Q. In VMware High Availability (HA) admission control policies, what does "Host failures cluster tolerates" mean?

My last Q&A discussed the notion of "wasted resources" that get set aside by VMware HA's admission control. It also explained the idea that a specific quantity of cluster resources should be reserved to protect the performance of every virtual machine in case a host dies.

You might be asking yourself exactly how many resources get set aside when admission control is enabled.

Figure: Admission control policy

The answer depends on which admission control policy you select. Three policies become available upon enabling admission control. Those policies, shown in the figure above, are Host failures cluster tolerates, Percentage of cluster resources reserved as failover spare capacity, and Specify a failover host. I'll talk about the first of these in this Q&A and save the others for a later article.

Staff in most environments are likely to choose the first of these policies and set its value to 1. While confusingly named, this policy in fact sets aside a quantity of resources that are equal to one cluster host's contribution. Thus the name Host failures cluster tolerates.

Your reason for using this policy has to do with the amount of host failure protection you'll likely want. To explain, turn back to that last Q&A, where I made the following statement:

In a small ESX cluster of only four hosts, admission control might set aside 25 percent of the cluster's total resources—25 percent represents one host's resource contribution. On the other hand, a cluster of ten hosts might only need 10 percent of its total resources.

A typical mid-sized cluster will rarely see more than a single host perish at a time. By setting this policy's value to 1, admission control will always monitor how much processor and memory resources are available throughout the entire cluster and reserve one host's worth as unused. If your cluster is bigger, the probability that more than one host will die simultaneously goes up. In that case, you might consider raising this value to 2 or more.

It's important to recognize that clusters with different kinds of servers pay an unexpected burden. Admission control policy wants to protect against any host failing, which really means the biggest host failing. As a result, this setting will set aside a quantity of resources equal to that biggest host's contribution. This behavior means that clusters with dissimilar hardware will always lose the contribution of their most-powerful host. Plan accordingly!

Catch up with @ConcentratdGreg on Twitter!