X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/94f980dcb3d25cdf2f85dd9d811de906d2f49eb8..4636ae2abfffeed6d3c21c565f9fdd6908826402:/src/msg/msg_vm.c

diff --git a/src/msg/msg_vm.c b/src/msg/msg_vm.c
index 0bc20d94f7..4539a93497 100644
--- a/src/msg/msg_vm.c
+++ b/src/msg/msg_vm.c
@@ -1,26 +1,21 @@
-/* Copyright (c) 2012-2013. The SimGrid Team.
+/* Copyright (c) 2012-2014. 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. */
 
-// QUESTIONS:
-// 1./ check how and where a new VM is added to the list of the hosts
-// 2./ Diff between SIMIX_Actions and SURF_Actions
-// => SIMIX_actions : point synchro entre processus de niveau (theoretically speaking I do not have to create such SIMIX_ACTION
-// =>  Surf_Actions
+/* TODO:
+ * 1. add the support of trace
+ * 2. use parallel tasks to simulate CPU overhead and remove the very
+ *    experimental code generating micro computation tasks
+ */
+
 
-// TODO
-//	MSG_TRACE can be revisited in order to use  the host
-//	To implement a mixed model between workstation and vm_workstation,
-//     please give a look at surf_model_private_t model_private at SURF Level and to the share resource functions
-//     double (*share_resources) (double now);
-//	For the action into the vm workstation model, we should be able to leverage the usual one (and if needed, look at
-// 		the workstation model.
 
 #include "msg_private.h"
 #include "xbt/sysdep.h"
 #include "xbt/log.h"
+#include "simgrid/platf.h"
 
 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(msg_vm, msg,
                                 "Cloud-oriented parts of the MSG API");
@@ -57,7 +52,7 @@ xbt_dict_t MSG_vm_get_properties(msg_vm_t vm)
 /** \ingroup m_host_management
  * \brief Change the value of a given host property
  *
- * \param host a host
+ * \param vm a vm
  * \param name a property name
  * \param value what to change the property to
  * \param free_ctn the freeing function to use to kill the value on need
@@ -102,7 +97,7 @@ static inline int __MSG_vm_is_state(msg_vm_t vm, e_surf_vm_state_t state)
   return simcall_vm_get_state(vm) == state;
 }
 
-/** @brief Returns whether the given VM has just reated, not running.
+/** @brief Returns whether the given VM has just created, not running.
  *  @ingroup msg_VMs
  */
 int MSG_vm_is_created(msg_vm_t vm)
@@ -123,7 +118,8 @@ int MSG_vm_is_running(msg_vm_t vm)
  */
 int MSG_vm_is_migrating(msg_vm_t vm)
 {
-  return __MSG_vm_is_state(vm, SURF_VM_STATE_MIGRATING);
+  msg_host_priv_t priv = msg_host_resource_priv(vm);
+  return priv->is_migrating;
 }
 
 /** @brief Returns whether the given VM is currently suspended, not running.
@@ -170,30 +166,32 @@ int MSG_vm_is_restoring(msg_vm_t vm)
  *  All parameters are in MBytes
  *
  */
-msg_vm_t MSG_vm_create(msg_host_t ind_pm, const char *name, int ncpus, int ramsize,
-	                                     int net_cap, char *disk_path, int disksize,
-	                                     	 int mig_netspeed, int dp_intensity)
+msg_vm_t MSG_vm_create(msg_host_t ind_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 update_speed = ((double)dp_intensity/100) * mig_netspeed;
-	
-	msg_vm_t vm = MSG_vm_create_core(ind_pm, name);
-	s_ws_params_t params;
-	memset(&params, 0, sizeof(params));
-	params.ramsize = 1L * 1024 * 1024 * ramsize;
-	//params.overcommit = 0;
-	params.devsize = 0;
-	params.skip_stage2 = 0;
-	params.max_downtime = 0.03;
-	params.dp_rate = (update_speed * 1L * 1024 * 1024 ) / host_speed; 
-	params.dp_cap = params.ramsize / 0.9; // working set memory is 90%
-	params.mig_speed = 1L * 1024 * 1024 * mig_netspeed; // mig_speed
-
-   //XBT_INFO("dp rate %f migspeed : %f intensity mem : %d, updatespeed %f, hostspeed %f",params.dp_rate, params.mig_speed, dp_intensity, update_speed, host_speed);
-	simcall_host_set_params(vm, &params);
-
-	return vm;
+  /* For the moment, intensity_rate is the percentage against the migration
+   * bandwidth */
+  double host_speed = MSG_get_host_speed(ind_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;
+  memset(&params, 0, sizeof(params));
+  params.ramsize = (sg_size_t)ramsize * 1024 * 1024;
+  //params.overcommit = 0;
+  params.devsize = 0;
+  params.skip_stage2 = 0;
+  params.max_downtime = 0.03;
+  params.dp_rate = (update_speed * 1024 * 1024) / host_speed;
+  params.dp_cap = params.ramsize * 0.9; // assume working set memory is 90% of ramsize
+  params.mig_speed = (double)mig_netspeed * 1024 * 1024; // mig_speed
+
+  //XBT_INFO("dp rate %f migspeed : %f intensity mem : %d, updatespeed %f, hostspeed %f",params.dp_rate, params.mig_speed, dp_intensity, update_speed, host_speed);
+  simcall_host_set_params(vm, &params);
+
+  return vm;
 }
 
 
@@ -235,6 +233,9 @@ msg_vm_t MSG_vm_create_core(msg_host_t ind_pm, const char *name)
  */
 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));
+
   /* First, terminate all processes on the VM if necessary */
   if (MSG_vm_is_running(vm))
       simcall_vm_shutdown(vm);
@@ -291,67 +292,93 @@ void MSG_vm_shutdown(msg_vm_t vm)
 
 
 /* We have two mailboxes. mbox is used to transfer migration data between
- * source and destiantion PMs. mbox_ctl is used to detect the completion of a
+ * 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. */
-static inline char *get_mig_mbox_src_dst(const char *vm_name, const char *src_pm_name, const char *dst_pm_name)
+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);
+
   return bprintf("__mbox_mig_src_dst:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name);
 }
 
-static inline char *get_mig_mbox_ctl(const char *vm_name, const char *src_pm_name, const char *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);
+
   return bprintf("__mbox_mig_ctl:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name);
 }
 
-static inline char *get_mig_process_tx_name(const char *vm_name, const char *src_pm_name, const char *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);
+
   return bprintf("__pr_mig_tx:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name);
 }
 
-static inline char *get_mig_process_rx_name(const char *vm_name, const char *src_pm_name, const char *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);
+
   return bprintf("__pr_mig_rx:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name);
 }
 
-static inline char *get_mig_task_name(const char *vm_name, const char *src_pm_name, const char *dst_pm_name, int stage)
+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);
+
   return bprintf("__task_mig_stage%d:%s(%s-%s)", stage, vm_name, src_pm_name, dst_pm_name);
 }
 
 static void launch_deferred_exec_process(msg_host_t host, double computation, double prio);
 
+
+struct migration_session {
+  msg_vm_t vm;
+  msg_host_t src_pm;
+  msg_host_t dst_pm;
+
+  /* The miration_rx process uses mbox_ctl to let the caller of do_migration()
+   * know the completion of the migration. */
+  char *mbox_ctl;
+  /* The migration_rx and migration_tx processes use mbox to transfer migration
+   * data. */
+  char *mbox;
+};
+
+
 static int migration_rx_fun(int argc, char *argv[])
 {
   XBT_DEBUG("mig: rx_start");
 
-  xbt_assert(argc == 4);
-  const char *vm_name = argv[1];
-  const char *src_pm_name  = argv[2];
-  const char *dst_pm_name  = argv[3];
-  msg_vm_t vm = MSG_get_host_by_name(vm_name);
-  msg_host_t src_pm = MSG_get_host_by_name(src_pm_name);
-  msg_host_t dst_pm = MSG_get_host_by_name(dst_pm_name);
+  struct migration_session *ms = MSG_process_get_data(MSG_process_self());
 
 
   s_ws_params_t params;
-  simcall_host_get_params(vm, &params);
+  simcall_host_get_params(ms->vm, &params);
   const double xfer_cpu_overhead = params.xfer_cpu_overhead;
 
-
   int need_exit = 0;
 
-  char *mbox = get_mig_mbox_src_dst(vm_name, src_pm_name, dst_pm_name);
-  char *mbox_ctl = get_mig_mbox_ctl(vm_name, src_pm_name, dst_pm_name);
-  char *finalize_task_name = get_mig_task_name(vm_name, src_pm_name, dst_pm_name, 3);
+  char *finalize_task_name = get_mig_task_name(ms->vm, ms->src_pm, ms->dst_pm, 3);
 
   for (;;) {
     msg_task_t task = NULL;
-    MSG_task_recv(&task, mbox);
+    MSG_task_recv(&task, ms->mbox);
     {
       double received = MSG_task_get_data_size(task);
       /* TODO: clean up */
       // const double alpha = 0.22L * 1.0E8 / (80L * 1024 * 1024);
-      launch_deferred_exec_process(vm, received * xfer_cpu_overhead, 1);
+      launch_deferred_exec_process(ms->vm, received * xfer_cpu_overhead, 1);
     }
 
     if (strcmp(task->name, finalize_task_name) == 0)
@@ -365,33 +392,31 @@ static int migration_rx_fun(int argc, char *argv[])
 
 
   /* deinstall the current affinity setting */
-  simcall_vm_set_affinity(vm, src_pm, 0);
+  simcall_vm_set_affinity(ms->vm, ms->src_pm, 0);
 
-  simcall_vm_migrate(vm, dst_pm);
-  simcall_vm_resume(vm);
+  simcall_vm_migrate(ms->vm, ms->dst_pm);
+  simcall_vm_resume(ms->vm);
 
   /* install the affinity setting of the VM on the destination pm */
   {
-    msg_host_priv_t priv = msg_host_resource_priv(vm);
+    msg_host_priv_t priv = msg_host_resource_priv(ms->vm);
 
-    unsigned long affinity_mask = (unsigned long) 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_INFO("set affinity(0x%04lx@%s) for %s", affinity_mask, MSG_host_get_name(dst_pm), MSG_host_get_name(vm));
+    unsigned long affinity_mask = (unsigned long) xbt_dict_get_or_null_ext(priv->affinity_mask_db, (char *) ms->dst_pm, sizeof(msg_host_t));
+    simcall_vm_set_affinity(ms->vm, ms->dst_pm, affinity_mask);
+    XBT_INFO("set affinity(0x%04lx@%s) for %s", affinity_mask, MSG_host_get_name(ms->dst_pm), MSG_host_get_name(ms->vm));
   }
 
   {
-    char *task_name = get_mig_task_name(vm_name, src_pm_name, dst_pm_name, 4);
+    char *task_name = get_mig_task_name(ms->vm, ms->src_pm, ms->dst_pm, 4);
 
     msg_task_t task = MSG_task_create(task_name, 0, 0, NULL);
-    msg_error_t ret = MSG_task_send(task, mbox_ctl);
+    msg_error_t ret = MSG_task_send(task, ms->mbox_ctl);
     xbt_assert(ret == MSG_OK);
 
     xbt_free(task_name);
   }
 
 
-  xbt_free(mbox);
-  xbt_free(mbox_ctl);
   xbt_free(finalize_task_name);
 
   XBT_DEBUG("mig: rx_done");
@@ -453,7 +478,7 @@ static double get_computed(char *key, msg_vm_t vm, dirty_page_t dp, double remai
   double computed = dp->prev_remaining - remaining;
   double duration = clock - dp->prev_clock;
 
-  XBT_DEBUG("%s@%s: computated %f ops (remaining %f -> %f) in %f secs (%f -> %f)",
+  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);
 
   return computed;
@@ -469,7 +494,7 @@ static double lookup_computed_flop_counts(msg_vm_t vm, int stage_for_fancy_debug
   dirty_page_t dp = NULL;
   xbt_dict_foreach(priv->dp_objs, cursor, key, dp) {
     double remaining = MSG_task_get_remaining_computation(dp->task);
-   
+
 	 double clock = MSG_get_clock();
 
     // total += calc_updated_pages(key, vm, dp, remaining, clock);
@@ -500,7 +525,7 @@ 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);
-  char *key = bprintf("%s-%lld", task->name, task->counter);
+  char *key = bprintf("%s-%p", task->name, task);
 
   dirty_page_t dp = xbt_new0(s_dirty_page, 1);
   dp->task = task;
@@ -522,13 +547,13 @@ void MSG_host_del_task(msg_host_t host, msg_task_t task)
 {
   msg_host_priv_t priv = msg_host_resource_priv(host);
 
-  char *key = bprintf("%s-%lld", task->name, task->counter);
+  char *key = bprintf("%s-%p", task->name, task);
 
   dirty_page_t dp = xbt_dict_get_or_null(priv->dp_objs, key);
   xbt_assert(dp->task == task);
 
   /* If we are in the middle of dirty page tracking, we record how much
-   * computaion has been done until now, and keep the information for the
+   * 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);
@@ -557,7 +582,7 @@ static int deferred_exec_fun(int argc, char *argv[])
   double prio = atof(prio_str);
 
   msg_task_t task = MSG_task_create("__task_deferred", computaion, 0, NULL);
-  // XBT_INFO("exec deferred %f", computaion);
+  // XBT_INFO("exec deferred %f", computation);
 
   /* dpt is the results of the VM activity */
   MSG_task_set_priority(task, prio);
@@ -576,14 +601,12 @@ static void launch_deferred_exec_process(msg_host_t host, double computation, do
 
   int nargvs = 4;
   char **argv = xbt_new(char *, nargvs);
-  argv[0] = xbt_strdup(pr_name);
-  argv[1] = bprintf("%lf", computation);
-  argv[2] = bprintf("%lf", prio);
+  argv[0] = pr_name;
+  argv[1] = bprintf("%f", computation);
+  argv[2] = bprintf("%f", prio);
   argv[3] = NULL;
 
   MSG_process_create_with_arguments(pr_name, deferred_exec_fun, NULL, host, nargvs - 1, argv);
-
-  xbt_free(pr_name);
 }
 
 
@@ -626,15 +649,12 @@ static void start_overhead_process(msg_task_t comm_task)
 
   int nargvs = 3;
   char **argv = xbt_new(char *, nargvs);
-  argv[0] = xbt_strdup(pr_name);
-  argv[1] = xbt_strdup(mbox);
+  argv[0] = pr_name;
+  argv[1] = mbox;
   argv[2] = NULL;
 
   // XBT_INFO("micro start: mbox %s", mbox);
   MSG_process_create_with_arguments(pr_name, task_tx_overhead_fun, NULL, MSG_host_self(), nargvs - 1, argv);
-
-  xbt_free(pr_name);
-  xbt_free(mbox);
 }
 
 static void shutdown_overhead_process(msg_task_t comm_task)
@@ -797,10 +817,10 @@ const double alpha = 0.22L * 1.0E8 / (80L * 1024 * 1024);
 #endif
 
 
-static void send_migration_data(const char *vm_name, const char *src_pm_name, const char *dst_pm_name,
-    double size, char *mbox, int stage, int stage2_round, double mig_speed, double xfer_cpu_overhead)
+static void 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 xfer_cpu_overhead)
 {
-  char *task_name = get_mig_task_name(vm_name, src_pm_name, dst_pm_name, stage);
+  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);
 
   /* TODO: clean up */
@@ -833,9 +853,9 @@ static void send_migration_data(const char *vm_name, const char *src_pm_name, co
 
 
   if (stage == 2){
-    XBT_DEBUG("mig-stage%d.%d: sent %f duration %f actual_speed %f (target %f) cpu %f", stage, stage2_round, size, duration, actual_speed, mig_speed, cpu_utilization);}
+    XBT_DEBUG("mig-stage%d.%d: sent %llu duration %f actual_speed %f (target %f) cpu %f", stage, stage2_round, size, duration, actual_speed, mig_speed, cpu_utilization);}
   else{
-    XBT_DEBUG("mig-stage%d: sent %f duration %f actual_speed %f (target %f) cpu %f", stage, size, duration, actual_speed, mig_speed, cpu_utilization);
+    XBT_DEBUG("mig-stage%d: sent %llu duration %f actual_speed %f (target %f) cpu %f", stage, size, duration, actual_speed, mig_speed, cpu_utilization);
   }
 
   xbt_free(task_name);
@@ -870,27 +890,24 @@ static double get_updated_size(double computed, double dp_rate, double dp_cap)
   return updated_size;
 }
 
-static double send_stage1(msg_host_t vm, const char *src_pm_name, const char *dst_pm_name,
-    long ramsize, double mig_speed, double xfer_cpu_overhead, double dp_rate, double dp_cap, double dpt_cpu_overhead)
+static double send_stage1(struct migration_session *ms,
+    sg_size_t ramsize, double mig_speed, double xfer_cpu_overhead, double dp_rate, double dp_cap, double dpt_cpu_overhead)
 {
-  const char *vm_name = MSG_host_get_name(vm);
-  char *mbox = get_mig_mbox_src_dst(vm_name, src_pm_name, dst_pm_name);
 
-  // const long chunksize = 1024 * 1024 * 100;
-  const unsigned long chunksize = 1024u * 1024u * 100000;
-  long remaining = ramsize;
+  // const long chunksize = (sg_size_t)1024 * 1024 * 100;
+  const sg_size_t chunksize = (sg_size_t)1024 * 1024 * 100000;
+  sg_size_t remaining = ramsize;
   double computed_total = 0;
 
   while (remaining > 0) {
-    long datasize = chunksize;
+    sg_size_t datasize = chunksize;
     if (remaining < chunksize)
       datasize = remaining;
 
     remaining -= datasize;
 
-    send_migration_data(vm_name, src_pm_name, dst_pm_name, datasize, mbox, 1, 0, mig_speed, xfer_cpu_overhead);
-
-    double computed = lookup_computed_flop_counts(vm, 1, 0);
+    send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, datasize, ms->mbox, 1, 0, mig_speed, xfer_cpu_overhead);
+    double computed = lookup_computed_flop_counts(ms->vm, 1, 0);
     computed_total += computed;
 
     // {
@@ -919,17 +936,12 @@ static int migration_tx_fun(int argc, char *argv[])
 {
   XBT_DEBUG("mig: tx_start");
 
-  xbt_assert(argc == 4);
-  const char *vm_name = argv[1];
-  const char *src_pm_name  = argv[2];
-  const char *dst_pm_name  = argv[3];
-  msg_vm_t vm = MSG_get_host_by_name(vm_name);
-
+  struct migration_session *ms = MSG_process_get_data(MSG_process_self());
 
   s_ws_params_t params;
-  simcall_host_get_params(vm, &params);
-  const long ramsize        = params.ramsize;
-  const long devsize        = params.devsize;
+  simcall_host_get_params(ms->vm, &params);
+  const sg_size_t ramsize   = params.ramsize;
+  const sg_size_t devsize   = params.devsize;
   const int skip_stage1     = params.skip_stage1;
   const int skip_stage2     = params.skip_stage2;
   const double dp_rate      = params.dp_rate;
@@ -954,12 +966,11 @@ static int migration_tx_fun(int argc, char *argv[])
   if (ramsize == 0)
     XBT_WARN("migrate a VM, but ramsize is zero");
 
-  char *mbox = get_mig_mbox_src_dst(vm_name, src_pm_name, dst_pm_name);
 
   XBT_INFO("mig-stage1: remaining_size %f", remaining_size);
 
   /* Stage1: send all memory pages to the destination. */
-  start_dirty_page_tracking(vm);
+  start_dirty_page_tracking(ms->vm);
 
   double computed_during_stage1 = 0;
   if (!skip_stage1) {
@@ -968,13 +979,13 @@ static int migration_tx_fun(int argc, char *argv[])
     /* send ramsize, but split it */
     double clock_prev_send = MSG_get_clock();
 
-    computed_during_stage1 = send_stage1(vm, src_pm_name, dst_pm_name, ramsize, mig_speed, xfer_cpu_overhead, dp_rate, dp_cap, dpt_cpu_overhead);
+    computed_during_stage1 = send_stage1(ms, ramsize, mig_speed, xfer_cpu_overhead, dp_rate, dp_cap, dpt_cpu_overhead);
     remaining_size -= ramsize;
 
     double clock_post_send = MSG_get_clock();
     double bandwidth = ramsize / (clock_post_send - clock_prev_send);
     threshold = get_threshold_value(bandwidth, max_downtime);
-    XBT_INFO("actual banwdidth %f, threshold %f", bandwidth / 1024 / 1024, threshold);
+    XBT_INFO("actual bandwidth %f (MB/s), threshold %f", bandwidth / 1024 / 1024, threshold);
   }
 
 
@@ -996,7 +1007,7 @@ static int migration_tx_fun(int argc, char *argv[])
       /* just after stage1, nothing has been updated. But, we have to send the data updated during stage1 */
       updated_size = get_updated_size(computed_during_stage1, dp_rate, dp_cap);
     } else {
-      double computed = lookup_computed_flop_counts(vm, 2, stage2_round);
+      double computed = lookup_computed_flop_counts(ms->vm, 2, stage2_round);
       updated_size = get_updated_size(computed, dp_rate, dp_cap);
     }
 
@@ -1024,18 +1035,13 @@ static int migration_tx_fun(int argc, char *argv[])
 
     double clock_prev_send = MSG_get_clock();
 
-    send_migration_data(vm_name, src_pm_name, dst_pm_name, updated_size, mbox, 2, stage2_round, mig_speed, xfer_cpu_overhead);
+    send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, updated_size, ms->mbox, 2, stage2_round, mig_speed, xfer_cpu_overhead);
 
     double clock_post_send = MSG_get_clock();
 
     double bandwidth = updated_size / (clock_post_send - clock_prev_send);
     threshold = get_threshold_value(bandwidth, max_downtime);
-    XBT_INFO("actual banwdidth %f, threshold %f", bandwidth / 1024 / 1024, threshold);
-
-
-
-
-
+    XBT_INFO("actual bandwidth %f, threshold %f", bandwidth / 1024 / 1024, threshold);
 
 
     remaining_size -= updated_size;
@@ -1046,12 +1052,11 @@ static int migration_tx_fun(int argc, char *argv[])
 stage3:
   /* Stage3: stop the VM and copy the rest of states. */
   XBT_INFO("mig-stage3: remaining_size %f", remaining_size);
-  simcall_vm_suspend(vm);
-  stop_dirty_page_tracking(vm);
+  simcall_vm_suspend(ms->vm);
+  stop_dirty_page_tracking(ms->vm);
 
-  send_migration_data(vm_name, src_pm_name, dst_pm_name, remaining_size, mbox, 3, 0, mig_speed, xfer_cpu_overhead);
+  send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, remaining_size, ms->mbox, 3, 0, mig_speed, xfer_cpu_overhead);
 
-  xbt_free(mbox);
 
   XBT_DEBUG("mig: tx_done");
 
@@ -1062,49 +1067,52 @@ stage3:
 
 static void do_migration(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm)
 {
-  char *mbox_ctl = get_mig_mbox_ctl(sg_host_name(vm), sg_host_name(src_pm), sg_host_name(dst_pm));
+  struct migration_session *ms = xbt_new(struct migration_session, 1);
+  ms->vm = vm;
+  ms->src_pm = src_pm;
+  ms->dst_pm = dst_pm;
+  ms->mbox_ctl = get_mig_mbox_ctl(vm, src_pm, dst_pm);
+  ms->mbox = get_mig_mbox_src_dst(vm, src_pm, dst_pm);
+
+  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_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;
+ 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;
+ MSG_process_create_with_arguments(pr_tx_name, migration_tx_fun, ms, src_pm, 1, argv);
+ }
+#endif
+
 
-  {
-    char *pr_name = get_mig_process_rx_name(sg_host_name(vm), sg_host_name(src_pm), sg_host_name(dst_pm));
-    int nargvs = 5;
-    char **argv = xbt_new(char *, nargvs);
-    argv[0] = xbt_strdup(pr_name);
-    argv[1] = xbt_strdup(sg_host_name(vm));
-    argv[2] = xbt_strdup(sg_host_name(src_pm));
-    argv[3] = xbt_strdup(sg_host_name(dst_pm));
-    argv[4] = NULL;
-
-    MSG_process_create_with_arguments(pr_name, migration_rx_fun, NULL, dst_pm, nargvs - 1, argv);
-
-    xbt_free(pr_name);
-  }
 
-  {
-    char *pr_name = get_mig_process_tx_name(sg_host_name(vm), sg_host_name(src_pm), sg_host_name(dst_pm));
-    int nargvs = 5;
-    char **argv = xbt_new(char *, nargvs);
-    argv[0] = xbt_strdup(pr_name);
-    argv[1] = xbt_strdup(sg_host_name(vm));
-    argv[2] = xbt_strdup(sg_host_name(src_pm));
-    argv[3] = xbt_strdup(sg_host_name(dst_pm));
-    argv[4] = NULL;
-    MSG_process_create_with_arguments(pr_name, migration_tx_fun, NULL, src_pm, nargvs - 1, argv);
-
-    xbt_free(pr_name);
-  }
 
   /* wait until the migration have finished */
   {
     msg_task_t task = NULL;
-    msg_error_t ret = MSG_task_recv(&task, mbox_ctl);
+    msg_error_t ret = MSG_task_recv(&task, ms->mbox_ctl);
+
     xbt_assert(ret == MSG_OK);
 
-    char *expected_task_name = get_mig_task_name(sg_host_name(vm), sg_host_name(src_pm), sg_host_name(dst_pm), 4);
+    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);
   }
 
-  xbt_free(mbox_ctl);
+  xbt_free(ms->mbox_ctl);
+  xbt_free(ms->mbox);
+  xbt_free(ms);
 }
 
 
@@ -1140,12 +1148,18 @@ void MSG_vm_migrate(msg_vm_t vm, msg_host_t new_pm)
 
   msg_host_t old_pm = simcall_vm_get_pm(vm);
 
-  if (simcall_vm_get_state(vm) != SURF_VM_STATE_RUNNING)
+  if (!MSG_vm_is_running(vm))
     THROWF(vm_error, 0, "VM(%s) is not running", sg_host_name(vm));
 
-  do_migration(vm, old_pm, new_pm);
+  if (MSG_vm_is_migrating(vm))
+    THROWF(vm_error, 0, "VM(%s) is already migrating", sg_host_name(vm));
+
+  msg_host_priv_t priv = msg_host_resource_priv(vm);
+  priv->is_migrating = 1;
 
+  do_migration(vm, old_pm, new_pm);
 
+  priv->is_migrating = 0;
 
   XBT_DEBUG("VM(%s) moved from PM(%s) to PM(%s)", vm->key, old_pm->key, new_pm->key);
 
@@ -1158,13 +1172,16 @@ void MSG_vm_migrate(msg_vm_t vm, msg_host_t new_pm)
 /** @brief Immediately suspend the execution of all processes within the given VM.
  *  @ingroup msg_VMs
  *
- * This function stops the exection of the VM. All the processes on this VM
- * will pause. The state of the VM is perserved. We can later resume it again.
+ * This function stops the execution of the VM. All the processes on this VM
+ * will pause. The state of the VM is preserved. We can later resume it again.
  *
  * No suspension cost occurs.
  */
 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));
+
   simcall_vm_suspend(vm);
 
   XBT_DEBUG("vm_suspend done");
@@ -1193,8 +1210,8 @@ void MSG_vm_resume(msg_vm_t vm)
 /** @brief Immediately save the execution of all processes within the given VM.
  *  @ingroup msg_VMs
  *
- * This function stops the exection of the VM. All the processes on this VM
- * will pause. The state of the VM is perserved. We can later resume it again.
+ * This function stops the execution of the VM. All the processes on this VM
+ * will pause. The state of the VM is preserved. We can later resume it again.
  *
  * FIXME: No suspension cost occurs. If you want to simulate this too, you want to
  * use a \ref MSG_file_write() before or after, depending on the exact semantic
@@ -1202,6 +1219,9 @@ void MSG_vm_resume(msg_vm_t vm)
  */
 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));
+
   simcall_vm_save(vm);
   #ifdef HAVE_TRACING
   TRACE_msg_vm_save(vm);
@@ -1243,11 +1263,11 @@ msg_host_t MSG_vm_get_pm(msg_vm_t vm)
  * For example,
  *  On PM0, there are Task1 and VM0.
  *  On VM0, there is Task2.
- * Now we bound 75% to Task1@PM0 and bound 25% to Task2@VM0.
+ * Now we bound 75% to Task1\@PM0 and bound 25% to Task2\@VM0.
  * Then, 
- *  Task1@PM0 gets 50%.
- *  Task2@VM0 gets 25%.
- * This is NOT 75% for Task1@PM0 and 25% for Task2@VM0, respectively.
+ *  Task1\@PM0 gets 50%.
+ *  Task2\@VM0 gets 25%.
+ * This is NOT 75% for Task1\@PM0 and 25% for Task2\@VM0, respectively.
  *
  * This is because a VM has the dummy CPU action in the PM layer. Putting a
  * task on the VM does not affect the bound of the dummy CPU action. The bound
@@ -1255,7 +1275,7 @@ msg_host_t MSG_vm_get_pm(msg_vm_t vm)
  *
  * There are some solutions for this problem. One option is to update the bound
  * of the dummy CPU action automatically. It should be the sum of all tasks on
- * the VM. But, this solution might be costy, because we have to scan all tasks
+ * the VM. But, this solution might be costly, because we have to scan all tasks
  * on the VM in share_resource() or we have to trap both the start and end of
  * task execution.
  *