How to choose a VPS. Size it right, then deploy it.

1. 01

   What will it run?

   List every piece of software: the website, the database, the VPN, the mail server. This decides almost everything else.

2. 02

   How much RAM does that need?

   Add up each piece, plus the operating system, plus 25% headroom. RAM is usually the number that decides your plan.

3. 03

   How much disk and traffic?

   Your data, your backups, and how much your users download each month.

4. 04

   Where are your users?

   Put the server near them. Distance adds delay to every request. We run the same plans in 15 countries.

5. 05

   What’s it running on?

   Full virtualization (KVM), NVMe storage, and billing that lets you test before you commit.

Notice what’s missing: “the one in the middle”. A plan is just a box of resources. The right box is the smallest one your project fits in comfortably.

Every line shows how much memory it adds and why, so you can see exactly where your total comes from.

What popular software needs

For most VPS projects, RAM is the first resource you run out of. That’s why the picker sizes by RAM first.

The operating system takes its share

Ubuntu needs about 1 GB of RAM for a cloud image, and is happier with more for real work. Debian is lighter, at 512 MB to 1 GB. That’s before your own software.

Websites: count PHP workers

WordPress needs about 128 MB per PHP process for a bare install, 256 MB for a typical production site, and 512 MB or more for WooCommerce. Four workers at 256 MB is 1 GB, on top of the OS and the database.

Databases: they’ll use what you give them

Databases don’t have a fixed minimum. They grow into the RAM you give them: on a dedicated database server, MySQL often gets up to 80% of memory for its buffer pool, MariaDB 60 to 70%, and PostgreSQL starts at about 25% for shared_buffers.

On a small VPS running a website and its database together, give the database a modest slice and let the operating system cache the rest.

How to read your memory

Run free -h. Look at the available column, not “free”. Linux uses spare RAM as disk cache, so “free” always looks low. “Available” is what you can still use.

On KVM, every virtual machine is a regular Linux process, scheduled by the host like any other program. Your vCPUs are threads that take turns on the host’s real cores.

On a well-run host that works beautifully. On a crowded one, your server has to wait for a core. Linux measures that wait as steal time: time your vCPU wanted to run but didn’t.

How to check steal time

Open top and look for st on the CPU line. Or run vmstat 1 10 and read the last column.

Here’s the real output from one of our servers while idle:

$ vmstat 1 6
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa st
 0  0      0 7280744  75184 804544    0    0     0     0   58   49  0  0 100  0  0
 0  0      0 7280744  75184 804544    0    0     0     4   52   50  0  0 100  0  0

The last column, st, is 0. That’s what you want. Check it again while your server is busy: a few percent now and then is normal on shared vCPUs; a steady double-digit number means the host is crowded.

How many vCPUs?

Websites, VPNs, bots and most apps are idle most of the time, so a few shared vCPUs go a long way. The Lite range scales with you: two vCPUs on a Lite 2, six on a Lite 6, sixteen on a Lite 16. For builds, video encoding and heavy databases, go up the Lite range. Wondering whether you need a whole machine? Our VPS vs dedicated server comparison has a check for your server.

KVM is full virtualization: each VPS gets its own virtual hardware and runs its own Linux kernel. Container platforms such as OpenVZ and LXC run many customers on one shared kernel.

That last row isn’t theory. mailcow won’t install on OpenVZ, Virtuozzo or LXC, and plenty of other software assumes it has a real kernel to itself.

How to check what you’re on

Run systemd-detect-virt. On a LiteKeep server it prints kvm. Our KVM explainer has a fuller check and a side-by-side test.

Disk: type first, size second

An SSD is dramatically faster than a hard drive for the small, random reads that websites and databases make. NVMe SSDs are faster again. Lites and Vaults run on NVMe, big disks included.

Size it as: the OS (about 25 GB), plus your software, plus your data, plus room for logs and updates. Our Lite plans start at 40 GB; our Vaults give you 700 GB to 3 TB for backups, nodes and archives.

Traffic: allowance vs speed

Plans describe traffic in two ways. A monthly allowance (say 1 TB) is how much data can leave your server. A port speed is how fast. “Unlimited” plans usually cap the speed instead of the amount.

Ours: each plan’s allowance adds to a pool shared by all your servers, only outbound traffic counts, and anything beyond the pool is $2 per TB. If you serve video or big downloads, a Traffic Lite with 32 TB a month is built for it. Our bandwidth calculator tells you which.

Location: put it near your users

Every 1,000 km between your server and your users adds at least 10 ms to every request and reply. The country finder ranks our 15 locations by distance from your users, and every plan costs the same in all of them.

Root access, your way

LiteKeep servers are unmanaged: you get full root access and decide everything. Updates, a firewall and SSH keys take about 30 minutes to set up, and our VPS security guide builds the commands for you.

Backups and snapshots

Take a snapshot in the console before any big change. For real off-site backups, put a Vault in a second country: if one location has a bad day, your data doesn’t.

Billing that matches your plans

Hourly billing lets you test and throw away servers for cents, and a month never costs more than the monthly price. Yearly Keeps give you up to two months free on servers you’re sure about. Our VPS cost guide compares the options.

Payment and support

No card? No problem. Pay with Bitcoin, USDT, USDC, Monero and six more coins, with no ID: an email and a password is the whole account. Questions before you buy? Ask a person.

Three buying mistakes to skip

• Buying vCPUs for a slow site. Most sites run out of RAM or wait on the database long before they max out the CPU.
• Buying for the traffic you hope to have. With hourly billing, buy for today and move up to a bigger Lite when you need it.
• Treating every “4 GB VPS” as equal. Virtualization, disk type, steal time and location make two “identical” plans very different.

Here’s what each check showed on one of our own servers, so you know what “good” looks like:

Quick answers

How much RAM do I need for a VPS? For a VPN or a bot, 1 to 2 GB is plenty: a Lite 2. For a small WordPress site, about 2.5 to 3 GB once you count the OS, two to four PHP workers, the database and headroom: a Lite 6. For mailcow, at least 6 GiB plus swap: a Lite 8. Use the picker for your exact mix.

Can I move up later? Yes. Start small, and when your memory says so, move to a bigger Lite. You only pay for the hours each server exists.

Is 1 vCPU enough? For a single light service, often yes. Two vCPUs keep one busy task from freezing everything else, which is why every Lite starts at two.