About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
Headroom is the spare capacity between what a server uses at its busiest and what it has. Set it per resource, in writing: a trigger level at which an upgrade is ordered, chosen so that growth during the lead time, plus the next known peak, still fits below the ceiling. The right figure depends on how fast the resource grows, how sharp its peaks are and how long an upgrade takes, so there is no single correct percentage. What matters is that the trigger and its assumptions are written down and checked.
Why it matters
Without an agreed trigger, upgrades are decided in the middle of a problem: the disk is full, month-end is slow, someone approves an emergency change. With one, the conversation happens months earlier, with numbers. Google's Site Reliability Engineering book observes that "a surprising number of services and teams don’t take the steps necessary to ensure that the required capacity is in place by the time it is needed". The fix is not a bigger server; it is a rule that triggers in time.
How it works
Three things set the trigger
- Lead time. How long from "we need more" to "it is in place", including approval, budget and the change itself. Google lists as mandatory "An accurate organic demand forecast, which extends beyond the lead time required for acquiring capacity". If an upgrade takes two months, the forecast must see at least two months ahead, and the trigger must leave room for two months of growth.
- Peaks. The gap between a normal day and the busiest day. Microsoft's capacity guidance says to "Identify the peak usage times, user load, transaction rates, and other relevant metrics", and lists "Regular spikes (8:00 AM sign-in rush)" and "Seasonal change" among the times to plan for. Headroom must cover the next peak, not just the average.
- Failure and business events. Google describes capacity planning as ensuring "sufficient capacity and redundancy to serve projected future demand with the required availability", and separates growth from customer adoption from "inorganic growth (which results from events like feature launches, marketing campaigns, or other business-driven changes)". A new branch or a new application is a capacity event, not a surprise. The book describes Google's own very large services; this article applies the same ideas to a small business's servers.
The simple sum
A simple sum we suggest (our own construction, built on the three points above, not a formula from the sources):
Trigger level = ceiling − (growth per month × lead time in months) − peak allowance.
Work it out per resource, then round down to a figure people will remember.
What to choose and why
Storage
Storage grows steadily and, when full, stops things. Use month-end used space and the growth rate from your forecast.
Illustrative example: a 2,000 GB data volume grows 40 GB a month, lead time is two months and the business wants 10 per cent (200 GB) kept free for one-off jobs such as a large import. Trigger = 2,000 − 80 − 200 = 1,720 GB, about 86 per cent. Rounded down, order more at 85 per cent. (Our forecasting article's example arrives at 80 per cent for a different volume; each is worked from its own figures, which is the point.)
Processor
Processor use is spiky; a server at 100 per cent for a moment is normal. Watch sustained business-hours peaks. Set the trigger on the peak, not the average, and lower it if users notice slowness before the figure looks high.
Illustrative example: peaks of 60 per cent at end of quarter, rising about five points a year. A trigger of 70 per cent sustained peak leaves room for one more year's growth and a bad day.
Memory
Memory figures can mislead on database servers. Microsoft notes that SQL Server "can change its memory requirements dynamically based on available system resources", and that over time, under normal circumstances, it will "claim memory up to the limit set by max server memory (MB)". So a database server can show memory near full while working normally. Our suggestion, which follows from that: for those servers, trigger on symptoms the application owner agrees on, such as slower reports or more disk reads, rather than the raw memory figure. For other servers, a peak in-use trigger works like processor.
Write it down
A one-page headroom policy per server is enough:
| Resource | Measure | Trigger | Lead time assumed | Owner |
|---|---|---|---|---|
| Data volume | Month-end used space | 85% | 2 months | IT contact |
| Processor | Sustained business-hours peak | 70% | 1 month | IT contact |
| Memory (non-database) | Business-hours peak in use | 80% | 1 month | IT contact |
The figures in this table are illustrative; work out your own with the sum above. Microsoft's guidance adds to plan "before usage changes", so add a line for known business events.
Read next
Related articles you may want next.
- HostingForecasting capacity from a year of measurements: when is the next upgrade due?How a business turns twelve months of storage, memory and processor figures into a likely date and budget line for its next hosted-server upgrade, using two simple forecasting methods with worked arithmetic.Read article
- HostingRight-sizing over-provisioned servers: reclaiming CPU and memory safelyHow a business proves a hosted server has more processors or memory than it needs, reduces them in measured steps without hurting peak work, and keeps a way back if it goes wrong.Read article
- HostingSizing a business application server from measurements, not the vendor's minimumWhich Windows performance counters to collect on an existing server, for how long and at what interval, and how to turn the results into processor, memory and disk sizes for a hosted replacement.Read article
Risks and limits
- A fixed percentage is not a policy. 80 per cent of a small disk and 80 per cent of a large one leave very different amounts of room. Check the absolute figure too.
- Lead time may change during the year. Ask your provider, and your own approvers, what lead time applies near financial year end and holiday periods, and use that figure for those months.
- Triggers drift out of date. Revisit them when the forecast, the server size or the business changes.
- Peaks hide in averages. Daily or monthly averages smooth away the busy hour. Measure peaks.
- Too much headroom costs money. Where a plan charges for allocated resources, generous headroom is paid for every month. Balance it against the cost of an emergency.
Sources
Opened 15 November 2026; the SQL Server memory page and the published sizing article opened 16 November 2026.
- Google: Site Reliability Engineering, chapter 1: Introduction, no revision date shown.
- Microsoft Learn: Architecture strategies for capacity planning, Azure Well-Architected Framework, updated 30 October 2025.
- Microsoft Learn: Server memory configuration options, page shows last updated 11 May 2026 (page metadata: modified 19 October 2026).
Related article: Sizing a business application server from measurements, not the vendor's minimum covers measuring peaks and recording headroom when a server is first sized.
Related service: Cloud hosting.
Learning notes
- Headroom must cover growth during the lead time plus the next peak.
- Trigger = ceiling − growth × lead time − peak allowance.
- Set triggers on peaks, not averages.
- Database server memory can look full while healthy; trigger on agreed symptoms.
- Write the policy down with the lead time it assumes, and revisit it.
This article was originally published on .