XBT_DEBUG("A new VM (%s) has been created", name);
- #ifdef HAVE_TRACING
TRACE_msg_vm_create(name, ind_pm);
- #endif
return ind_vm;
}
__MSG_host_destroy(vm);
- #ifdef HAVE_TRACING
TRACE_msg_vm_end(vm);
- #endif
}
{
simcall_vm_start(vm);
- #ifdef HAVE_TRACING
TRACE_msg_vm_start(vm);
- #endif
}
/* msg_vm_t equals to msg_host_t */
simcall_vm_shutdown(vm);
- // #ifdef HAVE_TRACING
// TRACE_msg_vm_(vm);
- // #endif
}
msg_host_t dst_pm = ms-> dst_pm;
msg_host_priv_t priv = msg_host_resource_priv(vm);
-// TODO: we have an issue, if the DST node is turning off during the three next calls, then the VM is in an inconsistent state
-// I should check with Takahiro in order to make this portion of code atomic
- /* deinstall the current affinity setting for the CPU */
- simcall_vm_set_affinity(vm, src_pm, 0);
-
- /* Update the vm location */
- simcall_vm_migrate(vm, dst_pm);
-
- /* Resume the VM */
- simcall_vm_resume(vm);
+// // TODO: we have an issue, if the DST node is turning off during the three next calls, then the VM is in an inconsistent state
+// // I should check with Takahiro in order to make this portion of code atomic
+// /* deinstall the current affinity setting for the CPU */
+// simcall_vm_set_affinity(vm, src_pm, 0);
+//
+// /* Update the vm location */
+// simcall_vm_migrate(vm, dst_pm);
+//
+// /* Resume the VM */
+// simcall_vm_resume(vm);
+//
+ simcall_vm_migratefrom_resumeto(vm, src_pm, dst_pm);
/* install the affinity setting of the VM on the destination pm */
{
msg_host_priv_t priv = msg_host_resource_priv(vm);
priv->is_migrating = 0;
XBT_DEBUG("VM(%s) moved from PM(%s) to PM(%s)", ms->vm->key, ms->src_pm->key, ms->dst_pm->key);
- #ifdef HAVE_TRACING
- TRACE_msg_vm_change_host(ms->vm, ms->src_pm, ms->dst_pm);
- #endif
-
+ TRACE_msg_vm_change_host(ms->vm, ms->src_pm, ms->dst_pm);
}
// Inform the SRC that the migration has been correctly performed
{
xbt_dict_cursor_t cursor = NULL;
dirty_page_t dp = NULL;
xbt_dict_foreach(priv->dp_objs, cursor, key, dp) {
- double remaining = MSG_task_get_remaining_computation(dp->task);
+ double remaining = MSG_task_get_flops_amount(dp->task);
dp->prev_clock = MSG_get_clock();
dp->prev_remaining = remaining;
xbt_dict_cursor_t cursor = NULL;
dirty_page_t dp = NULL;
xbt_dict_foreach(priv->dp_objs, cursor, key, dp) {
- double remaining = MSG_task_get_remaining_computation(dp->task);
+ double remaining = MSG_task_get_flops_amount(dp->task);
double clock = MSG_get_clock();
void MSG_host_add_task(msg_host_t host, msg_task_t task)
{
msg_host_priv_t priv = msg_host_resource_priv(host);
- double remaining = MSG_task_get_remaining_computation(task);
+ double remaining = MSG_task_get_flops_amount(task);
char *key = bprintf("%s-%p", task->name, task);
dirty_page_t dp = xbt_new0(s_dirty_page, 1);
* computation has been done until now, and keep the information for the
* lookup_() function that will called soon. */
if (priv->dp_enabled) {
- double remaining = MSG_task_get_remaining_computation(task);
+ double remaining = MSG_task_get_flops_amount(task);
double clock = MSG_get_clock();
// double updated = calc_updated_pages(key, host, dp, remaining, clock);
double updated = get_computed(key, host, dp, remaining, clock);
/* At stage 1, we do not need timeout. We have to send all the memory
* pages even though the duration of this tranfer exceeds the timeout
* value. */
+ XBT_INFO("Stage 1: Gonna send %llu", 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);
sg_size_t sent = 0;
double clock_prev_send = MSG_get_clock();
TRY {
+ XBT_DEBUG("Stage 2, gonna send %llu", updated_size);
sent = send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, updated_size, ms->mbox, 2, stage2_round, mig_speed, mig_timeout);
} CATCH_ANONYMOUS {
//hostfailure (if you want to know whether this is the SRC or the DST please check directly in send_migration_data code)
stop_dirty_page_tracking(ms->vm);
TRY {
+ XBT_DEBUG("Stage 3: Gonna send %f", remaining_size);
send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, remaining_size, ms->mbox, 3, 0, mig_speed, -1);
} CATCH_ANONYMOUS {
//hostfailure (if you want to know whether this is the SRC or the DST please check directly in send_migration_data code)
}
// At that point the Migration is considered valid for the SRC node but remind that the DST side should relocate effectively the VM on the DST node.
- XBT_INFO("mig: tx_done");
+ XBT_DEBUG("mig: tx_done");
return 0;
}
char *pr_rx_name = get_mig_process_rx_name(vm, src_pm, dst_pm);
char *pr_tx_name = get_mig_process_tx_name(vm, src_pm, dst_pm);
- msg_process_t tx_process, rx_process;
+// msg_process_t tx_process, rx_process;
// MSG_process_create(pr_rx_name, migration_rx_fun, ms, dst_pm);
// MSG_process_create(pr_tx_name, migration_tx_fun, ms, src_pm);
#if 1
char **argv = xbt_new(char *, 2);
argv[0] = pr_rx_name;
argv[1] = NULL;
- rx_process = MSG_process_create_with_arguments(pr_rx_name, migration_rx_fun, ms, dst_pm, 1, argv);
+/*rx_process = */ MSG_process_create_with_arguments(pr_rx_name, migration_rx_fun, ms, dst_pm, 1, argv);
}
{
char **argv = xbt_new(char *, 2);
argv[0] = pr_tx_name;
argv[1] = NULL;
- tx_process = MSG_process_create_with_arguments(pr_tx_name, migration_tx_fun, ms, src_pm, 1, argv);
+/* tx_process = */MSG_process_create_with_arguments(pr_tx_name, migration_tx_fun, ms, src_pm, 1, argv);
}
#endif
msg_task_t task = NULL;
msg_error_t ret = MSG_TIMEOUT;
while (ret == MSG_TIMEOUT && MSG_host_is_on(dst_pm)) //Wait while you receive the message o
- ret = MSG_task_receive_with_timeout(&task, ms->mbox_ctl, 3);
+ ret = MSG_task_receive_with_timeout(&task, ms->mbox_ctl, 4);
xbt_free(ms->mbox_ctl);
xbt_free(ms->mbox);
// 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
XBT_INFO("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
+ //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;
}
else if((ret == MSG_TRANSFER_FAILURE) || (ret == MSG_TIMEOUT)){ // MSG_TIMEOUT here means that MSG_host_is_avail() returned false.
msg_host_t old_pm = simcall_vm_get_pm(vm);
+ if(MSG_host_is_off(old_pm))
+ THROWF(vm_error, 0, "SRC host(%s) seems off, cannot start a migration", sg_host_name(old_pm));
+
+ if(MSG_host_is_off(new_pm))
+ THROWF(vm_error, 0, "DST host(%s) seems off, cannot start a migration", sg_host_name(new_pm));
+
if (!MSG_vm_is_running(vm))
THROWF(vm_error, 0, "VM(%s) is not running", sg_host_name(vm));
// This part is done in the RX code, to handle the corner case where SRC can crash just at the end of the migration process
// In that case, the VM has been already assigned to the DST node.
//XBT_DEBUG("VM(%s) moved from PM(%s) to PM(%s)", vm->key, old_pm->key, new_pm->key);
- //#ifdef HAVE_TRACING
//TRACE_msg_vm_change_host(vm, old_pm, new_pm);
- //#endif
}
XBT_DEBUG("vm_suspend done");
- #ifdef HAVE_TRACING
TRACE_msg_vm_suspend(vm);
- #endif
}
{
simcall_vm_resume(vm);
- #ifdef HAVE_TRACING
TRACE_msg_vm_resume(vm);
- #endif
}
THROWF(vm_error, 0, "VM(%s) is migrating", sg_host_name(vm));
simcall_vm_save(vm);
- #ifdef HAVE_TRACING
TRACE_msg_vm_save(vm);
- #endif
}
/** @brief Restore the execution of the VM. All processes on the VM run again.
{
simcall_vm_restore(vm);
- #ifdef HAVE_TRACING
TRACE_msg_vm_restore(vm);
- #endif
}