Scheduled Cadence

Back to Performance Tuning and Capacity Planning · Forecast vs. Actual · Cost-Aware Scaling · Feeds Back Into Tuning · Service Offerings

Capacity reviewed on a set interval, not only after something already started running hot. The difference between the two halves of the illustration above is not how much work was done. It is when, and under what conditions.

1. Why an Interval at All

Reactive capacity work has one guaranteed property: it happens while the system is already under strain, which is the worst moment to be making decisions about it. The people who understand the system are busy restoring it, the options are narrower, and anything that takes a lead time is no longer available.

A scheduled review moves the same analysis to a moment when nothing is on fire. It costs an hour. The scramble costs an evening, a rushed purchase, and usually a change made without the usual review.

2. Picking the Interval

There is no universal number. The interval should be shorter than the time it takes you to outgrow your headroom, which is something per-resource runway already tells you.

• Monthly where growth is fast or unpredictable, or where the runway on any resource is under a year.
• Quarterly for most steady systems. This also lines up with the budget cycle, which is where capacity decisions eventually have to land.
• Before known events regardless of the interval — a campaign, a large client onboarding, a seasonal peak. See pre-launch validation.
• After any significant architectural change, because every previous measurement now describes a system that no longer exists.

A short review that actually happens is worth more than a thorough one that is scheduled annually and skipped twice.

3. The Standing Agenda

The review is the same questions every time, which is what makes it quick. It should be possible to run in forty minutes once the inputs are automated.

• Current utilisation per resource, from the utilisation breakdown — CPU, memory, disk, I/O, network, and the application-level limits that are usually the real ceiling.
• Runway on each, and which one is shortest. That one is the subject of the meeting.
• Last period's forecast against what happened — see forecast vs. actual.
• What changed: new features, new clients, new integrations, retired workloads. These explain the deviations.
• What is coming, from scenario modeling.
• Decisions with a lead time that must be started now to land in time.

4. Who Is In the Room

A capacity review attended only by infrastructure people can measure, but it cannot act. The two things it needs from elsewhere are the plans and the budget.

• Someone who knows what the product is shipping next quarter.
• Someone who knows what sales has signed or is about to.
• Someone who can approve spending, or at least start that conversation.

Three people for forty minutes, four times a year, is a small standing cost against the thing it replaces.

5. It Has to Produce Decisions

A review that produces a document is a review that will be cancelled within a year, because nobody can point to what it changed. Each one should end with a small number of named items: something to provision, something to tune, something to measure before next time, or an explicit decision to do nothing and why.

Where the finding is a bottleneck rather than a shortage, it belongs in the tuning cycle — see feeds back into tuning.

6. Automate the Inputs, Not the Judgement

The numbers should be waiting when the meeting starts. Nobody should be gathering graphs during it, because that is the part that makes the review feel expensive and the part that gets dropped first.

A generated report with utilisation, trends, runway, and last period's forecast error turns the meeting into a discussion about what to do, which is the only part that needs people.

How We Approach It

1. Set the interval from the shortest runway, not from the calendar's convenience.
2. Fix a standing agenda so the review is the same questions each time and stays short.
3. Automate the inputs into a report that exists before the meeting.
4. Get the plans and the budget in the room, not just the infrastructure.
5. End every review with named decisions, including the decision to do nothing.
6. Add event-triggered reviews on top of the interval, not instead of it.

What You Get

• A review interval justified by your own runway figures.
• A standing agenda and a generated input report, so the meeting is short enough to survive.
• A record of decisions per review, which is what makes the cadence defensible.
• Capacity conversations that happen with time to act, instead of under load.

The question worth asking: when was capacity last looked at deliberately, rather than because something made you look?