A client with two sites complained about slowness. The natural suspicion is an undersized server, and that conversation usually ends in a more expensive plan.
It was not that. It was two switches left off, in services he already paid for.
What was happening
On the first site, DNS was already on Cloudflare but the proxy was off on the records. The name resolved through Cloudflare and the visitor went straight on to the origin server. The whole network the client had already bought sat there idle, carrying no traffic at all.
On the second, the host’s CDN showed up in the panel and had never been switched on.
The numbers, before and after
Measured with curl, from Brazil. Five runs on the 17th and ten on the 18th, after the change.
| Time to first byte | Before | After |
|---|---|---|
| Median | 1.306 s | 0.256 s |
| Worst case | 14.738 s | 0.873 s |
| Best case | 0.606 s | 0.231 s |
The median fell 5.1 times. The worst case fell 17 times, and the worst case is what rules. In one of the five measurements on the 17th the site did not answer at all, and in another the TLS handshake alone took 8 seconds. One failure in every five attempts. After the change, none.
Of the static files tested, 14 out of 14 now come from the edge cache. Before, it was zero.
What was not touched
Turning the Cloudflare proxy on means editing DNS records, and that is where email lives. The MX records and the DKIM and DMARC entries stayed as they were. That is the classic mistake with this change: the site gets fast on Tuesday and accounting stops receiving invoices on Wednesday.
The uncomfortable part
This gain did not come from engineering. It came from opening two panels. The client paid for both services and received half of one, and nobody had measured.
That is why I measure before proposing anything. In a good share of slowness cases, the first thing to surface is neither bad code nor a small server. It is configuration somebody left half-finished, invisible until you put a stopwatch on it.
Need this solved at your company?
A twenty minute call is usually enough to tell whether it makes sense, and you talk straight to the engineer who builds it.