The vSphere HA feature offers additional protection for virtual machines running on a cluster. There are three scenarios it can handle:

HA is a cluster setting that applies to all running VMs. Once enabled, it protects all virtual machines and allows administrators to adjust additional settings such as VM prioritization and monitoring settings.

vCenter Server is required to configure the feature. It's not required for HA to operate and once the feature is enabled, it works even if vCenter is down.

Host monitoring prevents ESXi host failure. When it's enabled, HA monitors the host's heartbeats. If heartbeats are not received within a certain period of time, HA restarts VMs on another host.

VM monitoring is a protection against...

vSphere HA is a cluster-wide setting. This means that once it's enabled, it protects all virtual machines that are a part of the cluster.

In most cases, there are more important virtual machines running along with less critical VMs on the same cluster, and it's desirable that more important and critical VMs are restarted first. Also, depending on cluster resources available, vCenter may not be able to restart all virtual machines on other hosts, so it becomes even more important to make sure the most important servers will be backed up first.

Each environment has critical virtual machines, which in case of host failure need to be restarted before other VMs. A good example is domain controllers, which in most cases need to be up before other servers become available. Another example is database servers, which may be required online by applications and web servers. In such cases, the ability to configure VM restart priority can be very useful.

When it comes to VM monitoring and auto-restart of individual VMs, vCenter offers an option to adjust the VM monitoring and application monitoring setting for individual VMs as well as change monitoring sensitivity.

Sensitivity can be set to low, medium, or high. These settings correspond to the following heartbeat timeout intervals: 120 seconds, 60 seconds, and 30 seconds. In other words, VM monitoring responds to missing the VMware tools heartbeat after this period of time. There is also an option to set a custom value for the timeout.

The custom option also allows you to change the following values:

For VMs that need to be up 100 percent of the time and do not tolerate even a brief interruption caused by a reboot as a result of host failure, vSphere offers a continuous availability option called Fault Tolerance (FT).

FT creates and maintains an exact copy of a running virtual machine—secondary VM—on another host. Both VMs exchange heartbeats to monitor each other's status.

When a host with a primary VM fails, the secondary VM becomes active almost instantly. There is a little delay but clients do not see the interruption. As everything that happens on the primary VM replays on the secondary one, this failover happens transparently without the interruption of the existing network connections or in-progress transactions. The replication delay is typically less than 1 millisecond and it's unlikely that there will be transactions that have not yet been replicated. Such transactions, should they happen, will be lost.

When there is more than one host in a cluster, it's usually expected that these hosts are redundant. In other words, if one or more hosts fail, the remaining should still be able to accommodate all VMs.

For this reason, each time a new VM is powered on or more memory or CPU is added to a VM, the administrator needs to be sure that the required redundant capacity is still available.

VMware offers a feature called admission control, which helps to ensure that sufficient resources are available to provide the required reservations or redundancy.

Once enabled, admission control is automatic. It is available not only for hosts but also for resource pools and vSphere HA.

Admission control monitors the environment for certain actions and decides whether they should be allowed or not based on resources that will be available after the action. This is to make sure that once the action is allowed, there are still enough resources to provide the...

Admission control helps to make sure there is spare capacity available in a cluster to keep the required level, redundancy, or reservations.

There are three types of policies available, which are as follows:

When the requirement is to be able to lose one or more hosts and still be able to keep all the VMs needed running, the Host failures cluster tolerates option works quite well unless the hosts you are trying to protect have significantly different sizes—memory and CPU resources available.

This option for different hosts results in reserving an excessive capacity based on the size of the largest host. Reserving more capacity than required in its turn results in wasting resources.

For clusters with differently-sized hosts, VMware advice is to use the percentage of cluster resources reserved option. This option ensures...

In certain cases, administrators may decide to use a separate standby host, which will take over and accommodate VMs from a failed host. This can be accomplished by using the third available option in the Admission Control Policy settings—Specify failover hosts.

When this policy is chosen, vCenter doesn't allow any VMs on the failover host even when a VM is being migrated there. This host will be used only when a failure occurs.

One important requirement is a shared storage available to all hosts, including the failover host. VMs running on the host's local storage will not be migrated in case of host failure.

If, for some reason, the failover host cannot be used to accommodate VMs, HA will try to restart them on other available cluster hosts.

VMware offers a backup and recovery solution called vSphere Data Protection (VDP). While fully integrated with vCenter, this solution provides agentless disk-based backup of virtual machines.

Starting from vSphere 5.1:

Some of the core VDP features are as follows: