When marketing and virtualization collide, the result can foster confusion, especially when similar terms are used to describe both a general technology and a specific product.
In general IT terms, high availability refers to a system or strategy whose primary goal is to maintain uptime. Then there's VMware High Availability (HA), a specific technology in vSphere that automatically restarts a virtual machine (VM) in the event of a hardware or software failure.
Defining what exactly constitutes high availability can be subjective. The general consensus is high availability is achieved when a system has uptime of 99.999% -- a total of 5.39 minutes of downtime, either by design or by accident, over the course of a year. This vaunted "five nines" of availability typically means certain measures have been implemented to construct a near-bulletproof system.
VMware HA, introduced in 2006, works when several hosts and their respective VMs are placed in a cluster. One host, which is selected as the master host, monitors the network heartbeats of the other hosts in the cluster to detect any downtime. If the master host finds another host in the cluster is not responding, or if the VMs on a host power off, the master host will restart those VMs on another host in the cluster.
Systems using VMware HA are reactive and will experience some downtime following the failure of a host or an application before the VMs are restarted. System administrators who need to keep business-critical applications running without interruption should consider VMware Fault Tolerance (FT), a high-availability feature introduced in 2009. A system with FT enabled has identical VMs running on separate host servers; if the primary host server fails, the virtual machine will continue to run without interruption on the secondary server.