Records & measurement · Beginner · 20 minutes
The First Segment of Every Speed Test Belongs to Your DNS
The opening bar of the waterfall is not your server at all, and when it is long the repair is in your records rather than in your hosting plan.
The short answer
Run PageSpeed Insights and WebPageTest, then read the waterfall from the left: the first segment is name resolution, which is your nameservers, your TTLs and any CNAME chain the request has to follow. Connection and TLS come next, and only then does time to first byte — the server's own thinking — begin. Optimising the wrong segment is the most common way to spend an afternoon and move nothing.
Below: why the first request of the day looks slow and the rest do not, why www and the bare form give different numbers, and which figures are worth acting on.
By the Hosting & Domains team · Reviewed 18 August 2026
Beginner
Level assumed
5
Stages in the procedure
Free
Cost of asking us
Proven
Where it was checked
No technical background needed. This was written for people who manage names rather than servers, proven on the platform we actually run, and honest about which parts are genuinely fiddly rather than merely unfamiliar.
One promise before step one: nothing here is a one-way door. Any step that is awkward to reverse is flagged, and the way back is printed beside it.
Read the opening bar before anything else
A page arrives after several separate delays, and the first is entirely yours: the visitor's resolver has to find out where the name points. If it already knows, because somebody nearby asked recently and the answer is still inside its TTL, that costs nothing. If it does not, the query walks up to your authoritative nameservers and back, and the distance is real.
Two things lengthen it. Nameservers a long way from your audience, which is a delegation decision rather than a hosting one. And a chain of CNAME records, where each link is another lookup before an address is finally returned. Neither shows up as a server problem, and no amount of caching on your side will shorten either.
Two tools, one picture
PageSpeed Insights gives you Core Web Vitals and lab detail. WebPageTest gives you the waterfall, which is where the DNS, connect, TLS and first-byte segments are separated out. Between them you get the whole picture rather than half of it.
Set the test location to where your audience actually is. Testing a UK site from a North American node charges the transatlantic hop to your account, and the location sits in a dropdown on both tools.
Test more than the homepage, and both forms of the name
The homepage, a product or service page, a content page. Each template behaves differently and the homepage is frequently the least representative page on the site.
Then test www and the bare name separately. If one redirects to the other, that visitor pays for the whole opening sequence twice — a lookup, a connection, a TLS handshake, a 301 — before the real page even begins. It is a small number that only ever applies to first-time visitors, and it is worth knowing rather than guessing at.
Field data against lab data
Field numbers are gathered from real browsers over weeks, and those are what ranking feeds on. Lab runs are diagnostic instruments that wobble with the test server, the network and the moment. A lab score of 60 with green field vitals beats the reverse every time.
Chasing a perfect lab hundred is the classic way to lose a day. Run three, take the median, and make decisions from field data.
One change, one measurement. Attribution is the entire point of testing. Five changes at once leave you with a faster site and no idea which one earned it, which means you cannot repeat it anywhere else.
Keep a note of the figure before and after each change, and include the DNS segment in that note. It is the number most likely to move for reasons that have nothing to do with the work you just did.
Why the job is shorter when the name is here
Every walkthrough here is run on the platform we actually operate — cPanel with a full zone editor, LiteSpeed, NVMe, one-click installs — so the screen the words describe is the screen in front of you.
The rate you register at is the rate you renew at, so year two costs exactly what year one did and nothing is lying in wait on the invoice.
- Checked against the panel your account actually opens
- Registry, registrar and host kept firmly apart
- The trap named before you walk into it
- Someone awake at the hour a record breaks
Why Hosting & Domains
Standard on every plan
Someone awake when a name breaks
Expiry dates and DNS faults keep no office hours, so the desk answers at whichever hour you find one.
Five stages, none of them filler
Each stage is a short spell of deliberate clicking, and the ones that are genuinely fiddly are labelled fiddly.
The way back, printed beside the risk
Anything awkward to reverse is flagged, with the route back written next to it rather than three paragraphs later.
The trap named before step one
The classic mistake on this particular task is named before you begin, which is the difference between the time estimate above and a lost evening.
Screens that match the words
Written against the panel your own account opens, so the buttons sit where the page says they sit.
Written from the ticket queue
Every trap named here came out of a real support ticket, which is why the awkward ones get named at all.
Quick Start
Order placed to site online
- 1
Run both reference tools from your audience's location
PageSpeed Insights for Core Web Vitals and lab detail, WebPageTest for the waterfall. Set the test location deliberately; the default is rarely where your visitors are.
- 2
Read the DNS segment as its own number
It is the first bar and it belongs to your records rather than your server. A long one points at nameserver distance or a chain of CNAMEs, neither of which caching will help.
- 3
Test both forms of the name
www and bare, separately. Where one redirects to the other, first-time visitors pay for the opening lookup, connection and handshake twice before the page begins.
- 4
Treat TTFB as the server's own number
This is the thinking time before a single byte of content moves. A cached page on decent hosting serves nearby visitors comfortably under 200ms; readings persistently above roughly 600ms point at hosting, caching or the application.
- 5
Change one thing, then measure again
Note the figures before and after, DNS segment included. Attribution is why you are testing at all, and it survives exactly one change at a time.
Built In
Fitted to every plan
- Year one of the registration included with an annual plan
- A renewal figure identical to the one you first registered at
- WebP image optimisation built in at no extra charge
- Softaculous included, for installs that take one click
- Money back inside 30 days on hosting plans, 7 on reseller
- Your current site moved across by our engineers at no charge
- No set-up charge and no joining fee at any point
- A 99.9% uptime target, watched every hour of the day
- Spam and virus screening on every mailbox as standard
- LiteSpeed caching inside the server, not bolted on by a plugin
Frequently Asked
Questions we field again and again
The first request of the day is slow and every one after it is quick. Why?
The first caller pays for the lookup and everybody behind them gets it free. When no resolver nearby has a current answer for your name, the query has to reach your authoritative nameservers; once it has, the answer is reused until the TTL expires. That is why a test at eight in the morning and one at eleven can differ by a noticeable margin without anything having changed.
Does a chain of CNAME records cost measurable time?
Yes, though it is usually tens of milliseconds rather than hundreds. Each link is a further lookup before an address is finally returned, and chains grow quietly as services are layered on top of one another. Where a hostname points at a service that points at another service, it is worth knowing whether the middle link is still doing anything useful.
Should I lower my TTLs to make the site faster?
The opposite. A short TTL means every resolver asks your nameservers far more often than it needs to, so more visitors pay for a full lookup. Short TTLs are a tool for planned changes — drop them a day before a move so cached answers expire quickly — and should go back up afterwards. Speed and agility pull in different directions here, and the everyday setting should favour speed.
If the numbers say the hosting is the limit, what is the move?
You upgrade in place. Plan changes are made from the client area with no migration and no downtime, and the range runs from small shared accounts through VPS to full dedicated machines. Growth is a setting on the account rather than a search for a new host, and the names stay exactly where they are throughout.
Keep reading
How to Create an SPF Record
Declare publicly which servers may send under your name — intermediate level, about 10 minutes.
Business Hosting
Shared hosting carrying the whole developer kit, from SSH through Node.js and Python to PostgreSQL.
WordPress Hosting
WordPress looked after for you — LiteSpeed caching, staging copies and a daily backup.
Changing provider? Work through this checklist beforehand.
A straightforward running order for a migration your visitors never spot: which files travel first, how to bring the mail across without dropping a single message, the right moment to repoint DNS, and the two errors that sit behind almost every outage we get called in to fix.
Put the site on a name you own outright.
NVMe hosting with the certificate and the migration included, people answering at any hour, and a figure that stays where it started.
View Business Hosting plans