When a site slows down, the first idea is usually a bigger plan. It takes a few clicks and it feels like doing something. But most slowness has nothing to do with server capacity, and adding capacity while leaving the cause in place spends money without changing what anyone feels.
This part has one job — telling apart the situations where an upgrade is the answer from the ones where it is not. Getting that right saves a meaningful share of the budget.
What an upgrade fixes, and what it never touches
The reason the left column is so common is simple: the server is usually idle, and the time goes into transferring files and doing wasted work. Double the CPU and memory in that state and the idle server is merely idle at a larger size.
What the real signals look like
A genuine capacity problem does not present as “slow”. It presents as a specific failure. If any of the following repeats, capacity really is the constraint.
Three signals genuinely mean capacity: pages that fail to open at all under a rush, processes dying mid-task and leaving a blank screen, and disk headroom shrinking steadily towards warnings. What they share is that they present as failure rather than as mild sluggishness.
Two others are easy to mistake for the same thing — an admin that is slow regardless of visitor numbers, and responses that wobble in the middle of the night when nobody is visiting. Read those two differently. A slow admin only is usually a database problem in the area caching cannot reach, so clearing out accumulated junk beats a bigger machine. Wobbling when nobody is visiting more often comes from neighbours on shared infrastructure or from scheduled jobs.
Three boxes to pass through first
Even when an upgrade looks necessary, walking the steps below often removes the need. And there is a stronger reason to walk them — upgrade without doing these three and you meet the same problem again in the next growth phase, on a more expensive plan.
The order is deliberate. Image work costs nothing and gains a great deal, and caching lets the same machine serve far more visitors. If the real signals above survive both, the upgrade that follows is a purchase you make with confidence rather than a guess.
The image and asset order was covered in where the least money buys the most speed. Broader performance material is in the performance archive, and if you would rather measure the bottleneck before spending, our process page publishes the order in which we measure and fix.
Next part
The final part. Everything decided so far goes stale with time — the closing part covers the thirty-minute review that puts it right twice a year.