What this calculator works out
This calculator works out how many virtualisation hosts a set of servers needs, taking the greater of the CPU and memory requirements, then adds a reserve host.
The reserve is not padding. Without it, a single host failure has nowhere to move its workloads, which removes most of the reason for virtualising in the first place.
Enter your details
Before you plan a consolidation
- Usable host resources are less than installed. The hypervisor, management agents and memory overhead all take a share.
- Memory is usually the binding constraint. CPU overcommits well; memory does not, and overcommitting it causes swapping and severe performance loss.
- The reserve host should be able to absorb the largest host's workload, which the N+1 model assumes.
- Storage performance, network bandwidth and licensing all constrain consolidation independently of CPU and memory.
Consolidating twenty servers onto one host is not resilience, it is a bigger single point of failure. The value comes from having somewhere to move workloads when a host fails or needs patching, which requires at least one host's worth of spare capacity across the cluster. A cluster running at full capacity cannot tolerate the failure it exists to survive.
How this calculator works
The greater of two constraints, plus a reserve:
Hosts by CPU = total vCPU ÷ usable vCPU per host, rounded up
Hosts by RAM = total RAM ÷ usable RAM per host, rounded up
Minimum hosts = the greater of the two
Total = minimum + reserveTaking the greater of the two is correct because both constraints must be satisfied simultaneously — a cluster with ample CPU and insufficient memory cannot run the workload.
Worked example: twenty servers
Using the default figures — 20 servers averaging 2 vCPU and 8 GB RAM, hosts with 48 usable vCPU and 256 GB usable RAM, 1 reserve host:
- Hosts by CPU: 1 · Hosts by RAM: 1
- Minimum active hosts: 1
- Hosts including reserve: 2
Twenty servers onto two hosts, one of which is spare. The reserve doubles the hardware and is the whole point — with one host, twenty workloads stop when it fails or needs a firmware update. It also shows how much headroom a modern host has: 40 vCPU and 160 GB against 48 and 256 available, leaving room for growth on the active host alone.
Common mistakes
- Consolidating without a reserve host. It creates a larger single point of failure.
- Using installed rather than usable resources. The hypervisor takes a share.
- Overcommitting memory. CPU tolerates it; memory does not.
- Ignoring storage performance. Twenty workloads on one array can saturate it.
- Forgetting licensing. Some products license per host or per core, which changes the economics.
Frequently asked questions
How much can I overcommit CPU?
Ratios of two or three virtual CPUs per physical core are common for general workloads, since most servers are idle most of the time. Latency-sensitive and consistently busy workloads need far less overcommitment or none. Monitoring CPU ready time rather than utilisation is what reveals whether a ratio is too aggressive.
Why can I not overcommit memory the same way?
Because a virtual machine's allocated memory is generally in use rather than idle, and when a host runs short the hypervisor resorts to ballooning, compression and finally swapping to disk — which is orders of magnitude slower than RAM. The performance collapse is severe and sudden, which is why memory is usually the constraint that determines host count.
What is N+1?
A cluster sized so that it can lose one host and still run every workload. It requires at least one host's worth of spare capacity distributed across the cluster, not merely an idle machine. Larger clusters achieve it more efficiently — N+1 on ten hosts costs 10% of capacity, on two hosts it costs 50%.
Does consolidation always save money?
Usually on power, cooling and rack space, and often on hardware. Licensing can move the other way, since some products charge per physical host or core regardless of how many virtual machines run on it. The full comparison should include licensing, support, and the cost of the reserve capacity resilience requires.
Is what I enter stored?
No. Figures are processed in your browser and never transmitted or retained.
Related tools
References
- National Cyber Security Centre — resilience, availability and secure system design
- GOV.UK — guidance on technology procurement and service resilience
Sources are checked at publication and can change — how I choose and check references.