A dedicated host is a single-tenant physical server that runs on the Alibaba Cloud virtualization stack. This topic describes the features of Dedicated Host (DDH).
Physical isolation
DDH provides the physical isolation feature:
The CPU, memory, and network resources of a dedicated host are physically isolated from other hosts. This guarantees a high level of security. As the only tenant of a host, the physical resources of the host are not shared with other tenants.
You can obtain the physical attributes of the host, including the number of physical CPU cores, the memory size, the number of vCPUs, and a unique machine code. You can also view the resource usage in real time.
Automatic deployment
After you enable automatic deployment for a dedicated host, the dedicated host resides in the resource pool of your Alibaba Cloud account for automatic deployment. When you create an ECS instance, you do not need to specify a dedicated host. The system selects a dedicated host from the resource pool to deploy the instance. For more information, see Enable automatic deployment. Automatic deployment allows you to flexibly deploy your business.
The following table describes the use scenarios of automatic deployment.
Scenario | Operation | Result | References |
Create an ECS instance | Specify a dedicated host | The ECS instance is deployed on the dedicated host. |
|
Select Allow Automatic Deployment | The ECS instance is automatically deployed on a dedicated host. | ||
Migrate an ECS instance from a shared host to a dedicated host | Specify a dedicated host | The ECS instance is deployed on the dedicated host. |
|
Do not specify a dedicated host | The ECS instance is automatically deployed on a dedicated host. |
| |
Migrate an ECS instance between two dedicated hosts | Specify a dedicated host | The ECS instance is deployed on the dedicated host. |
|
Do not specify a dedicated host | The ECS instance is automatically deployed on a dedicated host. Note If the dedicated host where the ECS instance resides allows automatic deployment and has the most available resources, the instance is still deployed on the dedicated host. |
|
CPU overcommit ratio
You can configure the CPU overcommit ratio for g6s, c6s, and r6s dedicated hosts. If you change the CPU overcommit ratio of a dedicated host, the number of its available vCPUs is changed. The CPU overcommit ratio of a dedicated host is calculated based on the following formula: Number of available vCPUs of the dedicated host = Number of CPU cores × 2 × CPU overcommit ratio. In scenarios where lower CPU stability is required or the CPU load is not heavy, you can increase the CPU overcommit ratio of a dedicated host. This way, you can increase the number of available vCPUs on the dedicated host. You can also deploy more ECS instances of the same specifications on the dedicated host and reduce the unit deployment cost. For more information, see Configure the CPU overcommit ratio.
Automatic instance migration upon hardware failures
A dedicated host is based on a physical server. The dedicated host may shut down if a failure occurs on the server hardware. To minimize the impact of hardware failures on your business, Alibaba Cloud supports automatic instance migration upon hardware failures. If a hardware failure occurs after you enable this feature for a dedicated host, the ECS instances on the dedicated host are migrated to another host.
The following table describes the use scenarios of automatic deployment.
Scenario | Operation |
Enable automatic instance migration for a dedicated host. |
|
A dedicated host has potential risks. | Migrate the dedicated host to another dedicated host to protect your services. For more information, see Migrate a dedicated host with hidden failures. |
A dedicated host fails. |
|
You cannot perform manual migration of ECS instances on local SSD dedicated hosts such as Local SSD type i2 upon hardware failures. In addition, you cannot perform automatic migration of ECS instances on local SSD dedicated hosts. If local SSD dedicated hosts fail, you can submit a ticket to apply for manual migration. However, the data in the local disk is lost after manual migration.
After all the ECS instances on a failed dedicated host are migrated to another dedicated host, the host ID and the metadata of the ECS instances remain unchanged. The metadata includes the ID, private IP address, and public IP address of each ECS instance. However, the machine ID of the dedicated host is changed.
Host association
Host association is suitable for ECS instances that have enabled the economical mode. Even if the instance enters the Stopped state, it is deployed on the original physical server. The feature also helps you maintain the same deployment method for ECS instances.
The following table describes the use scenarios of host association.
Scenario | Operation | Result |
Stop and restart an ECS instance that has enabled the economical mode | Enable the host association feature. For more information, see Associate an ECS instance with a dedicated host. | The ECS instance is still deployed on the original dedicated host. If the available resources of the dedicated host are insufficient, the instance fails to be restarted. |
Disable the host association feature. For more information, see Associate an ECS instance with a dedicated host. | ECS first attempts to deploy the ECS instance on the original dedicated host. If the available resources of the dedicated host are insufficient, ECS deploys the instance on another dedicated host. |
ECS instance migration
You can migrate an ECS instance between two dedicated hosts or between a shared host and a dedicated host.
Differences between ECS instances on a dedicated host and those on a shared host
The features of the ECS instances deployed on a dedicated host and the ECS instances deployed on a shared host are different. The following table lists the differences.
Item | ECS instances on a dedicated host | ECS instances on a shared host |
Network Type | Virtual private clouds (VPCs) are supported. | VPCs and the classic network are supported. For more information, see Network types. |
Billable item | See Billing overview. | |
Billing method | Subscription and pay-as-you-go are supported. | Subscription, pay-as-you-go, preemptible instances, and reserved instances are supported. |
Renewal | When you specify the renewal period, the expiration date of the instance cannot be later than that of the subscription dedicated host. For more information, see Limits. | You can specify the renewal period based on your business requirements. |
Economical mode | If an instance is stopped after the economical mode is enabled, the resources of the dedicated host are released. These resources include the vCPUs and memory. Resources are reallocated to the dedicated host after the instance is restarted. For more information, see Dedicated host billing. | See Economical mode. |
Switch the billing method from pay-as-you-go to subscription | Before you switch the billing method of an instance, the instance must meet the prerequisites. The feature is applicable only to the instances on a pay-as-you-go dedicated host. After the billing method of an ECS instance is switched to subscription, the expiration date of the instance cannot be later than that of the dedicated host. | Before you switch the billing method of an instance, the instance must meet the prerequisites. For more information, see Change the billing method of an ECS instance from pay-as-you-go to subscription. |
Create an ECS instance | ||
ECS instance type |
For more information, see Dedicated host types. | You can create an ECS instance based on a predefined instance family. For more information, see Overview of instance families. |
Upgrade or downgrade an ECS instance | ||
Release an ECS instance | If you do not renew expired dedicated hosts, the ECS instances on the hosts are automatically released. |
|
Adjust public bandwidth |