Origin timing · Bucharest
Bucharest: the last mile is fast, so measure ours
When the visitor's connection is genuinely quick, every remaining millisecond belongs to the server, and that makes the diagnosis unusually clean.
The short answer
Romania has some of the quickest consumer internet in Europe, which removes the excuse most hosting conversations lean on. If a page feels slow to a Bucharest visitor on gigabit fibre, the delay is at the origin, and it will show up in a single curl run split into its four gaps.
Ask curl to print time_namelookup, time_connect, time_appconnect and time_starttransfer, then read the differences rather than the total. DNS, the TCP handshake and the TLS negotiation are one bucket; the space before the first byte is the server thinking, and that is the number NVMe, a LiteSpeed cache and a current PHP version exist to shrink.
By the HostingFast team · Reviewed 12 August 2026
Free
Migration in, engineer-run
24/7
Human answers
Daily
Backup cadence
$0
To get started
A developer workforce deep enough to staff Western product teams outright, an IT services economy that outgrew the outsourcing label years ago, and a domestic ecommerce market that grew up alongside it and expects the same speed.
This page is for the person who will actually run the command. What each gap in the timing breakdown means, which of them you can move, what the London origin adds, and when the honest answer is a different tier rather than another plugin.
Read the four gaps, not the total
One curl run gives you the whole diagnosis. The gap up to time_namelookup is DNS, and if it is large the answer is your resolver or your TTL rather than your host. The gap to time_connect is the TCP handshake, which is roughly your round trip. The gap to time_appconnect is TLS, paid once per connection. Everything after that until time_starttransfer is the server building a response.
Most arguments about hosting speed are conducted without this breakdown, which is why they go round in circles. A 900 millisecond page with 40 milliseconds of network and 800 of server time is not a distance problem, and a plan upgrade is not automatically the fix either.
Which of those gaps you can actually move
The server gap is the one worth your attention, and it splits again. A cached response is measured in the time it takes to read a prepared file from NVMe. An uncached response is PHP plus database work, and that is where an object cache, a current PHP version and an index on the column your listing filters earn their keep. On the top shared tier, PHP X-Ray on CloudLinux Pro names the slow call directly rather than leaving you to bisect the plugin list.
The network gaps are mostly fixed. You can reduce how many times you pay them — one hostname instead of four, connection reuse, no redirect chain in front of the page — but you cannot negotiate physics. Spend your effort where the numbers say it goes.
What London adds from Romania
Romania sits in the Eastern Europe and Turkey band we publish for the London origin: 30 to 60 milliseconds of round trip. On a page with a healthy cache that is one handshake's worth of delay and it is not the interesting part of the waterfall.
Where it is worth planning around is your own working day. Uncached admin screens, SSH and file-manager sessions speak straight to the origin every time, so an editor in Bucharest feels the distance that a reader never notices. That is a real cost and we would rather you priced it in before you buy than discover it during a busy week.
When the answer is a bigger tier
Sometimes the profiling comes back and says the code is fine and the machine is busy. The signal is a first byte that flies at seven in the morning and drags at four in the afternoon on a page that is already cached — that pattern is contention, and no front-end work touches it.
The route out is deliberately dull. Move up a shared tier in place, since the account is upgraded rather than migrated, and step to a KVM VPS with root access when shared execution limits genuinely bind. .ro registers on the same account either way, with DNS aimed correctly and the certificate issued as the name connects.
The parts of the number we own
NVMe storage under every account with a LiteSpeed cache built into the server, so a cached response is a disk read rather than a rebuild.
Per-site PHP switching, SSH with Git and Composer, and PHP X-Ray on CloudLinux Pro on the top shared tier for the requests that cannot be cached at all.
- PHP X-Ray on CloudLinux Pro on the top shared tier
- SSH with Git and Composer from the first day
- In-place tier upgrades with no migration involved
- .ro registered and renewed on the hosting account
Why HostingFast
Standard on every plan
Profiling rather than bisecting
PHP X-Ray on CloudLinux Pro on the Nitro tier points at the slow call, turning a week of disabling plugins into an afternoon of fixing one.
Cached responses are disk reads
A LiteSpeed cache in the server plus NVMe storage means the fast path is genuinely fast rather than merely less slow.
A terminal, not a ticket
SSH with Git and Composer on the developer-focused plans, so the slow query log is something you read yourself.
Runtimes beyond PHP
Node.js, Python and Ruby deploy natively on the developer tiers, with PostgreSQL available alongside MySQL.
Upgrades that do not move you
Tier changes happen in place — the account is upgraded, nothing is copied and nothing changes address.
Migration measured, not promised
Engineers move the site free with the old one still serving, and DNS waits on your sign-off so before and after are comparable.
Quick Start
From order to online
- 1
Run the breakdown
One curl printing time_namelookup, time_connect, time_appconnect and time_starttransfer. Read the gaps between them, not the total.
- 2
Attack the server gap
Object cache, current PHP, an index on what the listing filters, and X-Ray on the top tier when the culprit will not confess.
- 3
Repeat it at four in the afternoon
A wide spread between quiet and peak on an already-cached page is contention, and that is a tier decision rather than a code one.
Built In
Loaded onto every plan
- PHP X-Ray on CloudLinux Pro on the top shared tier
- SSH, Git and Composer available from the first day
- Per-site PHP version switching from the control panel
- LiteSpeed caching inside the server rather than added by plugin
- NVMe drives under the cheapest plan as well as the dearest
- Node.js, Python and Ruby on the developer-focused plans
- Daily backups with self-service restores from the panel
- In-place tier upgrades, with no migration when you change plan
- Free SSL on every plan, reissued before it can lapse
- Human support on duty every hour, on the awkward questions too
Frequently Asked
What people ask us most often
What exactly should I run to get a useful first-byte figure?
curl against a real page with a write-out format printing time_namelookup, time_connect, time_appconnect and time_starttransfer, then subtract each from the next. That gives you DNS, round trip, TLS and server time as four separate numbers. A single total tells you nothing about which of the four to argue with.
Our fibre is gigabit and the site still feels heavy. What does that rule out?
It rules out the last mile, which is genuinely useful. On a quick connection the remaining delay is either the origin thinking or the browser working, and those are distinguished in one look: server time shows up before the first byte, browser time shows up after it. Take the curl reading first, then open the performance panel only if the server side comes back clean.
Does a UK-registered supplier make anything easier for an agency billing Western Europe?
Often it shortens the vendor-approval chain your own clients insist on. English-law terms, itemised invoices and a single named jurisdiction are what a procurement questionnaire is asking for, and having them answered in advance is time you would rather spend billing than corresponding.
How do I tell contention apart from bad code?
Timing, not intuition. Measure the same already-cached page at a quiet hour and again mid-afternoon. Code is slow at a consistent rate; a shared machine under pressure is slow in a pattern. If the spread is hundreds of milliseconds, that is contention, and the fix is a larger tier or a VPS rather than another caching plugin.
Keep reading
Web Hosting in Brighton
Brighton, where the same profiling routine runs inside the shortest band.
Web Hosting in Istanbul
Istanbul, and the same origin measured from a two-market catalogue.
TTFB (Time To First Byte) (Glossary)
Time to first byte defined properly, with the worked example this page uses.
Magento Hosting
Plans with the memory and PHP headroom Magento actually demands.
Domain Names
Registering and pointing the .ro, on the same account as the hosting.
Changing hosts? Run through our checklist first.
A straightforward sequence for a switch your visitors never feel: which files move first, how to shift email across without losing a single message, the right moment to repoint DNS, and the two mistakes behind almost all the downtime we get asked to rescue.
Read the gaps, then decide.
NVMe under a server-level cache, profiling on the top tier, and a documented step up when contention is the answer.
View Magento Hosting plans