Skip to main content
Swap is a file on disk that the system uses as “spare” memory when RAM runs out. Without swap, a process that runs out of memory gets killed by the OOM-killer — the service crashes. With swap, the kernel pushes inactive pages out to disk and the process keeps running.
Swap is not a replacement for RAM. Disk is tens of times slower than RAM. Swap saves you from sudden crashes and smooths out spikes, but if the server constantly “lives” in swap, that’s a sign it’s genuinely short on memory.

​
Check the current state

First, see whether swap already exists and how much free memory there is: 
free -h
swapon --show
free -h shows the Mem (RAM) and Swap lines. If swapon --show outputs nothing, there’s no swap.

​
Create a swap file

Connect to the server over SSH as root (the IP and password are in the server card in the bot) and run the commands step by step.
1

Create a file of the size you need

The quick way is fallocate:
fallocate -l 2G /swapfile
If fallocate isn’t available or the file system doesn’t support it, use dd:
dd if=/dev/zero of=/swapfile bs=1M count=2048 status=progress
count=2048 with bs=1M gives exactly 2 GB.
2

Lock out access for others

The swap file must be accessible to root only, otherwise the system will refuse to enable it:
chmod 600 /swapfile
3

Format it as swap

mkswap /swapfile
4

Enable it

swapon /swapfile
Check that swap has appeared:
swapon --show
free -h
5

Enable it at boot

So that swap is mounted after a reboot, add a line to /etc/fstab:
echo '/swapfile none swap sw 0 0' >> /etc/fstab
Check that the line was added only once: grep swapfile /etc/fstab — duplicates in fstab cause errors at boot.

​
How much to allocate

There’s no hard rule; here’s a rough guide:
Server RAMRecommended swap
1–2 GBequal to RAM (1–2 GB)
4–8 GB2–4 GB
16 GB and up2–4 GB is enough
The logic: on the smaller plans (VPS-0, VPS-1 with 1–2 GB of RAM) swap genuinely helps you out; on the larger ones it’s more of “insurance” against rare spikes, so there’s no point chasing size.

​
Tune swappiness

vm.swappiness (0–100) determines how eagerly the kernel reaches for swap. The default is usually 60. For a server, it makes sense to lower it to 10 — then swap is used only when memory really runs short, not “just in case”. Current value: 
cat /proc/sys/vm/swappiness
Apply it right away (until reboot): 
sysctl vm.swappiness=10
Make it permanent — add it to /etc/sysctl.conf: 
echo 'vm.swappiness=10' >> /etc/sysctl.conf
Lumi’s disks are on NVMe, so swap here is noticeably faster than on regular HDD/SATA-SSD. But even fast NVMe is still slower than RAM. If you’re constantly hitting the memory ceiling and swap is “on fire”, that’s not a reason to inflate the file — it’s a reason to upgrade your plan through support.

​
Where to next

Run an LLM on the server

Ollama and local language models on a VPS.

VPS not working

Connection issues, load, and other common situations.