});
if (TRACE_msg_vm_is_enabled()) {
- container_t container = PJ_container_get(vm->name().c_str());
+ container_t container = PJ_container_get(vm->cname());
PJ_container_remove_from_parent(container);
PJ_container_free(container);
}
if (pm_ramsize && !pm_overcommit) { /* Only verify that we don't overcommit on need */
/* Retrieve the memory occupied by the VMs on that host. Yep, we have to traverse all VMs of all hosts for that */
long total_ramsize_of_vms = 0;
- for (simgrid::s4u::VirtualMachine* ws_vm : simgrid::surf::VirtualMachineImpl::allVms_)
+ for (simgrid::s4u::VirtualMachine* ws_vm : simgrid::vm::VirtualMachineImpl::allVms_)
if (pm == ws_vm->pimpl_vm_->getPm())
total_ramsize_of_vms += ws_vm->pimpl_vm_->getRamsize();
if (vm_ramsize > pm_ramsize - total_ramsize_of_vms) {
XBT_WARN("cannnot start %s@%s due to memory shortage: vm_ramsize %ld, free %ld, pm_ramsize %ld (bytes).",
- sg_host_get_name(vm), sg_host_get_name(pm), vm_ramsize, pm_ramsize - total_ramsize_of_vms, pm_ramsize);
+ vm->cname(), pm->cname(), vm_ramsize, pm_ramsize - total_ramsize_of_vms, pm_ramsize);
THROWF(vm_error, 0, "Memory shortage on host '%s', VM '%s' cannot be started", pm->cname(), vm->cname());
}
}
});
if (TRACE_msg_vm_is_enabled()) {
- container_t vm_container = PJ_container_get(vm->name().c_str());
+ container_t vm_container = PJ_container_get(vm->cname());
type_t type = PJ_type_get("MSG_VM_STATE", vm_container->type);
val_t value = PJ_value_get_or_new("start", "0 0 1", type); // start is blue
new_pajePushState(MSG_get_clock(), vm_container, type, value);
snprintf(key, INSTR_DEFAULT_STR_SIZE, "%lld", counter++);
// start link
- container_t msg = PJ_container_get(vm->name().c_str());
+ container_t msg = PJ_container_get(vm->cname());
type_t type = PJ_type_get("MSG_VM_LINK", PJ_type_get_root());
new_pajeStartLink(MSG_get_clock(), PJ_container_get_root(), type, msg, "M", key);
// destroy existing container of this vm
- container_t existing_container = PJ_container_get(vm->name().c_str());
+ container_t existing_container = PJ_container_get(vm->cname());
PJ_container_remove_from_parent(existing_container);
PJ_container_free(existing_container);
// create new container on the new_host location
- PJ_container_new(vm->cname(), INSTR_MSG_VM, PJ_container_get(sg_host_get_name(ms->dst_pm)));
+ PJ_container_new(vm->cname(), INSTR_MSG_VM, PJ_container_get(ms->dst_pm->cname()));
// end link
- msg = PJ_container_get(vm->name().c_str());
+ msg = PJ_container_get(vm->cname());
type = PJ_type_get("MSG_VM_LINK", PJ_type_get_root());
new_pajeEndLink(MSG_get_clock(), PJ_container_get_root(), type, msg, "M", key);
}
static void start_dirty_page_tracking(msg_vm_t vm)
{
- simgrid::surf::VirtualMachineImpl* pimpl = static_cast<simgrid::s4u::VirtualMachine*>(vm)->pimpl_vm_;
+ simgrid::vm::VirtualMachineImpl* pimpl = static_cast<simgrid::s4u::VirtualMachine*>(vm)->pimpl_vm_;
pimpl->dp_enabled = 1;
if (!pimpl->dp_objs)
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, sg_host_get_name(vm), computed, dp->prev_remaining, remaining, duration, dp->prev_clock, clock);
+ XBT_DEBUG("%s@%s: computed %f ops (remaining %f -> %f) in %f secs (%f -> %f)", key, vm->cname(), computed,
+ dp->prev_remaining, remaining, duration, dp->prev_clock, clock);
return computed;
}
static double lookup_computed_flop_counts(msg_vm_t vm, int stage_for_fancy_debug, int stage2_round_for_fancy_debug)
{
- simgrid::surf::VirtualMachineImpl* pimpl = static_cast<simgrid::s4u::VirtualMachine*>(vm)->pimpl_vm_;
+ simgrid::vm::VirtualMachineImpl* pimpl = static_cast<simgrid::s4u::VirtualMachine*>(vm)->pimpl_vm_;
double total = 0;
char *key = nullptr;
simgrid::s4u::VirtualMachine* vm = dynamic_cast<simgrid::s4u::VirtualMachine*>(host);
if (vm == nullptr)
return;
- simgrid::surf::VirtualMachineImpl* pimpl = static_cast<simgrid::s4u::VirtualMachine*>(vm)->pimpl_vm_;
+ simgrid::vm::VirtualMachineImpl* pimpl = static_cast<simgrid::s4u::VirtualMachine*>(vm)->pimpl_vm_;
double remaining = MSG_task_get_flops_amount(task);
char *key = bprintf("%s-%p", task->name, task);
dp->prev_remaining = remaining;
}
if (!pimpl->dp_objs)
- pimpl->dp_objs = xbt_dict_new();
+ pimpl->dp_objs = xbt_dict_new_homogeneous(nullptr);
xbt_assert(xbt_dict_get_or_null(pimpl->dp_objs, key) == nullptr);
xbt_dict_set(pimpl->dp_objs, key, dp, nullptr);
- XBT_DEBUG("add %s on %s (remaining %f, dp_enabled %d)", key, sg_host_get_name(host), remaining, pimpl->dp_enabled);
+ XBT_DEBUG("add %s on %s (remaining %f, dp_enabled %d)", key, host->cname(), remaining, pimpl->dp_enabled);
xbt_free(key);
}
simgrid::s4u::VirtualMachine* vm = dynamic_cast<simgrid::s4u::VirtualMachine*>(host);
if (vm == nullptr)
return;
- simgrid::surf::VirtualMachineImpl* pimpl = static_cast<simgrid::s4u::VirtualMachine*>(vm)->pimpl_vm_;
+ simgrid::vm::VirtualMachineImpl* pimpl = static_cast<simgrid::s4u::VirtualMachine*>(vm)->pimpl_vm_;
char *key = bprintf("%s-%p", task->name, task);
dirty_page_t dp = (dirty_page_t)(pimpl->dp_objs ? xbt_dict_get_or_null(pimpl->dp_objs, key) : NULL);
xbt_dict_remove(pimpl->dp_objs, key);
xbt_free(dp);
- XBT_DEBUG("del %s on %s", key, sg_host_get_name(host));
+ XBT_DEBUG("del %s on %s", key, host->cname());
xbt_free(key);
}
if(ret == MSG_HOST_FAILURE){
//XBT_DEBUG("SRC host failed during migration of %s (stage %d)", sg_host_name(vm), stage);
MSG_task_destroy(task);
- THROWF(host_error, 0, "SRC host failed during migration of %s (stage %d)", sg_host_get_name(vm), stage);
+ THROWF(host_error, 0, "SRC host failed during migration of %s (stage %d)", vm->cname(), stage);
}else if(ret == MSG_TRANSFER_FAILURE){
//XBT_DEBUG("DST host failed during migration of %s (stage %d)", sg_host_name(vm), stage);
MSG_task_destroy(task);
- THROWF(host_error, 0, "DST host failed during migration of %s (stage %d)", sg_host_get_name(vm), stage);
+ THROWF(host_error, 0, "DST host failed during migration of %s (stage %d)", vm->cname(), stage);
}
double clock_end = MSG_get_clock();
const double mig_speed = params.mig_speed;
double max_downtime = params.max_downtime;
- /* hard code it temporally. Fix Me */
-#define MIGRATION_TIMEOUT_DO_NOT_HARDCODE_ME 10000000.0
- double mig_timeout = MIGRATION_TIMEOUT_DO_NOT_HARDCODE_ME;
+ double mig_timeout = 10000000.0;
double remaining_size = (double) (ramsize + devsize);
double threshold = 0.0;
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 (%lf > %lf), skip stage 2",
- (clock_post_send - clock_prev_send), MIGRATION_TIMEOUT_DO_NOT_HARDCODE_ME);
+ XBT_VERB("The duration of stage 1 exceeds the timeout value, skip stage 2");
skip_stage2 = 1;
}
return 0;
}
-static int do_migration(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm)
+/** @brief Migrate the VM to the given host.
+ * @ingroup msg_VMs
+ */
+void MSG_vm_migrate(msg_vm_t vm, msg_host_t dst_pm)
{
+ /* some thoughts:
+ * - One approach is ...
+ * We first create a new VM (i.e., destination VM) on the destination physical host. The destination VM will
+ * receive the state of the source
+ * VM over network. We will finally destroy the source VM.
+ * - This behavior is similar to the way of migration in the real world.
+ * Even before a migration is completed, we will see a destination VM, consuming resources.
+ * - We have to relocate all processes. The existing process migration code will work for this?
+ * - The name of the VM is a somewhat unique ID in the code. It is tricky for the destination VM?
+ *
+ * - Another one is ...
+ * We update the information of the given VM to place it to the destination physical host.
+ *
+ * The second one would be easier.
+ */
+
+ simgrid::s4u::VirtualMachine* typedVm = static_cast<simgrid::s4u::VirtualMachine*>(vm);
+ simgrid::vm::VirtualMachineImpl* pimpl = typedVm->pimpl_vm_;
+ msg_host_t src_pm = pimpl->getPm();
+
+ if (src_pm->isOff())
+ THROWF(vm_error, 0, "Cannot migrate VM '%s' from host '%s', which is offline.", vm->cname(), src_pm->cname());
+ if (dst_pm->isOff())
+ THROWF(vm_error, 0, "Cannot migrate VM '%s' to host '%s', which is offline.", vm->cname(), dst_pm->cname());
+ if (!MSG_vm_is_running(vm))
+ THROWF(vm_error, 0, "Cannot migrate VM '%s' that is not running yet.", vm->cname());
+ if (typedVm->isMigrating())
+ THROWF(vm_error, 0, "Cannot migrate VM '%s' that is already migrating.", vm->cname());
+
+ pimpl->isMigrating = true;
+
struct migration_session *ms = xbt_new(struct migration_session, 1);
ms->vm = vm;
ms->src_pm = src_pm;
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_TIMEOUT;
- while (ret == MSG_TIMEOUT && MSG_host_is_on(dst_pm)) // Wait while you receive the message o
+ while (ret == MSG_TIMEOUT && dst_pm->isOn()) // The rx will tell me when he gots the VM
ret = MSG_task_receive_with_timeout(&task, ms->mbox_ctl, 4);
+ pimpl->isMigrating = false;
+
xbt_free(ms->mbox_ctl);
xbt_free(ms->mbox);
xbt_free(ms);
XBT_ERROR("SRC crashes, throw an exception (m-control)");
// MSG_process_kill(tx_process); // Adrien, I made a merge on Nov 28th 2014, I'm not sure whether this line is
// required or not
- return -1;
+ THROWF(host_error, 0, "Source host '%s' failed during the migration of VM '%s'.", src_pm->cname(), vm->cname());
} else if ((ret == MSG_TRANSFER_FAILURE) || (ret == MSG_TIMEOUT)) {
// MSG_TIMEOUT here means that MSG_host_is_avail() returned false.
XBT_ERROR("DST crashes, throw an exception (m-control)");
- return -2;
+ THROWF(host_error, 0, "Destination host '%s' failed during the migration of VM '%s'.", dst_pm->cname(),
+ vm->cname());
}
char* expected_task_name = get_mig_task_name(vm, src_pm, dst_pm, 4);
xbt_assert(strcmp(task->name, expected_task_name) == 0);
xbt_free(expected_task_name);
MSG_task_destroy(task);
- return 0;
-}
-
-/** @brief Migrate the VM to the given host.
- * @ingroup msg_VMs
- *
- * FIXME: No migration cost occurs. If you want to simulate this too, you want to use a MSG_task_send() before or after,
- * depending on whether you want to do cold or hot migration.
- */
-void MSG_vm_migrate(msg_vm_t vm, msg_host_t new_pm)
-{
- /* some thoughts:
- * - One approach is ...
- * We first create a new VM (i.e., destination VM) on the destination physical host. The destination VM will
- * receive the state of the source
- * VM over network. We will finally destroy the source VM.
- * - This behavior is similar to the way of migration in the real world.
- * Even before a migration is completed, we will see a destination VM, consuming resources.
- * - We have to relocate all processes. The existing process migration code will work for this?
- * - The name of the VM is a somewhat unique ID in the code. It is tricky for the destination VM?
- *
- * - Another one is ...
- * We update the information of the given VM to place it to the destination physical host.
- *
- * The second one would be easier.
- */
-
- simgrid::s4u::VirtualMachine* typedVm = static_cast<simgrid::s4u::VirtualMachine*>(vm);
- simgrid::surf::VirtualMachineImpl* pimpl = typedVm->pimpl_vm_;
- msg_host_t old_pm = pimpl->getPm();
-
- if (old_pm->isOff())
- THROWF(vm_error, 0, "Cannot migrate VM '%s' from host '%s', which is offline.", vm->cname(), old_pm->cname());
-
- if (new_pm->isOff())
- THROWF(vm_error, 0, "Cannot migrate VM '%s' to host '%s', which is offline.", vm->cname(), new_pm->cname());
-
- if (!MSG_vm_is_running(vm))
- THROWF(vm_error, 0, "Cannot migrate VM '%s' that is not running yet.", vm->cname());
-
- if (typedVm->isMigrating())
- THROWF(vm_error, 0, "Cannot migrate VM '%s' that is already migrating.", vm->cname());
-
- pimpl->isMigrating = true;
-
- int ret = do_migration(vm, old_pm, new_pm);
- if (ret == -1) {
- pimpl->isMigrating = false;
- THROWF(host_error, 0, "Source host '%s' failed during the migration of VM '%s'.", old_pm->cname(), vm->cname());
- } else if (ret == -2) {
- pimpl->isMigrating = false;
- THROWF(host_error, 0, "Destination host '%s' failed during the migration of VM '%s'.", new_pm->cname(),
- vm->cname());
- }
}
/** @brief Immediately suspend the execution of all processes within the given VM.
{
simgrid::simix::kernelImmediate([vm]() {
if (static_cast<simgrid::s4u::VirtualMachine*>(vm)->pimpl_vm_->getState() != SURF_VM_STATE_SAVED)
- THROWF(vm_error, 0, "VM(%s) was not saved", vm->name().c_str());
+ THROWF(vm_error, 0, "VM(%s) was not saved", vm->cname());
- XBT_DEBUG("restore VM(%s), where %d processes exist", vm->name().c_str(),
- xbt_swag_size(sg_host_simix(vm)->process_list));
+ XBT_DEBUG("restore VM(%s), where %d processes exist", vm->cname(), xbt_swag_size(sg_host_simix(vm)->process_list));
/* jump to vm_ws_restore() */
static_cast<simgrid::s4u::VirtualMachine*>(vm)->pimpl_vm_->restore();
