Running out of capacity almost never happens overnight. The signs are in your monitoring data months before users notice. A simple capacity plan, checked every quarter, turns a future emergency into a planned purchase.
1. Track Trends, Not Snapshots
Today's CPU use tells you very little. What matters is the trend over three, six and twelve months. Track these for each cluster:
- CPU use, and CPU Ready time on busy VMs, which shows when VMs are waiting for a CPU
- Memory use, including whether hosts have started to swap or compress memory
- Datastore used and free space, and how fast it is growing
- Number of VMs, and how many are added each month
Tools such as VMware Aria Operations, now part of VCF Operations, can collect these trends and project when a cluster will run out.
2. Know Your Growth Rate
Work out how much you add each quarter: new VMs, bigger databases, longer backup retention. Then add known projects, such as a new application or a planned migration, which will not show up in past trends.
3. Keep Headroom for a Failure
A cluster is only healthy if it can lose one host and still run everything. This is often called N+1. If your cluster runs at 85 percent across four hosts, losing one host leaves no room at all. Plan your limits so that the remaining hosts can carry the full load.
4. Plan Around Lead Times
New servers and storage can take 8 to 12 weeks to arrive, sometimes more. Add time for approval, racking, installation and testing. If your plan says you will run out in four months, you may already need to order.
5. Review Every Quarter
Put a short capacity review in the calendar every quarter. One page is enough: current use, trend, forecast date for running out, and any action needed. Small and regular beats a crisis once a year.
A Simple Example
A four-host cluster runs at 60 percent memory use and grows by 3 percent each quarter. To survive one host failure, the other three hosts must carry everything, so the safe limit is 75 percent across all four. At 3 percent a quarter, that limit is reached in about five quarters. With a 12-week hardware lead time plus approval and installation, the order should go in within the next year, not when the alarms start.
Common Questions
Is It Better to Buy Big Once?
Buying too much early ties up money in hardware that sits idle, and newer models may be better value later. Buying too late causes emergencies. A plan that looks 12 to 18 months ahead, reviewed every quarter, is usually the right balance.
What About Backup Storage?
Backup storage grows too, often faster than production because of retention rules. Include backup repositories and the DR site in the same quarterly review.
A Quick Checklist
- CPU, memory and storage trends for each cluster over at least 6 months
- Growth rate per quarter, plus known upcoming projects
- Clusters can lose one host and still run every workload
- Order dates worked back from hardware lead times
- A one-page capacity review every quarter
How We Help
We build a simple capacity model from your monitoring data, set clear limits for each cluster and tell you when to order. When it is time to grow, we size the new hardware and help you compare quotes.
