-/* Copyright (c) 2012-2014. The SimGrid Team.
+/* Copyright (c) 2012-2015. The SimGrid Team.
* All rights reserved. */
/* This program is free software; you can redistribute it and/or modify it
#include "xbt/sysdep.h"
#include "xbt/log.h"
#include "simgrid/platf.h"
+#include "simgrid/host.h"
XBT_LOG_NEW_DEFAULT_SUBCATEGORY(msg_vm, msg,
"Cloud-oriented parts of the MSG API");
msg_vm_t MSG_vm_get_by_name(const char *name)
{
- return MSG_get_host_by_name(name);
+ return MSG_host_by_name(name);
}
/** \ingroup m_vm_management
*/
int MSG_vm_is_migrating(msg_vm_t vm)
{
- msg_host_priv_t priv = msg_host_resource_priv(vm);
+ msg_host_priv_t priv = sg_host_msg(vm);
return priv->is_migrating;
}
/** @brief Create a new VM with specified parameters.
* @ingroup msg_VMs*
- * All parameters are in MBytes
+ * @param pm Physical machine that will host the VM
+ * @param net_cap Maximal bandwidth that the VM can consume (in MByte/s)
+ * @param disk_path (unused) Path to the image that boots
+ * @param disksize (unused) will represent the size of the VM (will be used during migrations)
+ * @param mig_netspeed Amount of Mbyte/s allocated to the migration (cannot be larger than net_cap). Use 0 if unsure.
+ * @param dp_intensity Dirty page percentage according to migNetSpeed, [0-100]. Use 0 if unsure.
*
*/
-msg_vm_t MSG_vm_create(msg_host_t ind_pm, const char *name,
+msg_vm_t MSG_vm_create(msg_host_t pm, const char *name,
int ncpus, int ramsize,
int net_cap, char *disk_path, int disksize,
int mig_netspeed, int dp_intensity)
{
/* For the moment, intensity_rate is the percentage against the migration
* bandwidth */
- double host_speed = MSG_get_host_speed(ind_pm);
+ double host_speed = MSG_get_host_speed(pm);
double update_speed = ((double)dp_intensity/100) * mig_netspeed;
- msg_vm_t vm = MSG_vm_create_core(ind_pm, name);
- s_ws_params_t params;
+ msg_vm_t vm = MSG_vm_create_core(pm, name);
+ s_vm_params_t params;
memset(¶ms, 0, sizeof(params));
params.ramsize = (sg_size_t)ramsize * 1024 * 1024;
//params.overcommit = 0;
/* make sure the VM of the same name does not exit */
{
xbt_dictelm_t ind_host_tmp = xbt_lib_get_elm_or_null(host_lib, name);
- if (ind_host_tmp && xbt_lib_get_level(ind_host_tmp, SIMIX_HOST_LEVEL) != NULL) {
+ if (ind_host_tmp && sg_host_simix(ind_host_tmp) != NULL) {
XBT_ERROR("host %s already exits", name);
return NULL;
}
XBT_DEBUG("A new VM (%s) has been created", name);
- #ifdef HAVE_TRACING
TRACE_msg_vm_create(name, ind_pm);
- #endif
return ind_vm;
}
void MSG_vm_destroy(msg_vm_t vm)
{
if (MSG_vm_is_migrating(vm))
- THROWF(vm_error, 0, "VM(%s) is migrating", sg_host_name(vm));
+ THROWF(vm_error, 0, "VM(%s) is migrating", sg_host_get_name(vm));
/* First, terminate all processes on the VM if necessary */
if (MSG_vm_is_running(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
}
* migration. The names of these mailboxes must not conflict with others. */
static inline char *get_mig_mbox_src_dst(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm)
{
- char *vm_name = sg_host_name(vm);
- char *src_pm_name = sg_host_name(src_pm);
- char *dst_pm_name = sg_host_name(dst_pm);
+ char *vm_name = sg_host_get_name(vm);
+ char *src_pm_name = sg_host_get_name(src_pm);
+ char *dst_pm_name = sg_host_get_name(dst_pm);
return bprintf("__mbox_mig_src_dst:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name);
}
static inline char *get_mig_mbox_ctl(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm)
{
- char *vm_name = sg_host_name(vm);
- char *src_pm_name = sg_host_name(src_pm);
- char *dst_pm_name = sg_host_name(dst_pm);
+ char *vm_name = sg_host_get_name(vm);
+ char *src_pm_name = sg_host_get_name(src_pm);
+ char *dst_pm_name = sg_host_get_name(dst_pm);
return bprintf("__mbox_mig_ctl:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name);
}
static inline char *get_mig_process_tx_name(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm)
{
- char *vm_name = sg_host_name(vm);
- char *src_pm_name = sg_host_name(src_pm);
- char *dst_pm_name = sg_host_name(dst_pm);
+ char *vm_name = sg_host_get_name(vm);
+ char *src_pm_name = sg_host_get_name(src_pm);
+ char *dst_pm_name = sg_host_get_name(dst_pm);
return bprintf("__pr_mig_tx:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name);
}
static inline char *get_mig_process_rx_name(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm)
{
- char *vm_name = sg_host_name(vm);
- char *src_pm_name = sg_host_name(src_pm);
- char *dst_pm_name = sg_host_name(dst_pm);
+ char *vm_name = sg_host_get_name(vm);
+ char *src_pm_name = sg_host_get_name(src_pm);
+ char *dst_pm_name = sg_host_get_name(dst_pm);
return bprintf("__pr_mig_rx:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name);
}
static inline char *get_mig_task_name(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm, int stage)
{
- char *vm_name = sg_host_name(vm);
- char *src_pm_name = sg_host_name(src_pm);
- char *dst_pm_name = sg_host_name(dst_pm);
+ char *vm_name = sg_host_get_name(vm);
+ char *src_pm_name = sg_host_get_name(src_pm);
+ char *dst_pm_name = sg_host_get_name(dst_pm);
return bprintf("__task_mig_stage%d:%s(%s-%s)", stage, vm_name, src_pm_name, dst_pm_name);
}
// The structure has been created in the do_migration function and should only be freed in the same place ;)
struct migration_session *ms = MSG_process_get_data(MSG_process_self());
- s_ws_params_t params;
+ s_vm_params_t params;
simcall_host_get_params(ms->vm, ¶ms);
int need_exit = 0;
msg_vm_t vm = ms->vm;
msg_host_t src_pm = ms->src_pm;
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);
+ msg_host_priv_t priv = sg_host_msg(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 */
{
- unsigned long affinity_mask = (unsigned long) xbt_dict_get_or_null_ext(priv->affinity_mask_db, (char *)dst_pm, sizeof(msg_host_t));
+ unsigned long affinity_mask = (unsigned long)(uintptr_t) xbt_dict_get_or_null_ext(priv->affinity_mask_db, (char *)dst_pm, sizeof(msg_host_t));
simcall_vm_set_affinity(vm, dst_pm, affinity_mask);
XBT_DEBUG("set affinity(0x%04lx@%s) for %s", affinity_mask, MSG_host_get_name(dst_pm), MSG_host_get_name(vm));
}
{
// Now the VM is running on the new host (the migration is completed) (even if the SRC crash)
- msg_host_priv_t priv = msg_host_resource_priv(vm);
+ msg_host_priv_t priv = sg_host_msg(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
{
static void reset_dirty_pages(msg_vm_t vm)
{
- msg_host_priv_t priv = msg_host_resource_priv(vm);
+ msg_host_priv_t priv = sg_host_msg(vm);
char *key = NULL;
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;
static void start_dirty_page_tracking(msg_vm_t vm)
{
- msg_host_priv_t priv = msg_host_resource_priv(vm);
+ msg_host_priv_t priv = sg_host_msg(vm);
priv->dp_enabled = 1;
reset_dirty_pages(vm);
static void stop_dirty_page_tracking(msg_vm_t vm)
{
- msg_host_priv_t priv = msg_host_resource_priv(vm);
+ msg_host_priv_t priv = sg_host_msg(vm);
priv->dp_enabled = 0;
}
double duration = clock - dp->prev_clock;
XBT_DEBUG("%s@%s: computed %f ops (remaining %f -> %f) in %f secs (%f -> %f)",
- key, sg_host_name(vm), computed, dp->prev_remaining, remaining, duration, dp->prev_clock, clock);
+ key, sg_host_get_name(vm), 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)
{
- msg_host_priv_t priv = msg_host_resource_priv(vm);
+ msg_host_priv_t priv = sg_host_msg(vm);
double total = 0;
char *key = NULL;
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();
// msg_process_t MSG_process_create(const char *name, xbt_main_func_t code, void *data, msg_host_t host)
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);
+ msg_host_priv_t priv = sg_host_msg(host);
+ 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);
xbt_assert(xbt_dict_get_or_null(priv->dp_objs, key) == NULL);
xbt_dict_set(priv->dp_objs, key, dp, NULL);
- XBT_DEBUG("add %s on %s (remaining %f, dp_enabled %d)", key, sg_host_name(host), remaining, priv->dp_enabled);
+ XBT_DEBUG("add %s on %s (remaining %f, dp_enabled %d)", key, sg_host_get_name(host), remaining, priv->dp_enabled);
xbt_free(key);
}
void MSG_host_del_task(msg_host_t host, msg_task_t task)
{
- msg_host_priv_t priv = msg_host_resource_priv(host);
+ msg_host_priv_t priv = sg_host_msg(host);
char *key = bprintf("%s-%p", task->name, task);
* 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);
xbt_dict_remove(priv->dp_objs, key);
xbt_free(dp);
- XBT_DEBUG("del %s on %s", key, sg_host_name(host));
+ XBT_DEBUG("del %s on %s", key, sg_host_get_name(host));
xbt_free(key);
}
{
sg_size_t sent = 0;
char *task_name = get_mig_task_name(vm, src_pm, dst_pm, stage);
- msg_task_t task = MSG_task_create(task_name, 0, size, NULL);
+ msg_task_t task = MSG_task_create(task_name, 0, (double)size, NULL);
/* TODO: clean up */
if (ret == MSG_OK) {
sent = size;
} else if (ret == MSG_TIMEOUT) {
- sg_size_t remaining = MSG_task_get_remaining_communication(task);
+ sg_size_t remaining = (sg_size_t)MSG_task_get_remaining_communication(task);
sent = size - remaining;
XBT_INFO("timeout (%lf s) in sending_migration_data, remaining %llu bytes of %llu",
timeout, remaining, size);
if(ret == MSG_HOST_FAILURE){
//XBT_INFO("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_name(vm), stage);
+ THROWF(host_error, 0, "SRC host failed during migration of %s (stage %d)", sg_host_get_name(vm), stage);
}else if(ret == MSG_TRANSFER_FAILURE){
//XBT_INFO("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_name(vm), stage);
+ THROWF(host_error, 0, "DST host failed during migration of %s (stage %d)", sg_host_get_name(vm), stage);
}
double clock_end = MSG_get_clock();
// Note that the ms structure has been allocated in do_migration and hence should be freed in the same function ;)
struct migration_session *ms = MSG_process_get_data(MSG_process_self());
- s_ws_params_t params;
+ s_vm_params_t params;
simcall_host_get_params(ms->vm, ¶ms);
const sg_size_t ramsize = params.ramsize;
const sg_size_t devsize = params.devsize;
#define MIGRATION_TIMEOUT_DO_NOT_HARDCODE_ME 10000000.0
double mig_timeout = MIGRATION_TIMEOUT_DO_NOT_HARDCODE_ME;
- double remaining_size = ramsize + devsize;
+ double remaining_size = (double) (ramsize + devsize);
double threshold = 0.0;
/* check parameters */
/* 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 {
- send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, remaining_size, ms->mbox, 3, 0, mig_speed, -1);
+ XBT_DEBUG("Stage 3: Gonna send %f", remaining_size);
+ send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, (sg_size_t)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)
// Stop the dirty page tracking an return (there is no memory space to release)
}
// 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_get_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_get_name(new_pm));
+
if (!MSG_vm_is_running(vm))
- THROWF(vm_error, 0, "VM(%s) is not running", sg_host_name(vm));
+ THROWF(vm_error, 0, "VM(%s) is not running", sg_host_get_name(vm));
if (MSG_vm_is_migrating(vm))
- THROWF(vm_error, 0, "VM(%s) is already migrating", sg_host_name(vm));
+ THROWF(vm_error, 0, "VM(%s) is already migrating", sg_host_get_name(vm));
- msg_host_priv_t priv = msg_host_resource_priv(vm);
+ msg_host_priv_t priv = sg_host_msg(vm);
priv->is_migrating = 1;
{
// 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
}
void MSG_vm_suspend(msg_vm_t vm)
{
if (MSG_vm_is_migrating(vm))
- THROWF(vm_error, 0, "VM(%s) is migrating", sg_host_name(vm));
+ THROWF(vm_error, 0, "VM(%s) is migrating", sg_host_get_name(vm));
simcall_vm_suspend(vm);
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
}
void MSG_vm_save(msg_vm_t vm)
{
if (MSG_vm_is_migrating(vm))
- THROWF(vm_error, 0, "VM(%s) is migrating", sg_host_name(vm));
+ THROWF(vm_error, 0, "VM(%s) is migrating", sg_host_get_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
}
*/
void MSG_vm_set_bound(msg_vm_t vm, double bound)
{
- return simcall_vm_set_bound(vm, bound);
+ simcall_vm_set_bound(vm, bound);
}
*/
void MSG_vm_set_affinity(msg_vm_t vm, msg_host_t pm, unsigned long mask)
{
- msg_host_priv_t priv = msg_host_resource_priv(vm);
+ msg_host_priv_t priv = sg_host_msg(vm);
if (mask == 0)
xbt_dict_remove_ext(priv->affinity_mask_db, (char *) pm, sizeof(pm));
else
- xbt_dict_set_ext(priv->affinity_mask_db, (char *) pm, sizeof(pm), (void *) mask, NULL);
+ xbt_dict_set_ext(priv->affinity_mask_db, (char *) pm, sizeof(pm), (void *)(uintptr_t) mask, NULL);
msg_host_t pm_now = MSG_vm_get_pm(vm);
if (pm_now == pm) {