Keep Capacity in Reserve
Avoid using every available resource. Free capacity gives the environment room for traffic spikes, new customers, hardware failures, and workload migrations without immediate performance trouble.
Do not plan for 100%
A server close to full capacity may survive normal operation, but even a small rise in demand can overload it. Customer workloads are not constant; reserve CPU, memory, storage, and network capacity to absorb temporary peaks.
Plan for failures
When a host fails, its workloads must run elsewhere. The surviving hosts need enough free capacity to take the additional load.
Recommended approach
Leave capacity unused across hosts and review total headroom before accepting new customers. Ask not only “Can this run now?” but also “Can it still run if demand rises or a host fails?”