dst_pm, 4);
msg_task_t task = MSG_task_create(task_name.c_str(), 0, 0, nullptr);
- msg_error_t ret = MSG_task_send(task, ms->mbox_ctl);
+ msg_error_t ret = MSG_task_send(task, ms->mbox_ctl.c_str());
if(ret == MSG_HOST_FAILURE){
// The DST has crashed, this is a problem has the VM since we are not sure whether SRC is considering that the VM
// has been correctly migrated on the DST node
static void start_dirty_page_tracking(msg_vm_t vm)
{
- vm->pimpl_vm_->dp_enabled = 1;
+ vm->pimpl_vm_->dp_enabled = true;
if (vm->pimpl_vm_->dp_objs.empty())
return;
for (auto const& elm : vm->pimpl_vm_->dp_objs) {
dirty_page_t dp = elm.second;
- double remaining = MSG_task_get_flops_amount(dp->task);
+ double remaining = MSG_task_get_flops_amount(dp->task);
dp->prev_clock = MSG_get_clock();
dp->prev_remaining = remaining;
XBT_DEBUG("%s@%s remaining %f", elm.first.c_str(), vm->getCname(), remaining);
static void stop_dirty_page_tracking(msg_vm_t vm)
{
- vm->pimpl_vm_->dp_enabled = 0;
+ vm->pimpl_vm_->dp_enabled = false;
}
-static double get_computed(const char* key, msg_vm_t vm, dirty_page_t dp, double remaining, double clock)
+static double get_computed(const std::string& key, msg_vm_t vm, dirty_page_t dp, double remaining, double clock)
{
double computed = dp->prev_remaining - remaining;
double duration = clock - dp->prev_clock;
- XBT_DEBUG("%s@%s: computed %f ops (remaining %f -> %f) in %f secs (%f -> %f)", key, vm->getCname(), computed,
+ XBT_DEBUG("%s@%s: computed %f ops (remaining %f -> %f) in %f secs (%f -> %f)", key.c_str(), vm->getCname(), computed,
dp->prev_remaining, remaining, duration, dp->prev_clock, clock);
return computed;
double total = 0;
for (auto const& elm : vm->pimpl_vm_->dp_objs) {
- const char* key = elm.first.c_str();
+ const std::string& key = elm.first;
dirty_page_t dp = elm.second;
- double remaining = MSG_task_get_flops_amount(dp->task);
+ double remaining = MSG_task_get_flops_amount(dp->task);
double clock = MSG_get_clock();
return;
double remaining = MSG_task_get_flops_amount(task);
- char *key = bprintf("%s-%p", task->name, task);
+ std::string key = simgrid::xbt::string_printf("%s-%p", task->name, task);
- dirty_page_t dp = xbt_new0(s_dirty_page, 1);
+ dirty_page_t dp = new s_dirty_page;
dp->task = task;
if (vm->pimpl_vm_->dp_enabled) {
dp->prev_clock = MSG_get_clock();
dp->prev_remaining = remaining;
}
vm->pimpl_vm_->dp_objs.insert({key, dp});
- XBT_DEBUG("add %s on %s (remaining %f, dp_enabled %d)", key, host->getCname(), remaining, vm->pimpl_vm_->dp_enabled);
-
- xbt_free(key);
+ XBT_DEBUG("add %s on %s (remaining %f, dp_enabled %d)", key.c_str(), host->getCname(), remaining,
+ vm->pimpl_vm_->dp_enabled);
}
void MSG_host_del_task(msg_host_t host, msg_task_t task)
if (vm == nullptr)
return;
- char *key = bprintf("%s-%p", task->name, task);
+ std::string key = simgrid::xbt::string_printf("%s-%p", task->name, task);
dirty_page_t dp = nullptr;
- if (vm->pimpl_vm_->dp_objs.find(key) != vm->pimpl_vm_->dp_objs.end())
- dp = vm->pimpl_vm_->dp_objs.at(key);
+ auto dp_obj = vm->pimpl_vm_->dp_objs.find(key);
+ if (dp_obj != vm->pimpl_vm_->dp_objs.end())
+ dp = dp_obj->second;
xbt_assert(dp && dp->task == task);
/* If we are in the middle of dirty page tracking, we record how much computation has been done until now, and keep
}
vm->pimpl_vm_->dp_objs.erase(key);
- xbt_free(dp);
+ delete dp;
- XBT_DEBUG("del %s on %s", key, host->getCname());
- xbt_free(key);
+ XBT_DEBUG("del %s on %s", key.c_str(), host->getCname());
}
-static sg_size_t send_migration_data(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm, sg_size_t size, char* mbox,
- int stage, int stage2_round, double mig_speed, double timeout)
+static sg_size_t send_migration_data(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm, sg_size_t size,
+ const std::string& mbox, int stage, int stage2_round, double mig_speed,
+ double timeout)
{
sg_size_t sent = 0;
std::string task_name = get_mig_task_name(vm, src_pm, dst_pm, stage);
msg_error_t ret;
if (mig_speed > 0)
- ret = MSG_task_send_with_timeout_bounded(task, mbox, timeout, mig_speed);
+ ret = MSG_task_send_with_timeout_bounded(task, mbox.c_str(), timeout, mig_speed);
else
- ret = MSG_task_send(task, mbox);
+ ret = MSG_task_send(task, mbox.c_str());
if (ret == MSG_OK) {
sent = size;
// Note that the ms structure has been allocated in do_migration and hence should be freed in the same function ;)
migration_session* ms = static_cast<migration_session*>(MSG_process_get_data(MSG_process_self()));
- double host_speed = ms->vm->pimpl_vm_->getPm()->getSpeed();
+ double host_speed = ms->vm->getPm()->getSpeed();
s_vm_params_t params;
ms->vm->getParameters(¶ms);
- const sg_size_t ramsize = params.ramsize;
- const sg_size_t devsize = params.devsize;
- const int skip_stage1 = params.skip_stage1;
- int skip_stage2 = params.skip_stage2;
+ const sg_size_t ramsize = ms->vm->getRamsize();
const double dp_rate = host_speed ? (params.mig_speed * params.dp_intensity) / host_speed : 1;
const double dp_cap = params.dp_cap;
const double mig_speed = params.mig_speed;
double max_downtime = params.max_downtime;
double mig_timeout = 10000000.0;
+ bool skip_stage2 = false;
- double remaining_size = static_cast<double>(ramsize + devsize);
- double threshold = 0.0;
+ size_t remaining_size = ramsize;
+ size_t threshold = 0.0;
/* check parameters */
if (ramsize == 0)
}
/* Stage1: send all memory pages to the destination. */
- XBT_DEBUG("mig-stage1: remaining_size %f", remaining_size);
+ XBT_DEBUG("mig-stage1: remaining_size %zu", remaining_size);
start_dirty_page_tracking(ms->vm);
double computed_during_stage1 = 0;
- if (not skip_stage1) {
- double clock_prev_send = MSG_get_clock();
-
- try {
- /* At stage 1, we do not need timeout. We have to send all the memory pages even though the duration of this
- * transfer exceeds the timeout value. */
- XBT_VERB("Stage 1: Gonna send %llu bytes", ramsize);
- sg_size_t sent = send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, ramsize, ms->mbox, 1, 0, mig_speed, -1);
- remaining_size -= sent;
- computed_during_stage1 = lookup_computed_flop_counts(ms->vm, 1, 0);
-
- if (sent < ramsize) {
- XBT_VERB("mig-stage1: timeout, force moving to stage 3");
- skip_stage2 = 1;
- } else if (sent > ramsize)
- XBT_CRITICAL("bug");
-
- }
- catch (xbt_ex& e) {
- //hostfailure (if you want to know whether this is the SRC or the DST check directly in send_migration_data code)
- // Stop the dirty page tracking an return (there is no memory space to release)
- stop_dirty_page_tracking(ms->vm);
- return 0;
- }
+ double clock_prev_send = MSG_get_clock();
- double clock_post_send = MSG_get_clock();
- mig_timeout -= (clock_post_send - clock_prev_send);
- if (mig_timeout < 0) {
- XBT_VERB("The duration of stage 1 exceeds the timeout value, skip stage 2");
- skip_stage2 = 1;
- }
+ try {
+ /* At stage 1, we do not need timeout. We have to send all the memory pages even though the duration of this
+ * transfer exceeds the timeout value. */
+ XBT_VERB("Stage 1: Gonna send %llu bytes", ramsize);
+ sg_size_t sent = send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, ramsize, ms->mbox, 1, 0, mig_speed, -1);
+ remaining_size -= sent;
+ computed_during_stage1 = lookup_computed_flop_counts(ms->vm, 1, 0);
+
+ if (sent < ramsize) {
+ XBT_VERB("mig-stage1: timeout, force moving to stage 3");
+ skip_stage2 = true;
+ } else if (sent > ramsize)
+ XBT_CRITICAL("bug");
+
+ } catch (xbt_ex& e) {
+ // hostfailure (if you want to know whether this is the SRC or the DST check directly in send_migration_data code)
+ // Stop the dirty page tracking an return (there is no memory space to release)
+ stop_dirty_page_tracking(ms->vm);
+ return 0;
+ }
- /* estimate bandwidth */
- double bandwidth = ramsize / (clock_post_send - clock_prev_send);
- threshold = bandwidth * max_downtime;
- XBT_DEBUG("actual bandwidth %f (MB/s), threshold %f", bandwidth / 1024 / 1024, threshold);
+ double clock_post_send = MSG_get_clock();
+ mig_timeout -= (clock_post_send - clock_prev_send);
+ if (mig_timeout < 0) {
+ XBT_VERB("The duration of stage 1 exceeds the timeout value, skip stage 2");
+ skip_stage2 = true;
}
+ /* estimate bandwidth */
+ double bandwidth = ramsize / (clock_post_send - clock_prev_send);
+ threshold = bandwidth * max_downtime;
+ XBT_DEBUG("actual bandwidth %f (MB/s), threshold %zu", bandwidth / 1024 / 1024, threshold);
/* Stage2: send update pages iteratively until the size of remaining states becomes smaller than threshold value. */
if (not skip_stage2) {
/* Check whether the remaining size is below the threshold value. If so, move to stage 3. */
remaining_size += updated_size;
- XBT_DEBUG("mig-stage2.%d: remaining_size %f (%s threshold %f)", stage2_round, remaining_size,
+ XBT_DEBUG("mig-stage2.%d: remaining_size %zu (%s threshold %zu)", stage2_round, remaining_size,
(remaining_size < threshold) ? "<" : ">", threshold);
if (remaining_size < threshold)
break;
/* timeout did not happen */
double bandwidth = updated_size / (clock_post_send - clock_prev_send);
threshold = bandwidth * max_downtime;
- XBT_DEBUG("actual bandwidth %f, threshold %f", bandwidth / 1024 / 1024, threshold);
+ XBT_DEBUG("actual bandwidth %f, threshold %zu", bandwidth / 1024 / 1024, threshold);
remaining_size -= sent;
stage2_round += 1;
mig_timeout -= (clock_post_send - clock_prev_send);
}
/* Stage3: stop the VM and copy the rest of states. */
- XBT_DEBUG("mig-stage3: remaining_size %f", remaining_size);
+ XBT_DEBUG("mig-stage3: remaining_size %zu", remaining_size);
simgrid::vm::VirtualMachineImpl* pimpl = ms->vm->pimpl_vm_;
pimpl->setState(SURF_VM_STATE_RUNNING); // FIXME: this bypass of the checks in suspend() is not nice
pimpl->isMigrating = false; // FIXME: this bypass of the checks in suspend() is not nice
stop_dirty_page_tracking(ms->vm);
try {
- XBT_DEBUG("Stage 3: Gonna send %f bytes", remaining_size);
- send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, static_cast<sg_size_t>(remaining_size), ms->mbox, 3, 0,
- mig_speed, -1);
+ XBT_DEBUG("Stage 3: Gonna send %zu bytes", remaining_size);
+ send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, remaining_size, ms->mbox, 3, 0, mig_speed, -1);
}
catch(xbt_ex& e) {
//hostfailure (if you want to know whether this is the SRC or the DST check directly in send_migration_data code)
// Stop the dirty page tracking an return (there is no memory space to release)
- ms->vm->pimpl_vm_->resume();
+ ms->vm->resume();
return 0;
}
* The second one would be easier.
*/
- msg_host_t src_pm = vm->pimpl_vm_->getPm();
+ msg_host_t src_pm = vm->getPm();
if (src_pm->isOff())
THROWF(vm_error, 0, "Cannot migrate VM '%s' from host '%s', which is offline.", vm->getCname(), src_pm->getCname());
vm->pimpl_vm_->isMigrating = true;
- migration_session* ms = xbt_new(migration_session, 1);
- ms->vm = vm;
- ms->src_pm = src_pm;
- ms->dst_pm = dst_pm;
+ migration_session ms;
+ ms.vm = vm;
+ ms.src_pm = src_pm;
+ ms.dst_pm = dst_pm;
/* We have two mailboxes. mbox is used to transfer migration data between source and destination PMs. mbox_ctl is used
* to detect the completion of a migration. The names of these mailboxes must not conflict with others. */
- ms->mbox_ctl = bprintf("__mbox_mig_ctl:%s(%s-%s)", vm->getCname(), src_pm->getCname(), dst_pm->getCname());
- ms->mbox = bprintf("__mbox_mig_src_dst:%s(%s-%s)", vm->getCname(), src_pm->getCname(), dst_pm->getCname());
+ ms.mbox_ctl =
+ simgrid::xbt::string_printf("__mbox_mig_ctl:%s(%s-%s)", vm->getCname(), src_pm->getCname(), dst_pm->getCname());
+ ms.mbox = simgrid::xbt::string_printf("__mbox_mig_src_dst:%s(%s-%s)", vm->getCname(), src_pm->getCname(),
+ dst_pm->getCname());
std::string pr_rx_name = get_mig_process_rx_name(vm, src_pm, dst_pm);
std::string pr_tx_name = get_mig_process_tx_name(vm, src_pm, dst_pm);
- MSG_process_create(pr_rx_name.c_str(), migration_rx_fun, ms, dst_pm);
+ MSG_process_create(pr_rx_name.c_str(), migration_rx_fun, &ms, dst_pm);
- MSG_process_create(pr_tx_name.c_str(), migration_tx_fun, ms, src_pm);
+ MSG_process_create(pr_tx_name.c_str(), migration_tx_fun, &ms, src_pm);
/* wait until the migration have finished or on error has occurred */
XBT_DEBUG("wait for reception of the final ACK (i.e. migration has been correctly performed");
msg_task_t task = nullptr;
- msg_error_t ret = MSG_task_receive(&task, ms->mbox_ctl);
+ msg_error_t ret = MSG_task_receive(&task, ms.mbox_ctl.c_str());
vm->pimpl_vm_->isMigrating = false;
- xbt_free(ms->mbox_ctl);
- xbt_free(ms->mbox);
- xbt_free(ms);
-
if (ret == MSG_HOST_FAILURE) {
// Note that since the communication failed, the owner did not change and the task should be destroyed on the
// other side. Hence, just throw the execption
*/
void MSG_vm_suspend(msg_vm_t vm)
{
- smx_actor_t issuer = SIMIX_process_self();
- simgrid::simix::kernelImmediate([vm, issuer]() { vm->pimpl_vm_->suspend(issuer); });
-
- XBT_DEBUG("vm_suspend done");
-
+ vm->suspend();
if (TRACE_msg_vm_is_enabled()) {
- container_t vm_container = simgrid::instr::Container::byName(vm->getName());
- simgrid::instr::Type* type = vm_container->type_->byName("MSG_VM_STATE");
- simgrid::instr::Value* val = simgrid::instr::Value::byNameOrCreate("suspend", "1 0 0", type); // suspend is red
- new simgrid::instr::PushStateEvent(MSG_get_clock(), vm_container, type, val);
+ simgrid::instr::StateType* state = simgrid::instr::Container::byName(vm->getName())->getState("MSG_VM_STATE");
+ state->addEntityValue("suspend", "1 0 0"); // suspend is red
+ state->pushEvent("suspend");
}
}
*/
void MSG_vm_resume(msg_vm_t vm)
{
- vm->pimpl_vm_->resume();
-
- if (TRACE_msg_vm_is_enabled()) {
- container_t vm_container = simgrid::instr::Container::byName(vm->getName());
- simgrid::instr::Type* type = vm_container->type_->byName("MSG_VM_STATE");
- new simgrid::instr::PopStateEvent(MSG_get_clock(), vm_container, type);
- }
+ vm->resume();
+ if (TRACE_msg_vm_is_enabled())
+ simgrid::instr::Container::byName(vm->getName())->getState("MSG_VM_STATE")->popEvent();
}
/** @brief Get the physical host of a given VM.