Isvara
·visits
Tools/08 · Failover Capacity

Lose a host at peak.

Size the cluster, set the host failures it must survive, and see what is left.

N+1 verdict
protected
80%
Memory on the survivors after losing 1 of 8 hosts
8 hosts · lose 1
The clusterpeak load, per cluster
Host failures to tolerate
Peak CPU consumed55 %
Peak RAM consumed70 %
When it happenshatched: HA reserve
CPU549 of 998 GHz used
today
after
reserve 125 GHz63% of survivors
Memory5,734 of 8,192 GB used
today
after
reserve 1,024 GB80% of survivors
N+1: protected.
Losing 1 of 8 hosts leaves the memory of the survivors at 80%: the surviving hosts absorb the failed host’s load with headroom to spare.
Stay at or below
78 %
of the full cluster to stay green after 1 failure
HA reserve
13 %
1/8, the default cluster resource percentage
Survivors
7 hosts
874 GHz · 7,168 GB

Green means the survivors run at or below 90%. Slot policy and dedicated failover hosts reserve differently from the percentage policy modelled here.

In the guidethe book on HA

The book reads availability at the provider layer from the vSphere Availability dashboard, which assumes an N+1 design, and recommends enabling HA admission control.

Official Broadcom docstechdocs.broadcom.com