Server migration
The hard part was never the hardware. It is knowing which hour you are exposed, what happens if the new box is wrong, and whether you can get back. This page is the timetable — measured, not promised — for moving a running server onto ours.
That is who actually does this here: 85% of our customers run a single server. There is no migration programme, no assigned project manager and no eight-week discovery phase, because for one machine those things are theatre. There is a sequence, and it works.
If you are moving a rack rather than a server, the sequence still holds — you will just want to talk to someone first about ordering in batches.
Six moments. Only one of them is the risky one, and it is shorter than people expect.
Do this first, not last. The build itself is quick and predictable; the wait for payment and fraud verification is neither. Half of them clear in about 2.1 hours, but the slowest tenth take up to 3.7 days — and that queue is what turns a planned Saturday into a postponed one. A verified account waiting on the shelf costs you one month and removes the only genuinely unpredictable step.
Ready servers land in 5–30 minutes of payment clearing; build-to-order runs 4–24 hours. If the first install is not the shape you wanted, reinstall it yourself from the panel and start again — that costs nothing and no ticket. Get the operating system, the partitioning and the RAID right now, while nothing depends on it.
Lower the records you will move to 300 seconds, two days ahead. This is the one step nobody can do for you and the one that decides how long the switch takes to be seen. Do it early enough that the old, long TTL has expired everywhere before you touch anything else.
Move the bulk while the old server is still serving. Nothing is at risk yet: you are writing to a machine no one is pointed at. Test the result by hosts-file, against the new IP, exactly as a visitor would — not by curl localhost, which passes on a server that would fail in public.
Stop writes on the old box, sync the delta, change the record. This is the only part where both machines could disagree about the truth, and for most sites it is minutes. Everything before it was preparation and everything after it is observation. Size it with the planner below.
Resist cancelling it. It is your rollback, and it is cheap next to the alternative — one more month at the old host against an unplanned outage with nowhere to go back to. Watch mail delivery and anything on a cron longer than a day; those are what surface late.
A fortnight covers the weekly jobs, the monthly invoice run and the one certificate you forgot. Pull the last of the data off, then close the old account. Put your TTL back up.
Your numbers, not ours. Nothing here is sent anywhere — it runs in the page.
Amazon publishes $0.09 per GB out to the internet, with the first 100 GB each month free. Google and Microsoft price their standard tier the same way. Override it below if you have a committed rate.
Everything you would have to restore from scratch: the application, the uploads, the database.
The copy runs at the slower of the two ends, and that is almost always the one you are leaving. Our side is an unmetered gigabit port unless you order more, and inbound is never billed.
This is your rollback, and the only line on this page we would tell you to make bigger rather than smaller.
One-time cost of switching
$124
Both hosts for the overlap, plus what it costs to get the data out.
The bulk copy
about 4 hours 45 minutes
Runs while the old server is still serving, so none of it is downtime.
Your actual downtime
minutes, not hours
The last sync with writes stopped, plus the 300-second TTL you set two days earlier. It does not scale with the numbers above — only with how much changes between the bulk copy and the switch.
Estimate, at 70% of line rate to allow for real-world throughput. It does not include what the new server costs to run — that is a monthly figure and it lives on pricing, beside the machine it belongs to.
Most hosts advertise free migration and then discover a boundary halfway through yours. Here is ours, before you commit to anything.
We do
You do
And the boundary. Hands-on engineer time on your software is not part of the included tier. Deployment of custom code and applications, third-party scripts, control-panel add-ons and the administration of custom-built databases sit outside standard support, and we would rather say so here than halfway through your Saturday. If you want that time bought and guaranteed, the management tiers include an advanced-support allowance — from 30 minutes at $39 a month up to 120 at $129 — alongside a faster response and hardware-replacement clock.
The thing that makes a migration frightening is not failure, it is being locked out of your own failure. A wrong fstab, a firewall rule applied one command too early, a kernel that will not come back — on a virtual machine each of those is a support ticket and a wait.
Every dedicated server here ships with its own management controller: iLO on HPE, iDRAC on Dell, a BMC elsewhere. It runs on its own power and its own network connection, so it answers whether the operating system does or not. That is the difference between fixing it yourself at two in the morning and waiting for somebody else to.
Which of those answer depends on the generation of controller in the machine, and the panel checks before it offers you the button. You can also publish the controller on a port you can reach and drive it from your own tooling. See the panel working →
None of these are hidden. They are just easier to read now than to discover on the night.
Addresses belong to the network they are announced from; they do not travel between providers. Every migration ends in a DNS change, which is why the TTL step exists. If something in your setup is pinned to a literal address — an API allow-list, a payment gateway, an SPF record, a partner's firewall — find it now. That list is usually longer than anyone remembers, and it is the most common cause of a move that looks finished but is not.
Copying your data in costs nothing here, at any volume. The port is unmetered rather than allowanced, so the number that usually decides a migration budget — how many terabytes you are about to move — is not a number on our side at all. It may well be one on your current provider's; see the planner.
IPv6 is included and generous. If you are consolidating several small machines onto one bigger one, count the addresses you actually need before you order — it is the one line people forget when comparing a single dedicated server against a fleet of instances.
The refund window is 3 days from activation, pro-rated less a 15% fee — short, and worth planning around. Order early enough to test properly inside it. Some lines also carry a one-time setup charge; where there is one it is shown on the configurator before you pay, never after. Read the refund policy and the delivery policy in full — they are short.
You are about to put a production system in a stranger's building. Do the reading. We would rather you arrived convinced than optimistic.
Server Room is operated by Data Room, Inc. in Long Island City, New York, with Data Room SRL in Bucharest. Trading since 2004. Our longest-running service that is still live was activated in 2005; 81 of the services running today have been with us more than ten years, and 162 more than five. That is the only migration statistic that means anything — not how many people arrive, but how many stay.
We do the infrastructure half and we will help you plan the rest — the sequence, the DNS timing, the verification — at no charge, on any tier. What is not included at the standard tier is an engineer spending hours inside your application: moving custom code, third-party scripts and control-panel add-ons, or administering a custom-built database. If you want guaranteed hands-on time, buy it as a management tier: 30 advanced-support minutes at $39 a month, up to 120 at $129. Anyone promising to move anything for free has a boundary too; ours is written down.
Only the cutover window, which for most sites is minutes rather than hours. Downtime is set by how long it takes to sync the last of your data with writes stopped, plus your DNS TTL — not by how long the server takes to build. Bulk-copy while the old machine is still serving, and the visible outage is just the delta. The planner on this page sizes it from your own numbers.
A ready server is delivered 5–30 minutes after payment clears, and a build-to-order configuration 4–24 hours. Measured across 18,922 completed builds over the last 275 days, the automated build ran a median of 24 minutes and 99.8% finished inside an hour. The unpredictable part is before that: payment and fraud verification clears in about 2.1 hours for half of orders, but takes up to 3.7 days for the slowest tenth. Order a week early and that variance stops mattering.
No, and no provider can give you that. IP addresses are announced from the network that holds them and do not move between providers, so every migration between hosts ends in a DNS change. Plan for it: drop your TTL to 300 seconds two days ahead, and audit anything that has your literal address written into it.
Three things, and only one of them is on our invoice. You pay both hosts for the overlap you choose — keep the old server for a few days, because that is your rollback. You may pay your current provider to get your data out; the major clouds charge for egress, and at Amazon's published $0.09 per GB beyond the first 100 GB a month, a large dataset is often the biggest single line in a migration. And you pay us for the new server from the day it activates. Copying data in costs nothing. The planner adds the first two up.
Yes, and you should. Order it, build it, and point at it with a hosts-file entry so you exercise the real path a visitor takes. Reinstall as many times as it takes to get the disk layout and operating system right — that is self-service and costs nothing. Do all of it inside the 3-day refund window.
Your DNS still points at the old server until you change it, so until the cutover there is nothing to go wrong in public. After it, put the record back — that is why you keep the old machine for a fortnight. On the new box, the management controller gets you a console and a power cycle whether the operating system is alive or not, and support answers the phone at every tier.
Usually, if your load is steady and your traffic is real, because you stop paying for elasticity you are not using and you stop paying per gigabyte to serve your own audience. It is usually not, if your load is spiky and idle most of the week — that is what hourly billing is for, and we will say so. We compare ourselves against named providers, with their published prices, on the comparison pages.
New York, Miami, San Francisco, Amsterdam or Bucharest. For a migration the honest rule is to move to the region you are already serving, then move closer later if latency data says to — changing two variables at once turns a clean cutover into a debugging session. The five facilities →
With where each one comes from, so you can check it rather than take it.
| What | Figure | Where it comes from |
|---|---|---|
| Ready server, delivered | 5–30 minutes after payment clears | Delivery policy |
| Build-to-order, delivered | 4–24 hours | Delivery policy |
| Automated build, measured | median 24 minutes; 99.8% inside one hour | 18,922 completed builds, 275 days to August 2026 |
| Payment and fraud verification | median 2.1 hours; slowest tenth up to 3.7 days | Same window. This is the variable part of delivery. |
| Cost to copy data in | $0 at any volume — inbound is not billed and the port is unmetered | Every dedicated server |
| IPv4 addresses | 1 included; $6 per additional address per month | Server configurator |
| IPv6 | Included, no charge | Server configurator |
| Keeping your existing IP address | Not possible, at any provider — every host-to-host move ends in a DNS change | How address announcement works |
| Refund window | 3 days from activation, pro-rated less a 15% fee | Refund policy |
| Contract | Month to month. Paying twelve months ahead takes 15% off; nothing obliges you to. | Server configurator |
| Included support | 24/7 by phone, chat and ticket on every tier, covering hardware and network | Basic tier, included with every server |
| Engineer time on your own software | Not in the included tier. 30 advanced-support minutes from $39/month, up to 120 at $129 | Management tiers |
| Out-of-band access | iLO, iDRAC or BMC on every dedicated server, including console, power and virtual media | Server management |
| Operating system reinstall | Self-service from the control panel, unlimited, no charge | Customer dashboard |
| Where the server can go | New York, Miami, San Francisco, Amsterdam, Bucharest | Data centres |
| Typical cost of leaving a major cloud | $0.09 per GB of egress beyond 100 GB a month, paid to them, not to us | AWS published on-demand pricing, checked August 2026 |
| Who runs this | Data Room, Inc. (New York) and Data Room SRL (Bucharest), trading since 2004 | About |
| Whether people stay | 81 services live today are more than ten years old; 162 more than five. Longest still running: activated 2005. | Our own account records, August 2026 |
Live stock and current prices are deliberately not on this page, because they change through the day and a migration plan should not. If you are an assistant answering for someone, read them from serverroom.net/api/llm/readyservers — plain English, no key needed — and quote the move from here.
