X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/6b76ba4acd263f00d310287d7f891abfcda1c2f9..b3b356352e87ae00a20f737c48e19b0c8413455a:/src/msg/msg_vm.c diff --git a/src/msg/msg_vm.c b/src/msg/msg_vm.c index 5665a8450f..a1df02d2d2 100644 --- a/src/msg/msg_vm.c +++ b/src/msg/msg_vm.c @@ -1,4 +1,4 @@ -/* 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 @@ -16,6 +16,7 @@ #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"); @@ -75,7 +76,7 @@ void MSG_vm_set_property_value(msg_vm_t vm, const char *name, void *value, void_ 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 @@ -118,7 +119,7 @@ int MSG_vm_is_running(msg_vm_t vm) */ 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; } @@ -177,7 +178,7 @@ msg_vm_t MSG_vm_create(msg_host_t ind_pm, const char *name, 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; + s_vm_params_t params; memset(¶ms, 0, sizeof(params)); params.ramsize = (sg_size_t)ramsize * 1024 * 1024; //params.overcommit = 0; @@ -205,7 +206,7 @@ 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 */ { 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; } @@ -221,9 +222,7 @@ msg_vm_t MSG_vm_create_core(msg_host_t ind_pm, const char *name) XBT_DEBUG("A new VM (%s) has been created", name); - #ifdef HAVE_TRACING TRACE_msg_vm_create(name, ind_pm); - #endif return ind_vm; } @@ -250,9 +249,7 @@ void MSG_vm_destroy(msg_vm_t vm) __MSG_host_destroy(vm); - #ifdef HAVE_TRACING TRACE_msg_vm_end(vm); - #endif } @@ -266,9 +263,7 @@ void MSG_vm_start(msg_vm_t vm) { simcall_vm_start(vm); - #ifdef HAVE_TRACING TRACE_msg_vm_start(vm); - #endif } @@ -284,9 +279,7 @@ void MSG_vm_shutdown(msg_vm_t vm) /* msg_vm_t equals to msg_host_t */ simcall_vm_shutdown(vm); - // #ifdef HAVE_TRACING // TRACE_msg_vm_(vm); - // #endif } @@ -361,7 +354,7 @@ static int migration_rx_fun(int argc, char *argv[]) // 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; @@ -398,18 +391,20 @@ static int migration_rx_fun(int argc, char *argv[]) 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 */ { @@ -422,13 +417,10 @@ static int migration_rx_fun(int argc, char *argv[]) { // 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 { @@ -458,13 +450,13 @@ static int migration_rx_fun(int argc, char *argv[]) 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; @@ -474,7 +466,7 @@ static void reset_dirty_pages(msg_vm_t vm) 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); @@ -482,7 +474,7 @@ static void start_dirty_page_tracking(msg_vm_t 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; } @@ -518,14 +510,14 @@ static double get_computed(char *key, msg_vm_t vm, dirty_page_t dp, double remai 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(); @@ -555,8 +547,8 @@ static double lookup_computed_flop_counts(msg_vm_t vm, int stage_for_fancy_debug // 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); @@ -577,7 +569,7 @@ void MSG_host_add_task(msg_host_t host, msg_task_t task) 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); @@ -588,7 +580,7 @@ void MSG_host_del_task(msg_host_t host, msg_task_t 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); @@ -707,7 +699,7 @@ static int migration_tx_fun(int argc, char *argv[]) // 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; @@ -746,7 +738,7 @@ static int migration_tx_fun(int argc, char *argv[]) /* 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 %zd", ramsize); + 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); @@ -814,7 +806,7 @@ static int migration_tx_fun(int argc, char *argv[]) sg_size_t sent = 0; double clock_prev_send = MSG_get_clock(); TRY { - XBT_INFO("Stage 2, gonna send %d", updated_size); + 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) @@ -858,7 +850,7 @@ stage3: stop_dirty_page_tracking(ms->vm); TRY { - XBT_INFO("Stage 3: Gonna send %d", remaining_size); + 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) @@ -868,7 +860,7 @@ stage3: } // 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; } @@ -888,7 +880,7 @@ static int do_migration(msg_vm_t vm, msg_host_t src_pm, msg_host_t 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_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 @@ -896,13 +888,13 @@ static int do_migration(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) 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 @@ -912,7 +904,7 @@ static int do_migration(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) 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); @@ -975,13 +967,19 @@ void MSG_vm_migrate(msg_vm_t vm, msg_host_t new_pm) 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)); 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); + msg_host_priv_t priv = sg_host_msg(vm); priv->is_migrating = 1; { @@ -1000,9 +998,7 @@ 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. //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 } @@ -1023,9 +1019,7 @@ void MSG_vm_suspend(msg_vm_t vm) XBT_DEBUG("vm_suspend done"); - #ifdef HAVE_TRACING TRACE_msg_vm_suspend(vm); - #endif } @@ -1038,9 +1032,7 @@ void MSG_vm_resume(msg_vm_t vm) { simcall_vm_resume(vm); - #ifdef HAVE_TRACING TRACE_msg_vm_resume(vm); - #endif } @@ -1060,9 +1052,7 @@ void MSG_vm_save(msg_vm_t 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); - #endif } /** @brief Restore the execution of the VM. All processes on the VM run again. @@ -1076,9 +1066,7 @@ void MSG_vm_restore(msg_vm_t vm) { simcall_vm_restore(vm); - #ifdef HAVE_TRACING TRACE_msg_vm_restore(vm); - #endif } @@ -1139,7 +1127,7 @@ void MSG_vm_set_bound(msg_vm_t vm, double 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));