zram swap on low-memory systems

What you will achieve

zram compressed RAM swap on low-memory Raspberry Pi or VPS instances — reduces OOM kills without hammering SD card I/O like a huge swapfile.

1) Ubuntu zram-generator

sudo apt install zram-generator
# /etc/systemd/zram-generator.conf
[zram0]
zram-size = ram / 2
compression-algorithm = zstd

sudo systemctl daemon-reload
sudo systemctl start /dev/zram0

2) Manual zram (generic)

echo zstd > /sys/block/zram0/comp_algorithm
echo 2G > /sys/block/zram0/disksize
mkswap /dev/zram0
swapon -p 100 /dev/zram0

3) Trade-offs

zram uses CPU for compression — fine on 2+ cores with 1 GB RAM; less ideal under sustained memory pressure on heavy databases.

Verify

zramctl
swapon --show
free -h

5) Check if zram already active

cat /proc/swaps
zramctl

Ubuntu 24.04 may enable zram by default — avoid double swap layers.

6) Size tuning

zram-size = ram / 2 is common starting point. Monitor compression ratio in /sys/block/zram0/mm_stat.

7) Disable disk swap when using zram

sudo swapoff -a
sudo sed -i '/swap/d' /etc/fstab

Not for every workload

Database servers with strict latency SLAs should size RAM properly rather than rely on compressed swap — zram helps constrained edge nodes and Pis more than 64 GB DB hosts.

8) zram generator on Fedora

Fedora enables zram swap by default on many spins — verify before adding duplicate swapfile.

Prerequisites

Low RAM device (≤4 GB typical). Kernel zram module. Check existing swap — avoid duplicate layers. CPU headroom for compression — very CPU-bound workloads may suffer.

Verify swappiness with zram

cat /proc/sys/vm/swappiness

Default 60 may push to zram early — lowering to 100 on zram-only setups is sometimes recommended in Pi docs; test with your workload.

zram on Raspberry Pi OS

Pi OS enables zram via init — check dmesg | grep zram before manual zram-generator config duplicates effort.

zram generator size formula

min(ram/2, 4096) MB cap on small boards prevents zram consuming disproportionate CPU compressing gigabytes rarely swapped anyway — tune per workload measurements not forum defaults.

Memory cgroup interaction

systemd MemoryMax= on service limits RSS — zram swap still counts toward pressure — tune both not one in isolation on systemd 252+.

zram stats sysfs

cat /sys/block/zram0/comp_algorithm
 cat /sys/block/zram0/mm_stat

Compare orig_data_size vs compr_data_size for compression ratio — tune algorithm zstd vs lzo based on CPU budget.

disable zram test

Benchmark workload with zram off vs on same hardware — micro benchmarks lie, real app memory footprint tells truth for your Pi or VPS tier.

Related guides

• Find disk usage with du and ncdu
• Fix high load average on Linux