Server migration
You already know how to copy files. What you do not know is the hour your customers stop reaching the old machine and start reaching this one — and what you do if that hour goes wrong. So here is the plan written down: the ten days before the switch, the twenty minutes of the switch itself, and the week afterwards when the old server is still standing in case you want it back.
Four reasons people arrive on this page
Nothing is broken; the bill just stopped making sense — usually a bandwidth line, an add-on that became a tier, or a renewal that was never the advertised price. That is a comparison question, and we keep a page per host for it.
See us next to the host you are onRed Hat ended CentOS 7 on 30 June 2024, and a lot of working servers are still on it. We still install it, which matters more than it sounds: you can move first and upgrade second, instead of doing both frightening things in the same week.
What people actually run hereNoisy neighbours, a CPU allowance you keep hitting, an I/O ceiling nobody will explain. The fix is a machine with nobody else on it, and the move is the same move this page describes.
What a whole machine is likeThe month something you made got picked up, and the invoice arrived with a number on it nobody had budgeted. Ports here are not metered, which is also why copying your data in during the move costs nothing.
How unmetered is billedRead this part twice
This is the single fact that turns a calm migration into a bad one, and almost nobody plans for it. An IPv4 address belongs to the network that holds the allocation, not to you and not to the server. When you leave a provider, the number stays behind. Anything that refers to your service by number rather than by name has to be found and changed before the switch, not after it.
The usual suspects, in the order they usually bite:
Catalogue prices, per month, the same ones the order form charges.
| 1 × IPv4 | included |
|---|---|
| 2 × IPv4 | $6 |
| 4 × IPv4 | $18 |
| 6 × IPv4 | $30 |
| 8 × IPv4 | $42 |
The plan
Nothing below is specific to us until day seven. It is the ordinary shape of a careful move, and we have written it out because the version in most people's heads skips the two steps that cause all the trouble — the allow-lists and the TTL.
Day −10
Not what you think is running. Every listening port, every cron job, every certificate and when it expires, every credential that lives only in someone's head. This is the step people skip and it is the reason migrations get discovered three weeks later by a customer.
Bring the list to us and we will tell you which parts we would move first.
Day −9
Every partner who allow-lists your IP address needs a change request, and every one of them will take longer than you think. Send them all today with the new address once you have it, and chase them while you do the rest.
Day −7
The first operating system you pick is usually not the one you keep. There are 33 images you can install here and 32 of them cost nothing, you reinstall from the panel yourself, and there is no limit and no ticket. Rebuild it until it matches the old server, not until you run out of patience.
Every machine has out-of-band console access, so the firewall rule that locks you out costs you five minutes instead of an evening.
Day −5
Files first, databases later. Your users are still on the old machine and nothing about this step is visible to them, so take the time. Run it twice if it is large — the second pass only moves what changed and tells you how long the real one will take.
If your database engine can replicate to the new machine, set that up now rather than planning a dump on the night. It is the single thing that decides whether the freeze on day zero is seconds or an hour.
Inbound traffic is not metered and not billed, so the size of your data changes how long this takes and nothing else. The arithmetic is further down the page.
Day −3
Point your own machine at the new address with a hosts-file entry. You get the real hostname, the real certificate and the real application, on the new server, while everyone else still lands on the old one. Click through everything that takes money.
Your line, ready to paste
203.0.113.10 example.com www.example.com
/etc/hosts on macOS and Linux, C:\Windows\System32\drivers\etc\hosts on Windows. Take it out again when you are done.
Day −2
The TTL is how long the rest of the internet is allowed to keep believing your old answer. At a day, half your visitors are still on the old server the following morning. At five minutes, the switch is over before you have finished watching it. Lower it a day or two ahead so the long value has expired everywhere by the time you need the short one.
86400s — the world takes up to a day to notice the switch
Day 0
Stop writes on the old server. Run the last delta — minutes, because you already did the bulk. Change the DNS records. That is the whole cutover, and with the TTL down it is measured in minutes, not evenings. Pick an hour when your traffic is lowest and tell us which one; somebody will be on the phone for it.
Nobody honest will promise you zero downtime for this. The freeze is real. What we can tell you is that it is usually seconds to a couple of minutes, and that it is the only part of the ten days that is.
Day 0
+2h
The new one is easy: it either works or you hear about it. The interesting logs are on the machine you are leaving. Anything still arriving there after the TTL has expired is either a resolver behaving badly or — far more often — a hardcoded address you missed on day ten. Every hit is a to-do.
Day +7
Leave it powered down and intact for a week. It catches the mail that went to the old address, the order that came from a stale cache, and the file nobody remembered until they needed it. It is also your rollback, and rollbacks you can still perform are the reason cutovers are calm.
Day +30
By now you have gone a full billing cycle without touching it. Cancel before their cycle turns over rather than after, which is the one deadline in this whole plan that costs real money if you miss it.
The arithmetic
Three of the four numbers below are zero, and they are zero for a structural reason rather than as an offer: there is no migration product in our catalogue to bill you for, and the port is not metered so there is nothing to count. The fourth is the only real cost of any migration — the days you deliberately pay two hosts at once so you have somewhere to go back to.
Our end is a gigabit port unless you buy a bigger one. Most moves are limited by the host you are leaving.
Seven is the least we would run. Fourteen is what most people are glad they did.
Only used to price the overlap. Nothing is sent anywhere — the sum happens in your browser.
The copy takes about
1 h 15 m
At a realistic 60% of line rate. At the full rate it would be 45 m.
| Planning it with you, and the cutover call | $0 |
|---|---|
| The data you pull in | $0 |
| Rebuilding the machine until it matches | $0 |
| Paying your old host while you overlap for 14 days | $56 |
| What the move costs you | $56 |
Your new server is billed from the day it is handed over, like any other month — that is a server price and it lives on the pricing page, not here. What is above is the cost of moving, which is a different question and the one this page is for.
Our refund window is three calendar days from the moment a service is activated, and after that a minimum of one full month is not refundable. So the honest planning assumption is that the first month here is spent whatever happens, and the overlap you are pricing above is the money you spend on the other side. Everything is month to month with no contract; there is a discount for committing, and it is deliberately small.
What "we'll help you migrate" means here
Every machine includes the first row below. If your move needs someone watching it more closely than that, the tier is a line on the invoice you can change or drop in any month — it is not a sales conversation.
| Tier | What it promises | Per month |
|---|---|---|
| Basic - 24x7x12 (incl.) | Response time: 12H - H/W replace 24H | included |
| Bronze - 24x7x4 | Response time: 4H - H/W replace 4H - Adv. support 30min | $39 |
| Silver - 24x7x2 | Response time: 2H - H/W replace 3H - Adv. support 60min | $59 |
| Gold - 24x7x1 | Response time: 1H - H/W replace 2H - Adv. support 90min | $99 |
| Platinum - 24x7x1/2 | Response time: 30M - H/W replace 2H - Adv. support 120min | $129 |
Will my setup be an oddity here?
These are the operating systems and the cities our own live dedicated servers are running on right now, as proportions. We publish the shares rather than the counts because the useful part is whether you would be the only one — not how many machines we have.
Folded by family, so every Ubuntu release counts once. Rarer choices are left off rather than shown at one per cent.
For the copy itself, the city closest to the host you are leaving finishes soonest — but pick the one closest to your users, because you live with that choice for years and with the copy for an afternoon.
All five, with a latency testA gigabit is what a dedicated server comes with, at no charge and with nothing counted. Bigger ports are a catalogue line, priced per month:
| 1 Gbps | included |
|---|---|
| 2 Gbps | $189 |
| 3 Gbps | $299 |
| 5 Gbps | $459 |
| 10 Gbps | $529 |
| 20 Gbps | $1,629 |
| 40 Gbps | $2,799 |
| 100 Gbps | $3,999 |
Before you hand us something that is already working
You are about to trust a company you have not used with a system you cannot afford to lose. Read what other people say about us somewhere we cannot edit it. We have been doing this under this name since 2004, and the longest-running dedicated server on this brand's books has been up for 12 years.
Migrations are the one thing worth a conversation before an order. Book half an hour, bring the inventory from day ten, and we will tell you where we think it will go wrong.
Straight answers
No, and not as a promotion. There is no migration product in our catalogue, so there is no line to put on an invoice. The planning, the cutover call and the rebuilds are included with every dedicated server at every response tier, including the one that costs nothing.
Not from another provider. An IPv4 allocation belongs to the network holding it, so the number stays with the host you are leaving. You get one IPv4 with the server, IPv6 at no charge, more IPv4 as a catalogue line, and reverse DNS you can edit yourself.
A gigabit link moves about 450 GB an hour at its theoretical limit, and a realistic sustained rate between two data centres is usually 40–70% of that. So half a terabyte is an afternoon and ten terabytes is a few days. The planner above does the arithmetic for your numbers.
No. The gigabit port that comes with a dedicated server is unmetered in both directions and included at no charge, so the size of your data changes how long the move takes and nothing else. Faster ports are available and priced per month.
Yes, and you should — a week at minimum, a fortnight if the service takes money. Billing here is monthly with no contract, so an overlap needs no negotiation and no commitment. There is a discount for paying for longer up front and it is small on purpose.
You change the DNS records back. Your old server is still running, its data is untouched, and with the TTL low the rollback takes about as long as the switch did. That is the entire reason for the overlap, and it is why we would rather you kept the old machine a week longer than you think you need.
There are 33 images you can install from the panel and 32 of them are free: Ubuntu, Debian, AlmaLinux, CentOS, Proxmox VE, VMware vSphere Hypervisor and Windows Server from 2008 R2 through 2025. Reinstalls are unlimited and need no ticket.
Not for the move. Red Hat ended CentOS 7 on 30 June 2024, and we still install it — which lets you separate two frightening jobs that most people are told to do at once. Move onto identical software first, prove it works, then plan the distribution upgrade on a machine you can rebuild from the panel as often as you like.
Three hundred seconds, set 24 to 48 hours before the cutover. The lead time matters more than the value: the world is still holding whatever your old TTL said, so lowering it an hour before the switch buys you nothing. Put it back up a few days afterwards — a permanent five-minute TTL is a lot of extra queries for no benefit once you have landed.
Yes, in one way that catches people out: a listener already connected holds that connection open and will not follow DNS until they disconnect, so your audience migrates over hours rather than minutes and the old stream has to keep running the whole time. Everything else is the plan above, plus the places your stream is listed — directory entries, player embeds, and any app that hardcodes a host. Updating a directory listing is its own guide, and the radio and video pages cover what runs here.
The one nearest your users, not the one nearest your current host — you live with the first choice for years and with the copy for an afternoon. We run New York, Bucharest, Miami, San Francisco and Amsterdam. Each has a latency test you can run from where you are.
We will plan it, be on the call, and fix anything on our side at any hour. We will not log into your current provider on your behalf — partly because most providers' terms forbid it, and mostly because the person who knows what the application does should be the one typing.
Same plan, one extra rule: leave the old server accepting mail for at least a week after the switch, because some senders cache MX answers far longer than they should. Mail also carries a reputation attached to the old IP address that does not transfer, so warm the new one up rather than moving a full send volume on day one. Hosted Exchange has its own page.
cPanel and Plesk both ship transfer tools that move accounts between servers, and that is the route we would use. The licence itself is a separate matter: both are catalogue add-ons here, so you buy a new one rather than moving the old, and you can run the two in parallel for the length of the overlap. What comes with a panel server.
No, and be wary of anyone who says it is. There is one moment where writes stop on the old server and have not yet started on the new one. With the bulk copied days earlier and the TTL already low, that moment is usually seconds to a couple of minutes. Everything else in the ten days is invisible to your users.
Pick the machine, and pick the hour. Neither one commits you to the other, and neither one costs anything until you order — so most people book the call first and choose the hardware on it.
Where to go from here