When you create a new ESX cluster and turn on its VMware HA feature, one of the first configuration wizards deals with admission control. A screenshot of that wizard is shown here.
Figure: admission control
You'll find three settings boxes in its wizard. The first provides a place to enable Host Monitoring. This monitoring is necessary for the rest of HA to function properly. Thus, you'll need to enable it if you want to use HA. The second and third boxes provide locations to enable admission control and select the policy it will use.
Administrators often make mistakes with these settings. The problem is that incorrectly set admission control settings can have a major effect on the functionality of your cluster.
You can think of admission control as an automated tool to manage what I'll call "cluster reserve." That cluster reserve represents a quantity of resources—processing and memory—that must remain unused in case an ESX host goes down.
You already know that HA is the mechanism to automatically fail over virtual machines (VMs) as that host dies. But for those VMs to fail over, they need a quantity of unused resources located somewhere else in the cluster. Admission control sets aside those resources so they're available for use by failed over VMs. Without it, a host failure and its subsequent VM failovers could create a situation where a host is overloaded, forcing every VM to vie for resources, a situation often called contention.
Once it's enabled, admission control sets aside that amount of resources, spread across all the cluster hosts, so that contention won't happen when a cluster host fails.
Catch up with @ConcentratdGreg on Twitter!