X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/7f4e5b2ed2874cf5ab13b7fbe0cebff788cb5019..f7623851a023484d8ba1d5c26134ee7850bac134:/src/msg/msg_vm.cpp diff --git a/src/msg/msg_vm.cpp b/src/msg/msg_vm.cpp index e2bd5085cb..8d59a21358 100644 --- a/src/msg/msg_vm.cpp +++ b/src/msg/msg_vm.cpp @@ -11,21 +11,25 @@ #include +#include "src/instr/instr_private.h" +#include "src/msg/msg_private.h" #include "src/plugins/vm/VirtualMachineImpl.hpp" #include "src/plugins/vm/VmHostExt.hpp" -#include "src/simix/ActorImpl.hpp" -#include -#include -#include "msg_private.h" -#include "xbt/sysdep.h" -#include "xbt/log.h" #include "simgrid/host.h" +#include "simgrid/simix.hpp" -#include "src/simix/smx_host_private.h" /* don't ask me why the VM functions are in there (FIXME:KILLME) */ +SG_BEGIN_DECL() -XBT_LOG_NEW_DEFAULT_SUBCATEGORY(msg_vm, msg, "Cloud-oriented parts of the MSG API"); +struct dirty_page { + double prev_clock; + double prev_remaining; + msg_task_t task; +}; +typedef struct dirty_page s_dirty_page; +typedef struct dirty_page* dirty_page_t; +XBT_LOG_NEW_DEFAULT_SUBCATEGORY(msg_vm, msg, "Cloud-oriented parts of the MSG API"); /* **** ******** GENERAL ********* **** */ @@ -43,7 +47,7 @@ void MSG_vm_set_params(msg_vm_t vm, vm_params_t params) /** \ingroup m_vm_management * \brief Get the parameters of a given host * - * \param host a host + * \param vm the vm you are interested into * \param params a prameter object */ void MSG_vm_get_params(msg_vm_t vm, vm_params_t params) @@ -54,7 +58,8 @@ void MSG_vm_get_params(msg_vm_t vm, vm_params_t params) /* **** Check state of a VM **** */ static inline int __MSG_vm_is_state(msg_vm_t vm, e_surf_vm_state_t state) { - return static_cast(vm)->pimpl_vm_->getState() == state; + simgrid::s4u::VirtualMachine* castedVm = static_cast(vm); + return castedVm->pimpl_vm_ != nullptr && castedVm->pimpl_vm_->getState() == state; } /** @brief Returns whether the given VM has just created, not running. @@ -94,37 +99,38 @@ int MSG_vm_is_suspended(msg_vm_t vm) * @ingroup msg_VMs* * @param pm Physical machine that will host the VM * @param name Must be unique + * @param coreAmount Must be >= 1 * @param ramsize [TODO] * @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 ramsize, int mig_netspeed, int dp_intensity) +msg_vm_t MSG_vm_create(msg_host_t pm, const char* name, int coreAmount, int ramsize, int mig_netspeed, int dp_intensity) { simgrid::vm::VmHostExt::ensureVmExtInstalled(); /* For the moment, intensity_rate is the percentage against the migration bandwidth */ double host_speed = MSG_host_get_speed(pm); - double update_speed = ((double)dp_intensity/100) * mig_netspeed; + double update_speed = (static_cast(dp_intensity)/100) * mig_netspeed; - msg_vm_t vm = MSG_vm_create_core(pm, name); + msg_vm_t vm = MSG_vm_create_multicore(pm, name, coreAmount); s_vm_params_t params; memset(¶ms, 0, sizeof(params)); - params.ramsize = (sg_size_t)ramsize * 1024 * 1024; + params.ramsize = static_cast(ramsize) * 1024 * 1024; 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 + params.mig_speed = static_cast(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); + XBT_DEBUG("dp rate %f migspeed : %f intensity mem : %d, updatespeed %f, hostspeed %f", params.dp_rate, + params.mig_speed, dp_intensity, update_speed, host_speed); static_cast(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(). +/** @brief Create a new VM object with the default parameters * @ingroup msg_VMs* * * A VM is treated as a host. The name of the VM must be unique among all hosts. @@ -134,7 +140,19 @@ msg_vm_t MSG_vm_create_core(msg_host_t pm, const char* name) xbt_assert(sg_host_by_name(name) == nullptr, "Cannot create a VM named %s: this name is already used by an host or a VM", name); - return new simgrid::s4u::VirtualMachine(name, pm); + return new simgrid::s4u::VirtualMachine(name, pm, 1); +} +/** @brief Create a new VM object with the default parameters, but with a specified amount of cores + * @ingroup msg_VMs* + * + * A VM is treated as a host. The name of the VM must be unique among all hosts. + */ +msg_vm_t MSG_vm_create_multicore(msg_host_t pm, const char* name, int coreAmount) +{ + xbt_assert(sg_host_by_name(name) == nullptr, + "Cannot create a VM named %s: this name is already used by an host or a VM", name); + + return new simgrid::s4u::VirtualMachine(name, pm, coreAmount); } /** @brief Destroy a VM. Destroy the VM object from the simulation. @@ -149,8 +167,6 @@ void MSG_vm_destroy(msg_vm_t vm) if (MSG_vm_is_running(vm)) MSG_vm_shutdown(vm); - xbt_assert(MSG_vm_is_created(vm), "shutdown the given VM before destroying it"); - /* Then, destroy the VM object */ simgrid::simix::kernelImmediate([vm]() { vm->destroy(); @@ -166,7 +182,7 @@ void MSG_vm_destroy(msg_vm_t vm) /** @brief Start a vm (i.e., boot the guest operating system) * @ingroup msg_VMs * - * If the VM cannot be started (because of memory overprovisionning), an exception is generated. + * If the VM cannot be started (because of memory over-provisioning), an exception is generated. */ void MSG_vm_start(msg_vm_t vm) { @@ -182,7 +198,7 @@ void MSG_vm_start(msg_vm_t vm) int pm_overcommit = pm->extension()->overcommit; long vm_ramsize = typedVM->getRamsize(); - if (pm_ramsize && !pm_overcommit) { /* Only verify that we don't overcommit on need */ + if (pm_ramsize && not pm_overcommit) { /* Only verify that we don't overcommit on need */ /* Retrieve the memory occupied by the VMs on that host. Yep, we have to traverse all VMs of all hosts for that */ long total_ramsize_of_vms = 0; for (simgrid::s4u::VirtualMachine* ws_vm : simgrid::vm::VirtualMachineImpl::allVms_) @@ -203,7 +219,7 @@ void MSG_vm_start(msg_vm_t vm) container_t vm_container = PJ_container_get(vm->cname()); type_t type = PJ_type_get("MSG_VM_STATE", vm_container->type); val_t value = PJ_value_get_or_new("start", "0 0 1", type); // start is blue - new_pajePushState(MSG_get_clock(), vm_container, type, value); + new PushStateEvent(MSG_get_clock(), vm_container, type, value); } } @@ -263,7 +279,7 @@ static int migration_rx_fun(int argc, char *argv[]) bool received_finalize = false; char *finalize_task_name = get_mig_task_name(ms->vm, ms->src_pm, ms->dst_pm, 3); - while (!received_finalize) { + while (not received_finalize) { msg_task_t task = nullptr; int ret = MSG_task_recv(&task, ms->mbox); @@ -288,11 +304,10 @@ static int migration_rx_fun(int argc, char *argv[]) // 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) simgrid::s4u::VirtualMachine* vm = static_cast(ms->vm); - msg_host_t src_pm = ms->src_pm; msg_host_t dst_pm = ms->dst_pm; // Make sure that we cannot get interrupted between the migrate and the resume to not end in an inconsistent state - simgrid::simix::kernelImmediate([vm, src_pm, dst_pm]() { + simgrid::simix::kernelImmediate([vm, dst_pm]() { /* Update the vm location */ /* precopy migration makes the VM temporally paused */ xbt_assert(vm->pimpl_vm_->getState() == SURF_VM_STATE_SUSPENDED); @@ -302,52 +317,51 @@ static int migration_rx_fun(int argc, char *argv[]) vm->pimpl_vm_->resume(); }); - { - // Now the VM is running on the new host (the migration is completed) (even if the SRC crash) - vm->pimpl_vm_->isMigrating = false; - XBT_DEBUG("VM(%s) moved from PM(%s) to PM(%s)", ms->vm->cname(), ms->src_pm->cname(), ms->dst_pm->cname()); - - if (TRACE_msg_vm_is_enabled()) { - static long long int counter = 0; - char key[INSTR_DEFAULT_STR_SIZE]; - snprintf(key, INSTR_DEFAULT_STR_SIZE, "%lld", counter++); - - // start link - container_t msg = PJ_container_get(vm->cname()); - type_t type = PJ_type_get("MSG_VM_LINK", PJ_type_get_root()); - new_pajeStartLink(MSG_get_clock(), PJ_container_get_root(), type, msg, "M", key); - - // destroy existing container of this vm - container_t existing_container = PJ_container_get(vm->cname()); - PJ_container_remove_from_parent(existing_container); - PJ_container_free(existing_container); - - // create new container on the new_host location - PJ_container_new(vm->cname(), INSTR_MSG_VM, PJ_container_get(ms->dst_pm->cname())); - - // end link - msg = PJ_container_get(vm->cname()); - type = PJ_type_get("MSG_VM_LINK", PJ_type_get_root()); - new_pajeEndLink(MSG_get_clock(), PJ_container_get_root(), type, msg, "M", key); - } + + // Now the VM is running on the new host (the migration is completed) (even if the SRC crash) + vm->pimpl_vm_->isMigrating = false; + XBT_DEBUG("VM(%s) moved from PM(%s) to PM(%s)", ms->vm->cname(), ms->src_pm->cname(), ms->dst_pm->cname()); + + if (TRACE_msg_vm_is_enabled()) { + static long long int counter = 0; + char key[INSTR_DEFAULT_STR_SIZE]; + snprintf(key, INSTR_DEFAULT_STR_SIZE, "%lld", counter); + counter++; + + // start link + container_t msg = PJ_container_get(vm->cname()); + type_t type = PJ_type_get("MSG_VM_LINK", PJ_type_get_root()); + new StartLinkEvent(MSG_get_clock(), PJ_container_get_root(), type, msg, "M", key); + + // destroy existing container of this vm + container_t existing_container = PJ_container_get(vm->cname()); + PJ_container_remove_from_parent(existing_container); + PJ_container_free(existing_container); + + // create new container on the new_host location + PJ_container_new(vm->cname(), INSTR_MSG_VM, PJ_container_get(ms->dst_pm->cname())); + + // end link + msg = PJ_container_get(vm->cname()); + type = PJ_type_get("MSG_VM_LINK", PJ_type_get_root()); + new EndLinkEvent(MSG_get_clock(), PJ_container_get_root(), type, msg, "M", key); } + // Inform the SRC that the migration has been correctly performed - { - 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, nullptr); - 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 - // 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); + 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, nullptr); + 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 + // 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_DEBUG("mig: rx_done"); return 0; @@ -358,7 +372,7 @@ static void start_dirty_page_tracking(msg_vm_t vm) simgrid::vm::VirtualMachineImpl* pimpl = static_cast(vm)->pimpl_vm_; pimpl->dp_enabled = 1; - if (!pimpl->dp_objs) + if (not pimpl->dp_objs) return; char *key = nullptr; @@ -438,7 +452,7 @@ void MSG_host_add_task(msg_host_t host, msg_task_t task) dp->prev_clock = MSG_get_clock(); dp->prev_remaining = remaining; } - if (!pimpl->dp_objs) + if (not pimpl->dp_objs) pimpl->dp_objs = xbt_dict_new_homogeneous(nullptr); xbt_assert(xbt_dict_get_or_null(pimpl->dp_objs, key) == nullptr); xbt_dict_set(pimpl->dp_objs, key, dp, nullptr); @@ -481,7 +495,7 @@ static sg_size_t send_migration_data(msg_vm_t vm, msg_host_t src_pm, msg_host_t { 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, (double)size, nullptr); + msg_task_t task = MSG_task_create(task_name, 0, static_cast(size), nullptr); /* TODO: clean up */ @@ -498,18 +512,18 @@ static sg_size_t send_migration_data(msg_vm_t vm, msg_host_t src_pm, msg_host_t if (ret == MSG_OK) { sent = size; } else if (ret == MSG_TIMEOUT) { - sg_size_t remaining = (sg_size_t)MSG_task_get_remaining_communication(task); + sg_size_t remaining = static_cast(MSG_task_get_remaining_communication(task)); sent = size - remaining; 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_DEBUG("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)", vm->cname(), stage); MSG_task_destroy(task); THROWF(host_error, 0, "SRC host failed during migration of %s (stage %d)", vm->cname(), stage); }else if(ret == MSG_TRANSFER_FAILURE){ - //XBT_DEBUG("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)", vm->cname(), stage); MSG_task_destroy(task); THROWF(host_error, 0, "DST host failed during migration of %s (stage %d)", vm->cname(), stage); } @@ -538,7 +552,7 @@ static sg_size_t get_updated_size(double computed, double dp_rate, double dp_cap updated_size = dp_cap; } - return (sg_size_t) updated_size; + return static_cast(updated_size); } static int migration_tx_fun(int argc, char *argv[]) @@ -546,7 +560,7 @@ 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 = static_cast(MSG_process_get_data(MSG_process_self())); s_vm_params_t params; static_cast(ms->vm)->parameters(¶ms); @@ -561,7 +575,7 @@ static int migration_tx_fun(int argc, char *argv[]) double mig_timeout = 10000000.0; - double remaining_size = (double) (ramsize + devsize); + double remaining_size = static_cast(ramsize + devsize); double threshold = 0.0; /* check parameters */ @@ -578,7 +592,7 @@ static int migration_tx_fun(int argc, char *argv[]) start_dirty_page_tracking(ms->vm); double computed_during_stage1 = 0; - if (!skip_stage1) { + if (not skip_stage1) { double clock_prev_send = MSG_get_clock(); try { @@ -618,7 +632,7 @@ static int migration_tx_fun(int argc, char *argv[]) /* Stage2: send update pages iteratively until the size of remaining states becomes smaller than threshold value. */ - if (!skip_stage2) { + if (not skip_stage2) { int stage2_round = 0; for (;;) { @@ -693,7 +707,8 @@ static int migration_tx_fun(int argc, char *argv[]) try { 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); + send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, static_cast(remaining_size), ms->mbox, 3, 0, + mig_speed, -1); } catch(xbt_ex& e) { //hostfailure (if you want to know whether this is the SRC or the DST check directly in send_migration_data code) @@ -738,7 +753,7 @@ void MSG_vm_migrate(msg_vm_t vm, msg_host_t dst_pm) THROWF(vm_error, 0, "Cannot migrate VM '%s' from host '%s', which is offline.", vm->cname(), src_pm->cname()); if (dst_pm->isOff()) THROWF(vm_error, 0, "Cannot migrate VM '%s' to host '%s', which is offline.", vm->cname(), dst_pm->cname()); - if (!MSG_vm_is_running(vm)) + if (not MSG_vm_is_running(vm)) THROWF(vm_error, 0, "Cannot migrate VM '%s' that is not running yet.", vm->cname()); if (typedVm->isMigrating()) THROWF(vm_error, 0, "Cannot migrate VM '%s' that is already migrating.", vm->cname()); @@ -771,9 +786,7 @@ void MSG_vm_migrate(msg_vm_t vm, msg_host_t dst_pm) /* 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 = nullptr; - msg_error_t ret = MSG_TIMEOUT; - while (ret == MSG_TIMEOUT && dst_pm->isOn()) // The rx will tell me when he gots the VM - ret = MSG_task_receive_with_timeout(&task, ms->mbox_ctl, 4); + msg_error_t ret = MSG_task_receive(&task, ms->mbox_ctl); pimpl->isMigrating = false; @@ -822,7 +835,7 @@ void MSG_vm_suspend(msg_vm_t vm) container_t vm_container = PJ_container_get(vm->cname()); type_t type = PJ_type_get("MSG_VM_STATE", vm_container->type); val_t value = PJ_value_get_or_new("suspend", "1 0 0", type); // suspend is red - new_pajePushState(MSG_get_clock(), vm_container, type, value); + new PushStateEvent(MSG_get_clock(), vm_container, type, value); } } @@ -838,7 +851,7 @@ void MSG_vm_resume(msg_vm_t vm) if (TRACE_msg_vm_is_enabled()) { container_t vm_container = PJ_container_get(vm->cname()); type_t type = PJ_type_get("MSG_VM_STATE", vm_container->type); - new_pajePopState(MSG_get_clock(), vm_container, type); + new PopStateEvent(MSG_get_clock(), vm_container, type); } } @@ -881,3 +894,5 @@ void MSG_vm_set_bound(msg_vm_t vm, double bound) simgrid::simix::kernelImmediate( [vm, bound]() { static_cast(vm)->pimpl_vm_->setBound(bound); }); } + +SG_END_DECL()