X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/4a6b0a991a67e6f2f67c03fed43529e078da7115..a15797ea55151ddfdbae48147e74159efe01b411:/src/msg/msg_vm.cpp diff --git a/src/msg/msg_vm.cpp b/src/msg/msg_vm.cpp index cbfb739ae8..92f3cbd01e 100644 --- a/src/msg/msg_vm.cpp +++ b/src/msg/msg_vm.cpp @@ -6,24 +6,18 @@ /* 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 + * 2. use parallel tasks to simulate CPU overhead and remove the experimental code generating micro computation tasks */ - - #include "msg_private.h" #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"); +XBT_LOG_NEW_DEFAULT_SUBCATEGORY(msg_vm, msg, "Cloud-oriented parts of the MSG API"); /* **** ******** GENERAL ********* **** */ - /** \ingroup m_vm_management * \brief Returns the value of a given vm property * @@ -31,7 +25,6 @@ XBT_LOG_NEW_DEFAULT_SUBCATEGORY(msg_vm, msg, * \param name a property name * \return value of a property (or NULL if property not set) */ - const char *MSG_vm_get_property_value(msg_vm_t vm, const char *name) { return MSG_host_get_property_value(vm, name); @@ -69,10 +62,8 @@ void MSG_vm_set_property_value(msg_vm_t vm, const char *name, void *value, void_ * \param name the name of a vm. * \return the corresponding vm * - * Please note that a VM is a specific host. Hence, you should give a different name - * for each VM/PM. + * Please note that a VM is a specific host. Hence, you should give a different name for each VM/PM. */ - msg_vm_t MSG_vm_get_by_name(const char *name) { return MSG_host_by_name(name); @@ -82,15 +73,13 @@ msg_vm_t MSG_vm_get_by_name(const char *name) * * \brief Return the name of the #msg_host_t. * - * This functions checks whether \a host is a valid pointer or not and return - its name. + * This functions checks whether \a host is a valid pointer or not and return its name. */ const char *MSG_vm_get_name(msg_vm_t vm) { return MSG_host_get_name(vm); } - /* **** Check state of a VM **** */ static inline int __MSG_vm_is_state(msg_vm_t vm, e_surf_vm_state_t state) { @@ -154,13 +143,7 @@ int MSG_vm_is_restoring(msg_vm_t vm) return __MSG_vm_is_state(vm, SURF_VM_STATE_RESTORING); } - - -/* ------------------------------------------------------------------------- */ -/* ------------------------------------------------------------------------- */ - /* **** ******** MSG vm actions ********* **** */ - /** @brief Create a new VM with specified parameters. * @ingroup msg_VMs* * @param pm Physical machine that will host the VM @@ -172,16 +155,13 @@ int MSG_vm_is_restoring(msg_vm_t vm) * @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 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 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(pm); + double host_speed = MSG_host_get_speed(pm); double update_speed = ((double)dp_intensity/100) * mig_netspeed; msg_vm_t vm = MSG_vm_create_core(pm, name); @@ -196,13 +176,13 @@ msg_vm_t MSG_vm_create(msg_host_t pm, const char *name, 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); - vm->set_parameters(¶ms); + //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); + vm->setParameters(¶ms); return vm; } - /** @brief Create a new VM object. The VM is not yet started. The resource of the VM is allocated upon MSG_vm_start(). * @ingroup msg_VMs* * @@ -212,8 +192,7 @@ msg_vm_t MSG_vm_create_core(msg_host_t ind_pm, const char *name) { /* make sure the VM of the same name does not exit */ { - simgrid::s4u::Host* ind_host_tmp = - (simgrid::s4u::Host*) xbt_dict_get_or_null(host_list, name); + simgrid::s4u::Host* ind_host_tmp = (simgrid::s4u::Host*) xbt_dict_get_or_null(host_list, name); if (ind_host_tmp != nullptr && sg_host_simix(ind_host_tmp) != nullptr) { XBT_ERROR("host %s already exits", name); return nullptr; @@ -221,7 +200,6 @@ msg_vm_t MSG_vm_create_core(msg_host_t ind_pm, const char *name) } /* Note: ind_vm and vm_workstation point to the same elm object. */ - /* Ask the SIMIX layer to create the surf vm resource */ sg_host_t ind_vm_workstation = (sg_host_t) simcall_vm_create(name, ind_pm); @@ -257,12 +235,10 @@ void MSG_vm_destroy(msg_vm_t vm) TRACE_msg_vm_end(vm); } - /** @brief Start a vm (i.e., boot the guest operating system) * @ingroup msg_VMs * * If the VM cannot be started, an exception is generated. - * */ void MSG_vm_start(msg_vm_t vm) { @@ -271,13 +247,11 @@ void MSG_vm_start(msg_vm_t vm) TRACE_msg_vm_start(vm); } - - /** @brief Immediately kills all processes within the given VM. Any memory that they allocated will be leaked. * @ingroup msg_VMs * - * FIXME: No extra delay occurs. If you want to simulate this too, you want to - * use a #MSG_process_sleep() or something. I'm not quite sure. + * FIXME: No extra delay occurs. If you want to simulate this too, you want to use a #MSG_process_sleep() or something. + * I'm not quite sure. */ void MSG_vm_shutdown(msg_vm_t vm) { @@ -287,11 +261,8 @@ void MSG_vm_shutdown(msg_vm_t vm) // TRACE_msg_vm_(vm); } - - -/* 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. */ +/* 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. */ static inline char *get_mig_mbox_src_dst(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) { const char *vm_name = sg_host_get_name(vm); @@ -337,7 +308,6 @@ static inline char *get_mig_task_name(msg_vm_t vm, msg_host_t src_pm, msg_host_t return bprintf("__task_mig_stage%d:%s(%s-%s)", stage, vm_name, src_pm_name, dst_pm_name); } - struct migration_session { msg_vm_t vm; msg_host_t src_pm; @@ -346,12 +316,10 @@ struct migration_session { /* 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. */ + /* 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"); @@ -360,7 +328,7 @@ static int migration_rx_fun(int argc, char *argv[]) struct migration_session *ms = (migration_session *) MSG_process_get_data(MSG_process_self()); s_vm_params_t params; - ms->vm->get_parameters(¶ms); + ms->vm->parameters(¶ms); int need_exit = 0; @@ -372,7 +340,7 @@ static int migration_rx_fun(int argc, char *argv[]) ret = MSG_task_recv(&task, ms->mbox); { if (ret != MSG_OK) { - // An error occured, clean the code and return + // An error occurred, clean the code and return // The owner did not change, hence the task should be only destroyed on the other side xbt_free(finalize_task_name); return 0; @@ -392,14 +360,15 @@ static int migration_rx_fun(int argc, char *argv[]) // Hence complete the migration // Copy the reference to the vm (if SRC crashes now, do_migration will free ms) - // This is clearly ugly but I (Adrien) need more time to do something cleaner (actually we should copy the whole ms structure at the begining and free it at the end of each function) + // This is clearly ugly but I (Adrien) need more time to do something cleaner (actually we should copy the whole ms + // structure at the beginning and free it at the end of each function) 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 = 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 +// 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); // @@ -414,19 +383,18 @@ static int migration_rx_fun(int argc, char *argv[]) /* install the affinity setting of the VM on the destination pm */ { - 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)); + 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 = sg_host_msg(vm); priv->is_migrating = 0; - XBT_DEBUG("VM(%s) moved from PM(%s) to PM(%s)", - sg_host_get_name(ms->vm), sg_host_get_name(ms->src_pm), - sg_host_get_name(ms->dst_pm)); + XBT_DEBUG("VM(%s) moved from PM(%s) to PM(%s)", sg_host_get_name(ms->vm), sg_host_get_name(ms->src_pm), + sg_host_get_name(ms->dst_pm)); TRACE_msg_vm_change_host(ms->vm, ms->src_pm, ms->dst_pm); } // Inform the SRC that the migration has been correctly performed @@ -436,22 +404,20 @@ static int migration_rx_fun(int argc, char *argv[]) msg_error_t ret = MSG_task_send(task, ms->mbox_ctl); // xbt_assert(ret == MSG_OK); 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 + // 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 // TODO What does it mean ? What should we do ? MSG_task_destroy(task); } else if(ret == MSG_TRANSFER_FAILURE){ // The SRC has crashed, this is not a problem has the VM has been correctly migrated on the DST node MSG_task_destroy(task); } - xbt_free(task_name); } - xbt_free(finalize_task_name); XBT_DEBUG("mig: rx_done"); - return 0; } @@ -519,15 +485,10 @@ static double lookup_computed_flop_counts(msg_vm_t vm, int stage_for_fancy_debug total += priv->dp_updated_by_deleted_tasks; XBT_DEBUG("mig-stage%d.%d: computed %f flop_counts (including %f by deleted tasks)", - stage_for_fancy_debug, - stage2_round_for_fancy_debug, - total, priv->dp_updated_by_deleted_tasks); - - + stage_for_fancy_debug, stage2_round_for_fancy_debug, total, priv->dp_updated_by_deleted_tasks); priv->dp_updated_by_deleted_tasks = 0; - return total; } @@ -564,9 +525,8 @@ void MSG_host_del_task(msg_host_t host, msg_task_t task) dirty_page_t dp = (dirty_page_t) 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 - * computation has been done until now, and keep the information for the - * lookup_() function that will called soon. */ + /* If we are in the middle of dirty page tracking, we record how much 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_flops_amount(task); double clock = MSG_get_clock(); @@ -584,11 +544,8 @@ void MSG_host_del_task(msg_host_t host, msg_task_t task) xbt_free(key); } - - - -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, char *mbox, + int stage, int stage2_round, double mig_speed, double timeout) { sg_size_t sent = 0; char *task_name = get_mig_task_name(vm, src_pm, dst_pm, stage); @@ -611,17 +568,16 @@ static sg_size_t send_migration_data(msg_vm_t vm, msg_host_t src_pm, msg_host_t } else if (ret == MSG_TIMEOUT) { 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); + XBT_VERB("timeout (%lf s) in sending_migration_data, remaining %llu bytes of %llu", timeout, remaining, size); } /* FIXME: why try-and-catch is used here? */ if(ret == MSG_HOST_FAILURE){ - //XBT_INFO("SRC host failed during migration of %s (stage %d)", sg_host_name(vm), stage); + //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); }else if(ret == MSG_TRANSFER_FAILURE){ - //XBT_INFO("DST host failed during migration of %s (stage %d)", sg_host_name(vm), stage); + //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); } @@ -631,9 +587,11 @@ static sg_size_t send_migration_data(msg_vm_t vm, msg_host_t src_pm, msg_host_t double actual_speed = size / duration; if (stage == 2) - XBT_DEBUG("mig-stage%d.%d: sent %llu duration %f actual_speed %f (target %f)", stage, stage2_round, size, duration, actual_speed, mig_speed); + XBT_DEBUG("mig-stage%d.%d: sent %llu duration %f actual_speed %f (target %f)", + stage, stage2_round, size, duration, actual_speed, mig_speed); else - XBT_DEBUG("mig-stage%d: sent %llu duration %f actual_speed %f (target %f)", stage, size, duration, actual_speed, mig_speed); + XBT_DEBUG("mig-stage%d: sent %llu duration %f actual_speed %f (target %f)", + stage, size, duration, actual_speed, mig_speed); return sent; } @@ -643,17 +601,17 @@ static sg_size_t get_updated_size(double computed, double dp_rate, double dp_cap double updated_size = computed * dp_rate; XBT_DEBUG("updated_size %f dp_rate %f", updated_size, dp_rate); if (updated_size > dp_cap) { - // XBT_INFO("mig-stage2.%d: %f bytes updated, but cap it with the working set size %f", stage2_round, updated_size, dp_cap); + // XBT_INFO("mig-stage2.%d: %f bytes updated, but cap it with the working set size %f", stage2_round, updated_size, + // dp_cap); updated_size = dp_cap; } return (sg_size_t) updated_size; } -static double send_stage1(struct migration_session *ms, - sg_size_t ramsize, double mig_speed, double dp_rate, double dp_cap) +static double send_stage1(struct migration_session *ms, sg_size_t ramsize, double mig_speed, double dp_rate, + double dp_cap) { - // 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; @@ -673,8 +631,6 @@ static double send_stage1(struct migration_session *ms, return computed_total; } - - static double get_threshold_value(double bandwidth, double max_downtime) { return max_downtime * bandwidth; @@ -685,11 +641,10 @@ static int migration_tx_fun(int argc, char *argv[]) XBT_DEBUG("mig: tx_start"); // Note that the ms structure has been allocated in do_migration and hence should be freed in the same function ;) - migration_session *ms = - (migration_session *) MSG_process_get_data(MSG_process_self()); + migration_session *ms = (migration_session *) MSG_process_get_data(MSG_process_self()); s_vm_params_t params; - ms->vm->get_parameters(¶ms); + ms->vm->parameters(¶ms); const sg_size_t ramsize = params.ramsize; const sg_size_t devsize = params.devsize; const int skip_stage1 = params.skip_stage1; @@ -724,22 +679,21 @@ static int migration_tx_fun(int argc, char *argv[]) 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 tranfer exceeds the timeout - * value. */ - XBT_INFO("Stage 1: Gonna send %llu", ramsize); + /* 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", 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_INFO("mig-stage1: timeout, force moving to stage 3"); + XBT_VERB("mig-stage1: timeout, force moving to stage 3"); skip_stage2 = 1; } else if (sent > ramsize) XBT_CRITICAL("bug"); } CATCH_ANONYMOUS { - //hostfailure (if you want to know whether this is the SRC or the DST please check directly in send_migration_data code) + //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; @@ -748,7 +702,7 @@ static int migration_tx_fun(int argc, char *argv[]) double clock_post_send = MSG_get_clock(); mig_timeout -= (clock_post_send - clock_prev_send); if (mig_timeout < 0) { - XBT_INFO("The duration of stage 1 exceeds the timeout value (%lf > %lf), skip stage 2", + 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); skip_stage2 = 1; } @@ -760,8 +714,7 @@ static int migration_tx_fun(int argc, char *argv[]) } - /* Stage2: send update pages iteratively until the size of remaining states - * becomes smaller than the threshold value. */ + /* Stage2: send update pages iteratively until the size of remaining states becomes smaller than threshold value. */ if (! skip_stage2) { int stage2_round = 0; @@ -769,8 +722,7 @@ static int migration_tx_fun(int argc, char *argv[]) sg_size_t updated_size = 0; if (stage2_round == 0) { - /* just after stage1, nothing has been updated. But, we have to send the - * data updated during stage1 */ + /* 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(ms->vm, 2, stage2_round); @@ -780,23 +732,21 @@ static int migration_tx_fun(int argc, char *argv[]) XBT_DEBUG("mig-stage 2:%d updated_size %llu computed_during_stage1 %f dp_rate %f dp_cap %f", stage2_round, updated_size, computed_during_stage1, dp_rate, dp_cap); - - /* Check whether the remaining size is below the threshold value. If so, - * move to stage 3. */ + /* 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, (remaining_size < threshold) ? "<" : ">", threshold); if (remaining_size < threshold) break; - 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); + 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) + //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; @@ -816,7 +766,7 @@ static int migration_tx_fun(int argc, char *argv[]) } else if (sent < updated_size) { /* When timeout happens, we move to stage 3. The size of memory pages * updated before timeout must be added to the remaining size. */ - XBT_INFO("mig-stage2.%d: timeout, force moving to stage 3. sent %llu / %llu, eta %lf", + XBT_VERB("mig-stage2.%d: timeout, force moving to stage 3. sent %llu / %llu, eta %lf", stage2_round, sent, updated_size, (clock_post_send - clock_prev_send)); remaining_size -= sent; @@ -824,7 +774,6 @@ static int migration_tx_fun(int argc, char *argv[]) updated_size = get_updated_size(computed, dp_rate, dp_cap); remaining_size += updated_size; break; - } else XBT_CRITICAL("bug"); } @@ -839,20 +788,19 @@ static int migration_tx_fun(int argc, char *argv[]) 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) + //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) simcall_vm_resume(ms->vm); return 0; } - // 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. + // 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_DEBUG("mig: tx_done"); return 0; } - - static int do_migration(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) { struct migration_session *ms = xbt_new(struct migration_session, 1); @@ -861,7 +809,6 @@ static int do_migration(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_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); @@ -884,7 +831,7 @@ static int do_migration(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) } #endif - /* wait until the migration have finished or on error has occured */ + /* 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 = NULL; @@ -895,21 +842,22 @@ static int do_migration(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) xbt_free(ms->mbox_ctl); xbt_free(ms->mbox); xbt_free(ms); - + //xbt_assert(ret == MSG_OK); 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 - 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 + // 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_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; } - else if((ret == MSG_TRANSFER_FAILURE) || (ret == MSG_TIMEOUT)){ // MSG_TIMEOUT here means that MSG_host_is_avail() returned false. - XBT_INFO("DST crashes, throw an exception (m-control)"); + 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; } - 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); @@ -918,47 +866,38 @@ 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 * - * 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. + * 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 + * 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 migraion 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? + * 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. + * We update the information of the given VM to place it to the destination physical host. * * The second one would be easier. - * */ msg_host_t old_pm = (msg_host_t) 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_get_name(vm)); @@ -969,7 +908,6 @@ void MSG_vm_migrate(msg_vm_t vm, msg_host_t new_pm) priv->is_migrating = 1; { - int ret = do_migration(vm, old_pm, new_pm); if (ret == -1){ priv->is_migrating = 0; @@ -981,13 +919,12 @@ void MSG_vm_migrate(msg_vm_t vm, msg_host_t new_pm) } } - // 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. + // 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); //TRACE_msg_vm_change_host(vm, old_pm, new_pm); } - /** @brief Immediately suspend the execution of all processes within the given VM. * @ingroup msg_VMs * @@ -1008,7 +945,6 @@ void MSG_vm_suspend(msg_vm_t vm) TRACE_msg_vm_suspend(vm); } - /** @brief Resume the execution of the VM. All processes on the VM run again. * @ingroup msg_VMs * @@ -1028,9 +964,8 @@ void MSG_vm_resume(msg_vm_t vm) * 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 - * of VM save to you. + * 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 of VM save to you. */ void MSG_vm_save(msg_vm_t vm) { @@ -1044,9 +979,8 @@ void MSG_vm_save(msg_vm_t vm) /** @brief Restore the execution of the VM. All processes on the VM run again. * @ingroup msg_VMs * - * FIXME: No restore cost occurs. If you want to simulate this too, you want to - * use a \ref MSG_file_read() before or after, depending on the exact semantic - * of VM restore to you. + * FIXME: No restore cost occurs. If you want to simulate this too, you want to use a \ref MSG_file_read() before or + * after, depending on the exact semantic of VM restore to you. */ void MSG_vm_restore(msg_vm_t vm) { @@ -1055,7 +989,6 @@ void MSG_vm_restore(msg_vm_t vm) TRACE_msg_vm_restore(vm); } - /** @brief Get the physical host of a given VM. * @ingroup msg_VMs */ @@ -1064,12 +997,10 @@ msg_host_t MSG_vm_get_pm(msg_vm_t vm) return (msg_host_t) simcall_vm_get_pm(vm); } - /** @brief Set a CPU bound for a given VM. * @ingroup msg_VMs * - * 1. - * Note that in some cases MSG_task_set_bound() may not intuitively work for VMs. + * 1. Note that in some cases MSG_task_set_bound() may not intuitively work for VMs. * * For example, * On PM0, there are Task1 and VM0. @@ -1080,31 +1011,23 @@ msg_host_t MSG_vm_get_pm(msg_vm_t vm) * 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 - * of the dummy CPU action is unlimited. - * - * 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 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. + * 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 of the dummy CPU action is unlimited. * - * The current solution is to use MSG_vm_set_bound(), which allows us to - * directly set the bound of the dummy CPU action. + * 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 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. * + * The current solution is to use MSG_vm_set_bound(), which allows us to directly set the bound of the dummy CPU action. * - * 2. - * Note that bound == 0 means no bound (i.e., unlimited). But, if a host has - * multiple CPU cores, the CPU share of a computation task (or a VM) never - * exceeds the capacity of a CPU core. + * 2. Note that bound == 0 means no bound (i.e., unlimited). But, if a host has multiple CPU cores, the CPU share of a + * computation task (or a VM) never exceeds the capacity of a CPU core. */ void MSG_vm_set_bound(msg_vm_t vm, double bound) { simcall_vm_set_bound(vm, bound); } - /** @brief Set the CPU affinity of a given VM. * @ingroup msg_VMs *